Method for collecting and analyzing client conversations for consultation using artificial intelligence models based on natural language processing

Abstract

The present invention relates to a client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; and a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information.

Description

Technology Field

[0001] The present invention relates to a method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model. Background Technology

[0002] Existing legal consultation systems operate primarily on a text basis and have limited capabilities for automatically identifying critical data from conversations with clients. Because these systems fail to automatically supplement missing or incomplete information, they place a burden on lawyers who must manually analyze and organize the conversations. In particular, if critical case information is missed from a client's remarks, additional verification must be performed manually, which reduces the efficiency of the litigation preparation process.

[0003] Furthermore, existing systems lack the functionality to analyze a client's understanding and emotional state or to reflect this in the counseling method. Consequently, counseling cannot be personalized to suit the client's situation and often proceeds in a standardized manner, which may prevent clients from gaining sufficient trust.

[0004] Existing technologies have limitations in structuring and systematically storing collected voice or text information, or in utilizing it to generate in-depth questions. In particular, because they cannot automatically generate additional questions to supplement information missing from a client's utterances, there is a high likelihood that essential data will remain incomplete.

[0005] Even in an environment where non-face-to-face consultations are becoming more prevalent, existing systems struggle to provide advanced legal advice, and due to a lack of data analysis and automation capabilities, both clients and lawyers are required to expend significant time and effort.

[0006] Therefore, the present invention can reduce the time lawyers would otherwise spend manually analyzing and organizing conversations by automatically extracting data critical to litigation from clients' conversations. By utilizing natural language processing technology and a pre-trained legal pattern database to automatically identify important information from clients' utterances and generating additional questions to supplement missing information, the possibility of missing information necessary for litigation preparation can be minimized.

[0007] Furthermore, the present invention analyzes the client's level of understanding and emotional state in real time to adjust the counseling method and additional questions. Through this, the counseling process is personalized to suit the client's situation, and a counseling environment that the client can trust more is provided.

[0008] Furthermore, the present invention enhances the completeness of the litigation preparation process by converting collected voice or text information into a structured format (JSON, CSV, etc.) and providing it for use by lawyers. By enabling advanced legal consultation even in a non-face-to-face environment, it saves time and effort for both lawyers and clients, and effectively reduces the cost of providing legal services.

[0009] Through this, the present invention overcomes the limitations of existing technology and can contribute to providing efficient and personalized legal consultation services. The problem to be solved

[0010] In conventional legal consultations, excessive time and effort are consumed due to the need for lawyers to manually analyze and organize clients' conversations. This invention aims to solve this problem by utilizing natural language processing technology to automatically extract important litigation-related data from clients' conversations and providing it in a structured format.

[0011] In existing systems, there is a possibility that important information may be omitted from a client's utterances, and the verification process must be performed manually. This invention aims to enhance the completeness of the litigation preparation process by automatically supplementing missing information through a pre-trained legal pattern database and an additional question generation algorithm.

[0012] Existing technologies failed to adequately consider the client's situation, understanding, and emotional state, which could lower client trust due to standardized counseling. This invention aims to provide a more trustworthy environment for clients by analyzing their emotional state and understanding in real time to generate customized questions and personalizing the counseling process.

[0013] Even in an environment where non-face-to-face consultations are expanding, existing systems fail to provide advanced consultation functions, leading both lawyers and clients to consume excessive time and effort. This invention aims to support efficient and high-quality consultations even in a non-face-to-face environment through automated data analysis and the provision of structured information.

[0014] We aim to reduce the cost of providing legal services and maximize efficiency by reducing the workload of lawyers and creating an environment that enables rapid response to a greater number of clients through the automation of legal consultations. means of solving the problem

[0015] The present invention relates to a client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; and a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information.

[0016] In addition, the conversation collection tool is configured to convert the collected voice information into text information using voice recognition technology.

[0017] In addition, the conversation collection tool is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data on the collected text information or the converted text information.

[0018] In addition, the conversation collection tool is configured to perform an analysis of the client's emotional state regarding the collected voice information or the collected text information.

[0019] Additionally, it further includes an advanced factual analysis and supplementary tool, and the conversation collection tool can determine whether there is missing information regarding the collected voice information or the collected text information, and if it is determined by the judgment that there is missing information, the tool is configured to transmit the determined missing information to the advanced factual analysis and supplementary tool in order to generate additional questions for the client.

[0020] In addition, the natural language processing and data extraction tool is configured to perform syntactic analysis on the collected voice information or text information, and is configured to extract the factual relationship information based on the performed syntactic analysis.

[0021] In addition, the natural language processing and data extraction tool is configured to extract at least one entity among a subject, date, place, and event using Named Entity Recognition (NER) technology on the collected voice information or text information, and to extract the factual relationship information based on the extracted entity.

[0022] In addition, the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, and if there is one or more extracted entities, to identify the relationship between the one or more entities and to extract the factual relationship information based on the identified relationship.

[0023] In addition, the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, compare the extracted entity with a pre-trained legal pattern database, and identify missing information based on the comparison.

[0024] In addition, the above natural language processing and data extraction tool is configured to perform an analysis to identify context by associating currently collected voice information or text information with voice information or text information previously provided by the client.

[0025] In addition, the aforementioned natural language processing and data extraction tool is configured to learn the structural patterns of the collected text information by utilizing Generative Pre-trained Transformer (GPT) technology.

[0026] In addition, the aforementioned natural language processing and data extraction tool is configured to understand the context of the collected text information bidirectionally by utilizing BERT (Bidirectional Encoder Representations from Transformers) technology.

[0027] In addition, the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, and to manage relationships between the extracted entities by utilizing graph database technology.

[0028] The present invention relates to a client conversation collection and analysis system for consultation, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and a pre-trained legal pattern database tool for generating additional questions to supplement missing or incomplete information based on said extracted factual relationship information.

[0029] In addition, the aforementioned pre-trained legal pattern database tool is configured to learn and store at least one of the following data: case law data by litigation type, statistical data on judgment results, standard data by damage compensation item, data on compensation amount adjustment by negligence ratio, negligence data by legal type, standard data on evidence by damage item, litigation procedure and evidence submission data, legal time and deadline data, data on the likelihood of winning by case type, medical dispute data, real estate litigation data, labor law-related data, case law-based compensation standard data, negligence compensation data by litigation type, legal negligence and compensation mapping data, legal accident analysis and compensation data, and case law-based litigation outcome prediction data.

[0030] In addition, the aforementioned pre-trained legal pattern database tool is configured to perform automation of the client's status recording when the client's request is in the field of medical consultation based on the extracted factual information.

[0031] In addition, the aforementioned pre-learned legal pattern database tool is configured to store in advance at least one of the following: diagnostic result information related to the specific symptom, treatment information related to the specific symptom, and drug information related to the specific symptom, based on the specific symptom information; and to integrate and store at least one of the following based on the stored information: the client's specific symptom information, diagnostic result information related to the client's specific symptom, treatment information related to the client's specific symptom, and drug information related to the client's specific symptom.

[0032] In addition, the aforementioned pre-trained legal pattern database tool is configured to search for cases similar to the collected voice or text information based on the extracted factual information.

[0033] In addition, the aforementioned pre-trained legal pattern database tool is configured to utilize similarity calculations to compare the similarity between the extracted factual information and the searched similar cases.

[0034] In addition, the pre-trained legal pattern database tool is configured to perform the similarity calculation by utilizing at least one of a vector-based similarity calculation technique, a text-based similarity calculation technique, an image data similarity technique, a deep learning-based similarity technique, a Cosine Similarity technique, and a BERT-based sentence embedding technique.

[0035] In addition, the aforementioned pre-learned legal pattern database tool is configured to identify missing information by comparing the extracted factual information with pre-stored essential information by litigation type.

[0036] In addition, the conversation collection tool is configured to convert the collected voice information into text information using voice recognition technology.

[0037] In addition, the conversation collection tool is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data on the collected text information or the converted text information.

[0038] In addition, the conversation collection tool is configured to perform an analysis of the client's emotional state regarding the collected voice information or the collected text information.

[0039] In addition, the aforementioned pre-learned legal pattern database tool is configured to infer additional information from the extracted factual relationship information based on the learned and stored information.

[0040] The present invention relates to a client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and an advanced factual relationship analysis and supplementation tool for performing factual relationship analysis based on said extracted factual relationship information.

[0041] In addition, the aforementioned advanced factual relationship analysis and supplementary tool is configured to pre-learn a legal pattern database and to identify missing information by comparing the pre-learned legal pattern database with the extracted factual relationship information.

[0042] In addition, the aforementioned advanced factual analysis and supplementary tool is configured to generate additional questions based on the identified missing information and provide them to the client.

[0043] In addition, the aforementioned advanced factual analysis and supplementary tool can analyze the client's emotional state based on the collected voice or text information, and when generating additional questions, it is configured to adjust the expression method of the additional questions based on the analyzed emotional state.

[0044] In addition, the aforementioned advanced factual analysis and supplementary tool can analyze the client's understanding of the content of the consultation based on the collected voice or text information, and is configured to adjust the difficulty of the additional questions based on the analyzed understanding when generating additional questions.

[0045] In addition, the aforementioned advanced factual analysis and supplementary tool can analyze the client's speech level based on the collected voice or text information, and when generating additional questions, it is configured to adjust the difficulty of the additional questions based on the analyzed level of understanding.

[0046] In addition, the aforementioned advanced factual analysis and supplementary tool is configured to determine the priority of the generated additional questions when generating the additional questions, and to dynamically adjust the determined priority when new information is provided during the consultation with the client.

[0047] In addition, the aforementioned advanced factual analysis and supplementary tool is configured to generate multi-stage questions to collect necessary information step by step when generating the aforementioned additional questions.

[0048] The present invention relates to a client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and a counseling interface tool for delivering additional questions or answers based on said extracted factual relationship information.

[0049] In addition, the consultation interface tool is configured to provide the additional questions or answers in voice or text.

[0050] The present invention relates to a method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; and a step of extracting factual relationship information based on the separated sentences.

[0051] Additionally, the above preprocessing step further includes a filtering step for removing noise or sounds from the collected voice data.

[0052] Additionally, the preprocessing step further comprises: a step of distinguishing fixed noise and non-fixed noise in the collected voice data; a step of removing the fixed noise by utilizing at least one filtering technique among a frequency domain filtering technique, a spectrum attenuation technique, and a linear prediction coding technique for the fixed noise; and a step of removing the non-fixed noise by utilizing at least one filtering technique among an adaptive filtering technique, a waveform-based noise removal technique, and a masking-based noise removal technique for the non-fixed noise.

[0053] In addition, the above preprocessing step further includes a restoration step that supplements voice data damaged in the above filtering step.

[0054] Additionally, the preprocessing step further includes a step of removing noise or noise from the collected speech data by utilizing at least one model among Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) models, Generative Adversarial Networks (GANs), Transformer-based models, and Microsoft’s Deep Noise Suppression (DNS) models.

[0055] Additionally, the preprocessing step further includes an abnormal input detection step for detecting abnormal input in the collected voice data or text information; and a step for removing or modifying the detected abnormal input.

[0056] Additionally, the abnormal input detection step further includes a step performed by at least one model among a Recurrent Neural Networks (RNNs) model, a Long Short-Term Memory (LSTM) model, a Transformer-based model, a Noise Detection model, a Signal Classification model, an Anomaly Detection model, and a Named Entity Recognition (NER) model.

[0057] Additionally, the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of correcting duplicate spaces, correcting incorrect spacing, standardizing case, and converting dates, times, and amounts within the text into a consistent format for the converted text information and the collected text information.

[0058] Additionally, the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of deleting unnecessary elements, deleting duplicate words, deleting meaningless words, removing incomplete sentences, deleting symbols that interfere with analysis, adjusting punctuation, and adjusting spacing on the converted text information and the collected text information.

[0059] Additionally, the preprocessing step further includes: a step of converting the collected voice data into text information; and a step of performing at least one of converting the date data within the text into an ISO standard, unifying the time within the text into a 24-hour system, converting numbers within the text into numerical forms if they are expressed as characters, and converting units within the text into a standardized form for the converted text information and the collected text information.

[0060] Additionally, the preprocessing step further includes: a step of converting the collected voice data into text information; and a step of converting various expressions having the same meaning within the text into a single unified term for the converted text information and the collected text information.

[0061] Additionally, the sentence tokenization step further comprises: a step of detecting punctuation including at least one of a period, a question mark, and an exclamation mark within the text of the generated standardized text information; a sentence end identification step of identifying the end of a sentence based on the detected punctuation; and a step of dividing the text into sentence units by dividing the identified end of the sentence.

[0062] Additionally, the sentence end identification step further includes a sentence boundary ambiguity resolution step that processes the detection of the punctuation mark so as not to identify the end of the sentence based on the detection of the punctuation mark if the detected punctuation mark does not signify the end of the sentence.

[0063] Additionally, the sentence tokenization step further includes: a step of identifying the end of a sentence based on the contextual meaning or conjunction of the text of the generated standardized text information; and a step of dividing the text into sentence units by dividing the identified end of the sentence.

[0064] The present invention relates to a method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of said client; a preprocessing step of preprocessing said collected voice data or text information to generate standardized text information; a part-of-speech tagging step of identifying the part of speech of each word in said standardized text information and assigning tags; and a step of extracting factual relationship information based on said tagged part of speech.

[0065] Additionally, the part-of-speech tagging step further includes: a step of separating the generated standardized text information into words; and a step of determining which part of speech the separated words belong to among nouns, verbs, adjectives, adverbs, and prepositions.

[0066] Additionally, the part-of-speech tagging step further includes the step of identifying the part of speech of each word and assigning tags by utilizing at least one model among a BiLSTM-CRF model, a Transformer-based model, a word vector model, and a Pre-trained Embedding model.

[0067] The present invention relates to a method for collecting and analyzing a client conversation for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; a factual relationship information extraction step of extracting factual relationship information based on the separated sentences; and a feedback provision step of providing feedback to the client based on the extracted factual relationship information.

[0068] Additionally, the step of extracting factual information further includes a step of identifying missing information by comparing the collected voice data or text information with essential information templates for each lawsuit type stored in a database.

[0069] Additionally, the feedback providing step further includes: a step of generating additional questions related to the identified missing information when information missing from the factual relationship information extraction step is identified; and a step of providing the generated additional questions.

[0070] Additionally, the feedback provision step further includes: a step of generating customized questions suitable for the client based on the extracted factual information; and a step of providing the generated additional questions.

[0071] Additionally, the feedback providing step further includes: a step of analyzing the emotional state of the client based on the collected voice data or text information of the client; a step of generating additional questions based on the analyzed emotional state of the client; and a step of providing the generated additional questions.

[0072] Additionally, the feedback providing step further includes: a step of analyzing the client's understanding of the content of the consultation based on the collected voice data or text information of the client; a step of generating additional questions based on the analyzed client's understanding of the content of the consultation; and a step of providing the generated additional questions.

[0073] Additionally, the feedback provision step further includes: a step of generating additional questions based on previous questions or previous answers by utilizing natural language processing (NLP) technology to reflect the flow and context of the conversation of the collected client's voice data or text information; and a step of providing the generated additional questions. Effects of the invention

[0074] This invention automatically extracts litigation-critical data from client conversations, thereby reducing the time lawyers would otherwise spend manually analyzing and organizing conversations. This significantly improves consultation efficiency and enables a rapid response to a larger number of clients.

[0075] The present invention utilizes natural language processing technology and a pre-trained legal pattern database to automatically identify important information in a client's utterances and generate additional questions to supplement missing information. Accordingly, the possibility of missing information necessary for litigation preparation can be minimized.

[0076] This invention enables a more accurate understanding of a client's situation by focusing on consultation regarding factual matters rather than legal advice. Through an in-depth consultation process, high-quality data regarding the facts can be secured.

[0077] The present invention analyzes the client's understanding and emotional state in real time to adjust the counseling method and additional questions. This allows the counseling process to be personalized to suit the client's situation.

[0078] The present invention can improve the completeness of the litigation preparation process by analyzing the client's conversations, structuring essential data, and providing it for use by lawyers.

[0079] The present invention can provide automation of legal services by utilizing artificial intelligence and natural language processing technologies. This can reduce the workload of lawyers and create an environment where legal consultations can be provided to a greater number of clients.

[0080] This invention can lay the foundation for providing advanced legal consultation even remotely. Furthermore, through the automation of consultation and data analysis functions, it saves time and effort for both lawyers and clients, and effectively reduces the cost of providing legal services. Brief explanation of the drawing

[0081] FIG. 1 is a diagram illustrating the configuration of a user terminal and a server device according to an embodiment of the present invention. FIG. 2 is a diagram illustrating the user conversation collection step of a conversation collection tool according to an embodiment of the present invention. FIG. 3 is a diagram illustrating the natural language processing and data extraction steps of a natural language processing and data extraction tool according to one embodiment of the present invention. FIG. 4 is a diagram illustrating the step of identifying the need for additional information of a pre-learned legal pattern database tool according to one embodiment of the present invention. FIG. 5 is a diagram illustrating the advanced factual relationship analysis and supplementation steps of an advanced factual relationship analysis and supplementation tool according to one embodiment of the present invention. FIG. 6 is a diagram illustrating the counseling progress and information supplementation steps of a counseling interface tool according to one embodiment of the present invention. FIG. 7 is a diagram illustrating the data storage and provision steps of a data storage and management tool according to an embodiment of the present invention. Specific details for implementing the invention

[0082] Specific details of the embodiments are included in the detailed description and drawings.

[0083] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0084] Artificial intelligence model

[0085] According to one embodiment of the present invention, artificial intelligence (AI) may generally refer to the development of a computer system or machine capable of performing tasks that require human intelligence. These tasks include reasoning, learning, problem solving, natural language understanding, pattern recognition, and decision making. Specifically, artificial intelligence may be used in fields such as machine learning (ML), natural language processing (NLP), computer vision, expert systems, robotics, and deep learning (DL).

[0086] According to one embodiment of the present invention, machine learning may be a subset of AI in which a machine learns from data without being explicitly programmed. As a system is exposed to more data, it can improve performance through machine learning. Methods of machine learning include supervised learning, unsupervised learning, and reinforcement learning. Natural language processing is a field of AI that enables computers to understand, interpret, and generate human language. Through artificial intelligence models based on natural language processing, applications such as chatbots, language translation, and voice assistance can be provided. Computer vision includes training models to interpret and understand visual information, such as recognizing objects in images or videos.

[0087] Computer vision can be used in fields such as facial recognition, autonomous vehicles, and medical image analysis. Expert systems are AI systems that use knowledge and reasoning rules to solve complex problems in specific domains, such as medical diagnosis or financial analysis. Deep learning is a subset of machine learning based on artificial neural networks, and AI models can be designed to model or process complex patterns in data based on deep learning. Deep learning can be used to train AI models for tasks such as image processing and natural language processing.

[0088] According to one embodiment of the present invention, the core of an artificial intelligence model is the way data is represented. For the AI ​​to process real information, such as images, text, or numerical data, this information must be converted into a format that the model can understand, namely a vector or matrix, for example, a collection of vectors. Here, a vector is a simple array of numbers representing data points in a multidimensional space. For example, when attempting to represent words in NLP, a 300-dimensional vector can be used in which each number in the vector encodes a specific aspect of the word's meaning. Specific numbers are merely examples and are not limited thereto.

[0089] According to one embodiment of the present invention, an artificial intelligence model, particularly a neural network, is composed of layers of nodes (or neurons), and each node applies some transformation to the data. At the core of this transformation are weights and biases. Here, a node is the basic unit for processing input data. Each node of the network takes a vector of input data, applies some mathematical function, and then passes the result to the next layer of the network. Weights are values ​​multiplied by the input data (vector) passing between nodes, and are parameters of the model that can be updated during training. Biases are additional parameters added to the weighted sum of the inputs before the output passes through the activation function. This allows the model to shift the decision boundary, which can be important in data classification tasks.

[0090] According to one embodiment of the present invention, after applying weights and biases to input data, the output of each node is passed through an activation function. The activation function determines whether a node should be activated and introduces non-linearity into the model to enable the resolution of complex tasks. Activation functions include, for example, the sigmoid, which generally converts the output to a value between 0 and 1; the ReLU (Rectified Linear Unit), which outputs 0 for negative inputs and the input value itself for positive inputs; and the softmax, which converts the output into a probability distribution and is used for classification tasks, particularly based on probabilities across multiple classes.

[0091] According to one embodiment of the present invention, the loss function of an artificial intelligence model is a function for measuring how far the model's prediction deviates from the actual target value. The learning objective of the artificial intelligence model is to minimize this loss function. Adjusting the model's weights and biases can reduce the loss function and improve the model's accuracy. The loss function may include, for example, the Mean Squared Error (MSE), which measures the squared difference between the predicted value and the actual value, and the Cross-Entropy Loss, which is used to measure the difference between the predicted probability distribution and the actual distribution in classification tasks. MSE can be commonly used in regression tasks, and Cross-Entropy Loss can be used in the field of image processing, but these are merely simple examples and are not limited thereto.

[0092] According to one embodiment of the present invention, to minimize the loss function, an AI model uses an optimization process, such as Gradient Descent or Backpropagation. Gradient Descent involves iteratively adjusting the model's parameters (weights and biases) in a direction that reduces the loss function, and the gradient can inform the model of how much and in which direction each weight should be adjusted to reduce the loss. Backpropagation is the process of updating the weights of nodes in a neural network by propagating the error of the output backward through the layers. That is, through backpropagation, weights can be adjusted layer by layer while working from the output toward the input.

[0093] According to one embodiment of the present invention, an artificial neural network (artificial intelligence model) may include an input layer, hidden layers, and an output layer. The input layer is a layer into which data is input to the model, and each node of this layer represents a function of the input data. The hidden layer is a layer located between the input layer and the output layer, and the hidden layer applies transformations to the data to enable the model to learn complex patterns. The output layer is a layer that provides the final output of the model, and for example, in a classification task, the output may be a probability vector corresponding to various classes.

[0094] According to one embodiment of the present invention, types of artificial intelligence models include feedforward neural networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformer-based neural networks, in which information flows in one direction from input to output through neuron layers. CNNs are designed for image and video processing and use convolutional layers to scan images into small patches and learn spatial layers of features (edges, textures, etc.). CNNs can be used in fields such as image classification, object detection, and segmentation. RNNs are artificial neural networks used for sequential data such as time series or text. Since RNNs process one element of a sequence at a time and the output depends on the current input and previous calculations, they may be modified to include Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) for long-term dependencies. Since Transformers use an attention mechanism, the model can focus on relevant parts of the input at various processing stages; they are powerful at handling long-term dependencies and are widely used in natural language processing, such as BERT and GPT.

[0095] According to one embodiment of the present invention, an artificial intelligence model is trained using a large dataset that is typically divided into a training set, a validation set, and a test set. The training set is data used to train the model and update weights. The validation set is a subset of data that is not used during training but is used to tune hyperparameters and prevent overfitting. The test set is a completely separate dataset used to evaluate model performance after training. The training process involves multiple epochs in which the model checks the same data multiple times and updates weights at each step. Model performance can be monitored using metrics such as accuracy, precision, recall, or F1 score depending on the task.

[0096] According to one embodiment of the present invention, when a model performs well on training data but performs poorly on new data that has not yet been seen, this is called overfitting, which can occur because the model learns not only the patterns of the data but also the noise. Therefore, techniques such as regularization, which applies a penalty to large weights, dropout, which randomly deletes nodes during training, and early stopping, which stops training before overfitting occurs, can help prevent overfitting by simplifying the model.

[0097] The present invention aims to overcome the limitations of existing systems by integrally performing real-time conversation analysis, question generation, data extraction, and verification. Existing technologies suffered from the problem of reduced efficiency and accuracy in the litigation preparation process due to the omission or incomplete storage of important information during conversations. The present invention utilizes natural language processing technology and a pre-trained legal pattern database to analyze client utterances, automatically generate questions based on the case context, and verify extracted data to supplement missing information. Through this, the analysis and utilization of data generated in real-time during the consultation process are automated, enabling the provision of more reliable legal services to lawyers and clients. The present invention will be described in detail below with reference to the drawings.

[0098] The present invention may provide a client conversation collection and analysis system for counseling. Specifically, the client conversation collection and analysis system for counseling may be a system of a processor unit (130) of a user terminal (100). Additionally, the client conversation collection and analysis system for counseling may be a system of a server device (200). The client conversation collection and analysis system for counseling may include a conversation collection tool (131, 210) for collecting voice information or text information from a client, and a natural language processing and data extraction tool (132, 220) for analyzing the collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information. Additionally, the conversation collection tool (131, 210) is configured to perform conversion of the collected voice information into text information using voice recognition technology.

[0099] Additionally, the conversation collection tool (131, 210) is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data on the collected text information or the converted text information. Additionally, the conversation collection tool (131, 210) is configured to perform an analysis of the client's emotional state on the collected voice information or the collected text information.

[0100] According to one embodiment of the present invention, a client conversation collection and analysis system for consultation further includes an advanced factual relationship analysis and supplementary tool (134, 240). At this time, the conversation collection tool (131, 210) can make a determination regarding whether there is missing information regarding the collected voice information or the collected text information, and if it is determined by the determination made that there is missing information, the advanced factual relationship analysis and supplementary tool (134, 240) is configured to transmit the determined missing information in order to generate additional questions for the client.

[0101] Additionally, the natural language processing and data extraction tool (132, 220) is configured to perform syntactic analysis on the collected voice information or text information and to extract the factual relationship information by the performed syntactic analysis. Additionally, the natural language processing and data extraction tool (132, 220) is configured to extract at least one entity among a subject, date, place, and event using Named Entity Recognition (NER) technology on the collected voice information or text information, and to extract the factual relationship information based on the extracted entity. Additionally, the natural language processing and data extraction tool (132, 220) is configured to extract at least one entity among subject information, date information, place information, weather information, and event information on the collected voice information or text information, and if there is one or more extracted entities, to identify the relationship between the one or more entities and to extract the factual relationship information based on the identified relationship.

[0102] Additionally, the natural language processing and data extraction tool (132, 220) is configured to extract at least one entity among subject information, date information, place information, weather information and event information from the collected voice information or text information, compare the extracted entity with a pre-trained legal pattern database, and identify missing information based on the comparison.

[0103] Additionally, the natural language processing and data extraction tool (132, 220) is configured to perform an analysis to determine the context by associating the currently collected voice information or text information with the voice information or text information previously provided by the client. Additionally, the natural language processing and data extraction tool (132, 220) is configured to learn the structural patterns of the collected text information by utilizing Generative Pre-trained Transformer (GPT) technology.

[0104] Additionally, the natural language processing and data extraction tool (132, 220) is configured to understand the context of the collected text information in both directions by utilizing BERT (Bidirectional Encoder Representations from Transformers) technology. Additionally, the natural language processing and data extraction tool (132, 220) is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, and to manage the relationships between the extracted entities by utilizing Graph Database technology.

[0105] The client conversation collection and analysis system for consultation according to the present invention includes a conversation collection tool (131, 210) for collecting voice information or text information from the client, a natural language processing and data extraction tool (132, 220) for analyzing the collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information, and a pre-trained legal pattern database tool (133, 230) for generating additional questions to supplement missing or incomplete information based on the extracted factual relationship information.

[0106] Additionally, the pre-learned legal pattern database tool (133, 230) is configured to learn and store at least one of the following data: case law data by type of litigation, statistical data on judgment results, standard data by item of damages, data on adjustment of damages by percentage of negligence, negligence data by type of law, standard data on evidence by item of damages, litigation procedure and evidence submission data, legal time and deadline data, data on the likelihood of winning by case type, medical dispute data, real estate litigation data, labor law related data, case law-based damages standard data, negligence compensation data by type of litigation, legal negligence and compensation mapping data, legal accident analysis and compensation data, and case law-based litigation outcome prediction data.

[0107] Additionally, the pre-learned legal pattern database tool (133, 230) is configured to perform automation of the client's condition recording when the client's request is in the field of medical consultation based on the extracted factual information. Additionally, the pre-learned legal pattern database tool (133, 230) is configured to store at least one of the following: diagnostic result information related to the specific symptom, treatment information related to the specific symptom, and drug information related to the specific symptom, in advance according to the specific symptom information, and to integrate and store at least one of the following based on the stored information: the client's specific symptom information, diagnostic result information related to the client's specific symptom, treatment information related to the client's specific symptom, and drug information related to the client's specific symptom.

[0108] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) is configured to search for cases similar to the collected voice information or text information based on the extracted factual relationship information. Additionally, the pre-trained legal pattern database tool (133, 230) is configured to compare the similarity between the extracted factual relationship information and the searched similar cases by utilizing a similarity calculation.

[0109] Additionally, the pre-trained legal pattern database tool (133, 230) is configured to perform the similarity calculation using at least one of a vector-based similarity calculation technique, a text-based similarity calculation technique, an image data similarity technique, a deep learning-based similarity technique, a Cosine Similarity technique, and a BERT-based sentence embedding technique. Additionally, the pre-trained legal pattern database tool (133, 230) is configured to identify missing information by comparing the extracted factual information with pre-stored essential information by litigation type.

[0110] Additionally, the conversation collection tool (131, 210) is configured to convert the collected voice information into text information using voice recognition technology. Additionally, the conversation collection tool (131, 210) is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data on the collected text information or the converted text information. Additionally, the conversation collection tool (131, 210) is configured to perform an analysis of the client's emotional state on the collected voice information or the collected text information. Additionally, the pre-trained legal pattern database tool (133, 230) is configured to infer additional information from the extracted factual relationship information based on the learned and stored information.

[0111] The present invention includes a client conversation collection and analysis system for counseling. The client conversation collection and analysis system for counseling according to the present invention includes a conversation collection tool (131, 210) for collecting voice information or text information from the client, a natural language processing and data extraction tool (132, 220) for analyzing the collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information, and an advanced factual relationship analysis and supplementation tool (134, 240) for performing factual relationship analysis based on the extracted factual relationship information.

[0112] Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) is configured to pre-learn a legal pattern database and to identify missing information by comparing the pre-learned legal pattern database with the extracted factual relationship information. Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) is configured to generate additional questions based on the identified missing information and provide them to the client. Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's emotional state based on the collected voice information or text information, and when generating the additional questions, it is configured to adjust the expression method of the additional questions based on the analyzed emotional state.

[0113] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's understanding of the content of the consultation based on the collected voice information or text information, and when generating the additional question, is configured to adjust the difficulty of the additional question based on the analyzed understanding. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's speech level based on the collected voice information or text information, and when generating the additional question, is configured to adjust the difficulty of the additional question based on the analyzed understanding.

[0114] Additionally, the advanced factual analysis and supplementary tool (134, 240) is configured to determine the priority of the additional questions generated when the additional questions are generated, and is configured to dynamically adjust the determined priority when new information is provided during the consultation with the client. Additionally, the advanced factual analysis and supplementary tool (134, 240) is configured to generate multi-stage questions to collect necessary information step by step when the additional questions are generated.

[0115] The present invention relates to a client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and a counseling interface tool for delivering additional questions or answers based on said extracted factual relationship information. Additionally, the counseling interface tool (135, 250) is configured to provide said additional questions or answers in voice or text.

[0116] The present invention relates to a method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; and a step of extracting factual relationship information based on the separated sentences.

[0117] Additionally, the above preprocessing step further includes a filtering step for removing noise or sounds from the collected voice data.

[0118] Additionally, the preprocessing step further comprises: a step of distinguishing fixed noise and non-fixed noise in the collected voice data; a step of removing the fixed noise by utilizing at least one filtering technique among a frequency domain filtering technique, a spectrum attenuation technique, and a linear prediction coding technique for the fixed noise; and a step of removing the non-fixed noise by utilizing at least one filtering technique among an adaptive filtering technique, a waveform-based noise removal technique, and a masking-based noise removal technique for the non-fixed noise.

[0119] In addition, the above preprocessing step further includes a restoration step that supplements voice data damaged in the above filtering step.

[0120] Additionally, the preprocessing step further includes a step of removing noise or noise from the collected speech data by utilizing at least one model among Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) models, Generative Adversarial Networks (GANs), Transformer-based models, and Microsoft’s Deep Noise Suppression (DNS) models.

[0121] Additionally, the preprocessing step further includes an abnormal input detection step for detecting abnormal input in the collected voice data or text information; and a step for removing or modifying the detected abnormal input.

[0122] Additionally, the abnormal input detection step further includes a step performed by at least one model among a Recurrent Neural Networks (RNNs) model, a Long Short-Term Memory (LSTM) model, a Transformer-based model, a Noise Detection model, a Signal Classification model, an Anomaly Detection model, and a Named Entity Recognition (NER) model.

[0123] Additionally, the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of correcting duplicate spaces, correcting incorrect spacing, standardizing case, and converting dates, times, and amounts within the text into a consistent format for the converted text information and the collected text information.

[0124] Additionally, the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of deleting unnecessary elements, deleting duplicate words, deleting meaningless words, removing incomplete sentences, deleting symbols that interfere with analysis, adjusting punctuation, and adjusting spacing on the converted text information and the collected text information.

[0125] Additionally, the preprocessing step further includes: a step of converting the collected voice data into text information; and a step of performing at least one of converting the date data within the text into an ISO standard, unifying the time within the text into a 24-hour system, converting numbers within the text into numerical forms if they are expressed as characters, and converting units within the text into a standardized form for the converted text information and the collected text information.

[0126] Additionally, the preprocessing step further includes: a step of converting the collected voice data into text information; and a step of converting various expressions having the same meaning within the text into a single unified term for the converted text information and the collected text information.

[0127] Additionally, the sentence tokenization step further comprises: a step of detecting punctuation including at least one of a period, a question mark, and an exclamation mark within the text of the generated standardized text information; a sentence end identification step of identifying the end of a sentence based on the detected punctuation; and a step of dividing the text into sentence units by dividing the identified end of the sentence.

[0128] Additionally, the sentence end identification step further includes a sentence boundary ambiguity resolution step that processes the detection of the punctuation mark so as not to identify the end of the sentence based on the detection of the punctuation mark if the detected punctuation mark does not signify the end of the sentence.

[0129] Additionally, the sentence tokenization step further includes: a step of identifying the end of a sentence based on the contextual meaning or conjunction of the text of the generated standardized text information; and a step of dividing the text into sentence units by dividing the identified end of the sentence.

[0130] The present invention relates to a method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of said client; a preprocessing step of preprocessing said collected voice data or text information to generate standardized text information; a part-of-speech tagging step of identifying the part of speech of each word in said standardized text information and assigning tags; and a step of extracting factual relationship information based on said tagged part of speech.

[0131] Additionally, the part-of-speech tagging step further includes: a step of separating the generated standardized text information into words; and a step of determining which part of speech the separated words belong to among nouns, verbs, adjectives, adverbs, and prepositions.

[0132] Additionally, the part-of-speech tagging step further includes the step of identifying the part of speech of each word and assigning tags by utilizing at least one model among a BiLSTM-CRF model, a Transformer-based model, a word vector model, and a Pre-trained Embedding model.

[0133] The present invention may provide a method for collecting and analyzing client conversations for consultation by utilizing an artificial intelligence model based on natural language processing. The method for collecting and analyzing client conversations for consultation may include: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; a factual relationship information extraction step of extracting factual relationship information based on the separated sentences; and a feedback provision step of providing feedback to the client based on the extracted factual relationship information.

[0134] According to one embodiment of the present invention, standardized text information may be information converted into text from voice information input by a user. Additionally, standardized text information may be information converted into a consistent form by refining unstructured text information according to rules. Additionally, standardized text information may be information from which special characters, duplicate spaces, or unnecessary words that are not required for analysis have been removed. Additionally, standardized text information may be information from which various expressions having the same meaning have been converted into a single standard expression. For example, it may be information converted from January 1, 2024 to 1 / 1 / 2024. Additionally, standardized text information may be information from which spelling errors have been corrected. Additionally, standardized text information may be information from which the spelling of collected text information has been corrected. Additionally, standardized text information may be information from which abbreviations or informal expressions of collected text information have been converted into formal expressions.

[0135] Additionally, the step of extracting factual information further includes a step of identifying missing information by comparing the collected voice data or text information with essential information templates for each lawsuit type stored in a database.

[0136] Additionally, the feedback providing step further includes: a step of generating additional questions related to the identified missing information when information missing from the factual relationship information extraction step is identified; and a step of providing the generated additional questions.

[0137] Additionally, the feedback provision step further includes: a step of generating customized questions suitable for the client based on the extracted factual information; and a step of providing the generated additional questions.

[0138] Additionally, the feedback providing step further includes: a step of analyzing the emotional state of the client based on the collected voice data or text information of the client; a step of generating additional questions based on the analyzed emotional state of the client; and a step of providing the generated additional questions.

[0139] Additionally, the feedback providing step further includes: a step of analyzing the client's understanding of the content of the consultation based on the collected voice data or text information of the client; a step of generating additional questions based on the analyzed client's understanding of the content of the consultation; and a step of providing the generated additional questions.

[0140] Additionally, the feedback provision step further includes: a step of generating additional questions based on previous questions or previous answers by utilizing natural language processing (NLP) technology to reflect the flow and context of the conversation of the collected client's voice data or text information; and a step of providing the generated additional questions.

[0141] FIG. 1 is a diagram illustrating the configuration of a user terminal and a server device according to an embodiment of the present invention.

[0142] sensor unit (110)

[0143] As described above, the user terminal (100) of the present invention may include a sensor unit (110), a communication unit (120), a processor unit (130), a memory unit (140), and an output unit (150). Specifically, the sensor unit (110) may sense voice information of a user. That is, the sensor unit (110) may collect voice data generated from a user. At this time, voice information may be physically detected by including an acoustic sensor such as a microphone. In addition, the sensor unit (110) may include a function to remove or filter ambient noise or static to recognize voice data more accurately. Through this, the sensor unit (110) can transmit clear voice data to the processor unit (130). In addition, when used in a complex environment, the sensor unit (110) may perform a function to detect the direction of the user's voice and improve the quality of voice collection using a multi-microphone system.

[0144] According to one embodiment of the present invention, the sensor unit (110) may include a MEMS microphone (Micro-Electro-Mechanical Systems) or a condenser microphone. Specifically, the sensor unit (110) serves to convert sound into an electrical signal, and the MEMS microphone is suitable for small electronic devices and can be efficiently mounted on various user terminals. Additionally, the sensor unit (110) may include an amplifier that can amplify the signal and transmit it more clearly if the voice signal is too weak. Furthermore, since the sensor unit (110) must be able to accurately detect voice even in a noisy environment, a method of intensively collecting voice coming from a specific direction using beamforming technology, etc., may be applied, but is not limited thereto.

[0145] Communications Department (120)

[0146] As described above, the user terminal (100) of the present invention may include a communication unit (120). Specifically, the communication unit (120) may be composed of a network tool capable of using various communication technologies, such as Wi-Fi, Bluetooth, and cellular networks (LTE, 5G). That is, the communication unit (120) may provide a physical or wireless communication path for the user terminal (100) to be connected to a server device (200).

[0147] According to one embodiment of the present invention, the communication unit (120) can transmit information processed by the user terminal (100) or data collected by the sensor unit (110) to the server device (200). That is, the communication unit (120) can perform the function of receiving results or response data processed by the server device (200) back to the user terminal (100). That is, the communication unit (120) can provide bidirectional communication between the user and the server. Specifically, the communication unit (120) can transmit the voice analysis information to the server device (200) when the user terminal (100) collects voice data and analyzes it in the processor unit (130), and the processor unit (130) creates voice analysis information. Accordingly, the communication unit (120) can request the server device (200) to perform additional processing, such as complex natural language processing or data analysis. Additionally, the communication unit (120) can receive results, such as voice responses and analyzed information processed by the server device (200), back to the user terminal (100) through the communication unit (120). At this time, the transmitted data can be delivered to the output unit (150) by the processor unit (130) to provide the user with a response service, but this will be described in detail below.

[0148] Processor unit (130)

[0149] As described above, the user terminal (100) of the present invention may include a processor unit (130). Specifically, the processor unit (130) may receive voice information collected from the sensor unit (110), analyze it in real time, and provide text response information or voice response information based thereon. Additionally, the processor unit (130) may understand the user's voice information and analyze the content by utilizing a natural language processing-based artificial intelligence model implemented in the memory unit (140). This can be achieved through voice recognition and natural language processing (NLP) functions.

[0150] Conversation collection tool (131, 210)

[0151] Capture voice or text information in real time

[0152] According to one embodiment of the present invention, the system of the processor unit (130) or the server (200) may include a client conversation collection and analysis system for consultation. In the client conversation collection and analysis system for consultation, the processor unit (130) or the server (200) may include a conversation collection tool (131, 210). Specifically, the conversation collection tool (131, 210) can capture voice or text information provided by the client in real time. For example, the conversation collection tool (131, 210) can capture the client's voice data through the sensor unit (110) and convert it into text using voice recognition technology. In addition, the conversation collection tool (131, 210) can immediately collect data entered as text from the client. As a specific example, in the case of a traffic accident damage compensation lawsuit, if the client explains the date, time, location, etc. of the accident, the conversation collection tool (131, 210) can capture this in real time and convert it into text information. In addition, in a contract dispute lawsuit, when a client explains the breach of contract and the extent of damages, the conversation collection tool (131, 210) can collect such content and organize it into structured data.

[0153] According to one embodiment of the present invention, the conversation collection tool (131, 210) can be usefully utilized in various industries as well as in the legal field. For example, in the field of medical consultation, voice data in which a patient describes their condition can be captured and converted into text to be automatically entered into an electronic medical record (EMR). For example, if a patient says, "I have had pain in my right knee since last week. It is especially severe when climbing stairs," the conversation collection tool (131, 210) can convert this into text and structure date and symptom information. Additionally, in the field of customer service, the conversation collection tool (131, 210) can capture and analyze customer problems in real time to automate the problem-solving process.

[0154] For example, if a customer inquires, "My internet connection keeps dropping," the conversation collection tool (131, 210) can convert this into text, extract keywords related to the connection problem, and provide solutions such as appropriate network reset procedures based on this. Additionally, in the field of education, the conversation collection tool (131, 210) can capture and analyze student question or answer data in real time to provide personalized learning support. For example, if a student asks, "Why do we use this formula when solving this problem?", the conversation collection tool (131, 210) can analyze the question and provide appropriate explanations based on the learning content or recommend additional learning materials.

[0155] Removal of unnecessary symbols, removal of unnecessary whitespace, removal of duplicate data

[0156] According to one embodiment of the present invention, a conversation collection tool (131, 210) can organize collected data into a processable state. Specifically, the conversation collection tool (131, 210) is configured to convert voice collected through a sensor unit (110) into text, and then perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data. Afterward, the data organized in this way can be analyzed by a natural language processing and data extraction tool (132, 220). For example, if a client speaks, "There was an accident yesterday evening," the sensor unit (110) can sense this, and the conversation collection tool (131, 210) can convert it into a format of organized data such as "2024-11-14 accident occurred."

[0157] According to one embodiment of the present invention, a conversation collection tool (131, 210) can improve the quality of data by removing duplicate or incomplete sentences. For example, in a traffic accident damage compensation lawsuit, the conversation collection tool (131, 210) can integrate sentences repeatedly mentioned by a client, such as "The place of the accident was Seoul. Yes, Seoul," into a single sentence. In addition, in a contract dispute lawsuit, if contract clauses are duplicated, the conversation collection tool (131, 210) can organize them into a single contract clause information.

[0158] Analysis of the client's emotional state

[0159] According to one embodiment of the present invention, a conversation collection tool (131, 210) with respect to collected voice information or collected text information The client's emotional state can be analyzed. That is, the conversation collection tool (131, 210) can analyze the client's utterance to identify the intent of the utterance and analyze the emotional state to reflect it in subsequent question generation and counseling processes. For example, in a traffic accident damages lawsuit, if the client says, "It's too hard," the conversation collection tool (131, 210) can analyze this as an "anxious state." Also, in a contract dispute lawsuit, if the client says, "The damages are huge," the conversation collection tool (131, 210) can use this information as a basis for generating specific questions regarding the amount of damages.

[0160] Judging missing information

[0161] According to one embodiment of the present invention, a conversation collection tool (131, 210) can make a determination regarding whether there is missing information in the collected voice information or collected text information. Additionally, if the conversation collection tool (131, 210) determines by the determination made that there is missing information, it can transmit the determined missing information to the advanced factual analysis and supplementation tool (134, 240) so that the advanced factual analysis and supplementation tool (134, 240) can generate additional questions for the client. For example, in the case of a traffic accident damages lawsuit, if the weather or traffic conditions at the time of the accident are not mentioned in the conversation, the conversation collection tool (131, 210) can recognize this and transmit the information to the advanced factual analysis and supplementation tool (134, 240) so that it can generate additional questions. Additionally, in a contract dispute lawsuit, the conversation collection tool (131, 210) can identify the need for missing contract clauses or evidence. Additionally, the conversation collection tool (131, 210) transmits the collected data to the natural language processing and data extraction tool (132, 220), and can manage the flow of the conversation and maintain a record.

[0162] Natural language processing and data extraction tools (132, 220)

[0163] construe

[0164] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can understand the meaning of text information by analyzing text information collected from a conversation collection tool (131, 210). For example, in the case of a traffic accident damage compensation lawsuit, the natural language processing and data extraction tool (132, 220) can analyze accident date information, location information, time information, perpetrator information, damage details information, etc. from text information provided by the client.

[0165] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) is configured to perform syntactic analysis on collected voice information or text information and to extract factual relationship information through the performed syntactic analysis. As a specific example, the natural language processing and data extraction tool (132, 220) can analyze the sentence "An accident occurred in Seoul at 3:00 PM on January 1, 2024" and structure it by separating the date into "2024-01-01", the place into "Seoul", and the time into "3:00 PM". Additionally, in the case of a contract dispute lawsuit, the natural language processing and data extraction tool (132, 220) can analyze detailed information such as the contract contents, the counterparty's non-performance of obligations, and the amount of damages, and perform conversion into necessary data.

[0166] According to one embodiment of the present invention, the natural language processing and data extraction tool (132, 220) can be usefully utilized in various industries as well as in the legal field. For example, in the field of medical consultation, the recording of a patient's condition can be automated by analyzing text information in which the patient describes their symptoms and condition. The natural language processing and data extraction tool (132, 220) can analyze text information provided by the patient to extract symptoms, the time of occurrence, the location of pain, etc., and store this as structured data in an electronic medical record (EMR) system. For example, if a patient says, "I have had pain in my right shoulder since last week, and it is especially worse when I lift my arm," the natural language processing and data extraction tool (132, 220) can analyze this to extract and record data such as "Symptom: Pain," "Location: Right shoulder," and "Time of occurrence: Last week." In addition, in the field of customer service, customer inquiry data can be analyzed.

[0167] Specifically, the natural language processing and data extraction tool (132, 220) can analyze text information entered by a customer to quickly suggest the cause of the problem and a solution. For example, if a customer asks, "My internet connection has been unstable since last week," the natural language processing and data extraction tool (132, 220) can extract data such as "Problem: Unstable internet connection" and "Period: Since last week," and based on this, recommend an appropriate solution such as rebooting network settings. In addition, in the field of education, it can provide customized learning support by analyzing student question data. For example, if a student asks, "Why is this formula used here?", the natural language processing and data extraction tool (132, 220) can extract "formula" and "reason for application" from the question and provide an appropriate explanation in conjunction with textbook content or recommend related materials.

[0168] Factual Information

[0169] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform analysis of voice information or text information collected from a client by a conversation collection tool (131, 210) using a natural language processing-based artificial intelligence model. Accordingly, the natural language processing and data extraction tool (132, 220) can extract factual relationship information based on the voice information or text information collected from the client. Specifically, the factual relationship information may include information on the subject of the incident, information on the act or incident, temporal information, spatial information, causal relationship information, information on evidence and materials, and information on damage items. For example, the natural language processing and data extraction tool (132, 220) can extract information on the subject of the incident based on the voice information or text information collected from the client. At this time, the information on the subject of the incident may be information capable of identifying key figures or organizations involved in the incident. For example, the information on the subject of the incident may include information on the client, the opposing party who is the defendant or perpetrator, witnesses, and third parties. For a specific example, the information regarding the parties involved in the incident could be that the client's name is Kim Cheol-su, the opposing party is the driver of the offending vehicle, and the third party is the insurance company.

[0170] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual information based on voice information or text information collected from a client. Specifically, the factual information may include information on actions or events. For example, the information on actions or events may be information regarding actions or events that are central to the issue, such as information on contract conclusion, information on accident occurrence, or information on claims for damages. As a specific example, the information on actions or events may include information on traffic accident occurrence, information on contract violation, or information on industrial accident occurrence.

[0171] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual information based on voice information or text information collected from a client. Specifically, the factual information may include temporal information. For example, the temporal information may be specific date and time information where an event occurred, including information on the date of occurrence of the event, information on the date of signing the contract, information on the repayment deadline, etc. As a specific example, the temporal information may include information on January 1, 2024, information on 10:00 AM, etc.

[0172] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual information based on voice information or text information collected from a client. Specifically, the factual information may include spatial information. For example, the spatial information may include geographical information related to the location where an event occurs, such as information on the location where an accident occurred or information on the location where a contract was concluded. As a specific example, the spatial information may include information on Gangnam-gu, Seoul, crosswalk information, etc.

[0173] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual relationship information based on voice information or text information collected from a client. Specifically, the factual relationship information may include causal relationship information. For example, the causal relationship information may include information explaining the cause and effect between events, such as information on the cause of an accident and information on the circumstances of damage occurrence. As a specific example, the causal relationship information may include information on an accident caused by a signal violation, information on damage caused by a breach of contract conditions, etc.

[0174] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual information based on voice information or text information collected from a client. Specifically, the factual information may include evidence and data information. For example, the evidence and data information may include evidence and data information related to an incident, such as contract information, photographic information, medical certificate information, video data information, etc. As a specific example, the evidence and data information may include traffic accident scene photographic information, hospital medical certificate information, etc.

[0175] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract factual information based on voice information or text information collected from a client. Specifically, the factual information may include damage item information. For example, the damage item information may include specific damage item information resulting from an event, such as information on medical expenses, repair expenses, and mental damage. As a specific example, the damage item information may include information on alternative transportation costs of 500,000 won, information on hospitalization treatment costs of 2,000,000 won, etc.

[0176] Entity extraction

[0177] According to the present invention, an entity may refer to a specific person or substance in data analysis, natural language processing, etc., and may refer to an element containing specific information such as a person, place, object, event, or concept.

[0178] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract information related to an event from collected voice information or text information. Specifically, the natural language processing and data extraction tool (132, 220) can extract at least one entity among subject information, date information, place information, and event information using Named Entity Recognition (NER) technology. Additionally, the natural language processing and data extraction tool (132, 220) can extract factual relationship information based on the extracted entities.

[0179] According to one embodiment of the present invention, in a traffic accident damage compensation lawsuit, the natural language processing and data extraction tool (132, 220) extracts entities such as "accident," "date," and "place," and in a contract dispute lawsuit, it can identify specific information such as "contract date," "opponent's obligation," and "details of violation." For example, in the sentence "Kim Cheol-su had an accident in Seoul on January 1, 2024," the natural language processing and data extraction tool (132, 220) can generate data such as {"Subject": "Kim Cheol-su", "Date": "2024-01-01", "Accident Place": "Seoul", "Incident": "Accident occurred"}.

[0180] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can identify and extract key entities, such as subject information, date information, place information, and event information, from text using Named Entity Recognition (NER) technology. For example, when the sentence "On January 1, 2024, an accident occurred in Gangnam-gu, Seoul, when Kim Cheol-su collided with a vehicle that violated a traffic signal," the natural language processing and data extraction tool (132, 220) can extract "Kim Cheol-su" as subject information, "January 1, 2024" as date information, "Gangnam-gu, Seoul" as place information, and "traffic signal violation" and "accident occurred" as event information. Additionally, the natural language processing and data extraction tool (132, 220) "Kim Cheol-su was involved in an accident in Gangnam-gu, Seoul on January 1, 2024, and the cause of the accident was a vehicle that violated a traffic signal." This allows for structuring information such as the subject, time, place, cause, and result of the incident. In other words, the natural language processing and data extraction tool (132, 220) can extract factual information based on the extracted entities, and can use the extracted factual information to classify the type of incident as a "traffic accident" or to analyze liability by identifying that the vehicle that violated the traffic signal was the cause of the accident.

[0181] Extract relationships

[0182] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract at least one entity among subject information, date information, place information, weather information, and event information from collected voice information or text information, and if there is one or more extracted entities, it can identify the relationship between one or more entities and extract factual relationship information based on the identified relationship. For example, in the case of a traffic accident damage compensation lawsuit, the natural language processing and data extraction tool (132, 220) can extract the relationship between “subject (Kim Cheol-su) - action (accident) - place (Seoul) - date (2024-01-01)” from the sentence “Kim Cheol-su had an accident in Seoul on January 1, 2024.” In addition, in a contract dispute lawsuit, a natural language processing and data extraction tool (132, 220) can analyze the relationship between the other party, the breach of contract, and the date in the sentence “The other party violated the terms of the contract on May 1, 2023.”

[0183] Check for missing information

[0184] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract at least one entity among subject information, date information, place information, weather information and event information from collected voice information or text information, compare the extracted entity with a pre-trained legal pattern database, and identify missing information based on the comparison. For example, in a traffic accident damages lawsuit, the natural language processing and data extraction tool (132, 220) can detect when additional information, such as weather or traffic conditions at the time of the accident, is needed. Additionally, the natural language processing and data extraction tool (132, 220) can identify missing items in contract clauses or parts where relevant evidence is lacking in a contract dispute lawsuit. For example, if a comparison with a database reveals that the "location of the accident" exists but "weather information" is missing, the natural language processing and data extraction tool (132, 220) can generate a trigger to supplement this.

[0185] Contextual analysis

[0186] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform analysis to determine context by associating currently collected voice information or text information with voice information or text information previously provided by the client. Specifically, the natural language processing and data extraction tool (132, 220) can analyze new information emerging from the current conversation by associating it with information previously provided by the client. In the case of a traffic accident damage compensation lawsuit, if the client says, "There was an accident yesterday. The location is near my house," the natural language processing and data extraction tool (132, 220) can infer this as "2024-11-14, Gangnam-gu, Seoul" and store it in the data. Additionally, in a contract dispute lawsuit, the natural language processing and data extraction tool (132, 220) can analyze a sentence such as "I signed the contract 3 years ago" to calculate and store the date the contract was signed.

[0187] Data structuring

[0188] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) extracts information JSON, CSV, etc. It can be converted into a structured format and transmitted to an advanced factual relationship analysis and supplementation tool (134, 240) or a data storage and management tool (136, 260). The natural language processing and data extraction tool (132, 220) of the present invention can effectively analyze collected text information and understand its meaning by utilizing the latest natural language processing technology.

[0189] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) is Context can be understood bidirectionally using BERT (Bidirectional Encoder Representations from Transformers) technology. Through this, natural language processing and data extraction tools (132, 220) It accurately identifies specific information such as the date of the accident, and even the same expression can be interpreted differently depending on the context. For example, when the sentence “On January 1, 2024, a traffic accident occurred in Seoul when Kim Cheol-su collided with a vehicle that ran a red light,” is input, the natural language processing and data extraction tool (132, 220) By utilizing BERT, the relationship between words in a sentence can be analyzed bidirectionally. That is, a natural language processing and data extraction tool (132, 220) can accurately understand the context that a vehicle violated a traffic signal, as the word "violation" in the phrase "vehicle that violated the signal" is connected to both the preceding "signal" and the following "vehicle."

[0190] Based on such contextual understanding, BERT can extract key entities from a sentence. In the sentence above, "Kim Cheol-su" is extracted as subject information, "January 1, 2024" as date information, "Seoul" as place information, and "traffic accident" and "signal violation" as event information. Furthermore, BERT does not stop at merely identifying words but can differentiate the meanings of words and sentences by analyzing the context. For example, in "vehicle that violated the signal," "violation" refers to a violation of traffic laws, whereas in "person who violated the contract," it refers to a violation of contract conditions; BERT can understand this difference through context. Based on the extracted entities, the natural language processing and data extraction tool (132, 220) can derive factual relationship information of the event. For example, the natural language processing and data extraction tool (132, 220) It is possible to derive the factual relationship that the subject of the incident is "Kim Cheol-su," the time of the incident is "January 1, 2024," the place is "Seoul," the cause is "a vehicle that ran a red light," and the result is "a traffic accident."

[0191] This information can be used to classify types of traffic accidents or to analyze the causes of accidents to determine liability. For example, if a "vehicle running a red light" is identified as the cause of the accident, it can increase the likelihood that the perpetrator's liability will be set at 100%.

[0192] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) is configured to learn structural patterns of collected text information by utilizing Generative Pre-trained Transformer (GPT) technology. Accordingly, the natural language processing and data extraction tool (132, 220) enables the generation of natural responses by analyzing complex sentences. For example, it can analyze the sentence "There was an accident near my house" and infer additional location information based on the context. In addition, Named Entity Recognition (NER) technology is applied to extract entities such as names, dates, and places from the text, and the precision of the analysis is further enhanced through combination with BERT.

[0193] Graph Database

[0194] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can extract at least one entity among subject information, date information, place information, weather information and event information from collected voice information or text information, and utilize graph database technology to store the relationships between the extracted entities by representing them as nodes and dpt.

[0195] Specifically, natural language processing and data extraction tools (132, 220) can utilize a graph database to provide a data structure that allows for the visual and intuitive management of relationships between entities. For example, in a traffic accident damages lawsuit, relational information such as "Kim Cheol-su (subject) -> accident (action) -> Seoul (place)" can be stored by representing it as nodes and edges, thereby allowing for the effective tracking and analysis of associations, temporal flow, and causal relationships between events.

[0196] Furthermore, the extracted information can be stored in JSON format to facilitate data transfer and utilization to other tools; for example, data can be stored in a structure such as {"Subject": "Kim Cheol-su", "Date": "2024-01-01", "Accident Location": "Seoul", "Incident": "Accident Occurrence"}. Additionally, the present invention can provide the effect of enhancing search efficiency by classifying data according to the type of litigation and utilizing a domain-specific indexing method. Specifically, in a traffic accident damages lawsuit, "accident date" or "perpetrator information" can be set as key values, and in a contract dispute, "contract date" or "details of violation" can be set to quickly search for relevant information.

[0197] Model training and optimization

[0198] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can continuously optimize a trained model by utilizing a domain-specific database. For example, the natural language processing and data extraction tool (132, 220) can be trained to prioritize the extraction of essential information, such as the "date of the accident," in a traffic accident damages lawsuit. Additionally, the natural language processing and data extraction tool (132, 220) can provide a model optimized for identifying missing items in contract terms in a contract dispute lawsuit.

[0199] Pre-trained legal pattern database tool (133, 230)

[0200] Data Management by Litigation Type

[0201] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) is configured to learn and store at least one of the following: case law data by type of litigation, statistical data on judgment results, standard data by item of damages, data on adjustment of compensation amount by negligence ratio, negligence data by type of law, standard data on proof by item of damages, litigation procedure and evidence submission data, legal time and deadline data, data on likelihood of winning by case type, medical dispute data, real estate litigation data, labor law related data, case law-based compensation standard data, negligence compensation data by type of litigation, legal negligence and compensation mapping data, legal accident analysis and compensation data, and case law-based litigation result prediction data.

[0202] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can learn and store case law data by type of litigation. For example, the pre-learned legal pattern database tool (133, 230) can learn and store data that systematically organizes major case precedents and case results according to the type of litigation, such as case precedents for traffic accident damages, case precedents for breach of contract, and case precedents for industrial accidents.

[0203] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can learn and store statistical data on case results. For example, the pre-learned legal pattern database tool (133, 230) can learn and store statistical information such as the fault ratio, average compensation amount, and time required to resolve the case in judgments that occurred in specific types of incidents, such as rear-end collisions or traffic signal violations.

[0204] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can learn and store standard data for each compensation item. For example, in the case of a traffic accident, the pre-learned legal pattern database tool (133, 230) can learn and store standard data such as repair costs, replacement transportation costs, medical expenses, and consolation money. In addition, in the case of a contract dispute, it can learn and store standard data such as unpaid amounts, delayed interest, and mental damages.

[0205] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can learn and store data on adjusting compensation amounts according to the ratio of negligence. For example, the pre-learned legal pattern database tool (133, 230) can learn and store data on adjusting compensation amounts, such as adjusting the ratio of compensation when the perpetrator's negligence is 80% or reducing the compensation amount when the victim's negligence exceeds 20%.

[0206] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can learn and store traffic accident fault ratio data. For example, the pre-learned legal pattern database tool (133, 230) can learn and store crosswalk accident fault ratio data, accident fault ratio data while overtaking, highway accident fault ratio data, etc.

[0207] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can systematically manage data according to the type of lawsuit. For example, in a traffic accident damage compensation lawsuit, it is common for the fault of the offending vehicle to be recognized as 100% in a rear-end collision. In such cases, the pre-learned legal pattern database tool (133, 230) can provide criteria for calculating the compensation amount, including repair costs for the damaged vehicle, replacement transportation costs, and other damage compensation items. Additionally, the pre-learned legal pattern database tool (133, 230) can store data indicating that in signal violation accidents, the offending party often bears 80% or more of the fault, and that the fault ratio may be adjusted if the victim exceeded the speed limit. In addition, a pre-trained legal pattern database tool (133, 230) can store data indicating that in the case of a pedestrian accident, if it occurs at a crosswalk, the vehicle's fault is usually treated as 100%, and even if the pedestrian jaywalked, the vehicle's fault is likely to be recognized as 60% or more.

[0208] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) in a contract dispute lawsuit can learn that in the event of non-performance of a contract, a penalty is calculated in the range of 10 to 30 percent of the contract amount, or that the principle of disadvantage to the drafter may be applied if the contract contains unclear clauses. In addition, the pre-learned legal pattern database tool (133, 230) can store data on the possibility that the termination will be deemed invalid if the specified conditions (e.g., written notice 30 days in advance) are not observed at the time of contract termination, and accordingly, can provide cases in which damages resulting from the termination can be claimed.

[0209] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can recognize and store as a pattern that in labor contract lawsuits, if wage arrears occur, a delay interest of 15% per year is charged, and if unfair dismissal is recognized, a demand for reinstatement or compensation for damages after dismissal is possible. Additionally, the pre-learned legal pattern database tool (133, 230) can store information that in medical malpractice lawsuits, if misdiagnosis or surgical failure occurs, compensation for damages including actual damages and mental distress can be claimed, and generally, when there is clear negligence, the average amount of compensation for damages is calculated to be between 5 million won and 50 million won. Additionally, the pre-learned legal pattern database tool (133, 230) can learn that in trademark infringement lawsuits, if the shape, pronunciation, or meaning of trademarks is similar, the likelihood of confusion is considered high, and the amount of damages is calculated based on sales loss and brand image loss.

[0210] Additionally, the pre-trained legal pattern database tool (133, 230) can store patterns indicating that in construction dispute litigation, a delay penalty of 0.1% of the contract amount per day is applied in case of construction delay, and compensation for damages including additional costs is possible if rework is required due to quality issues. Additionally, the pre-trained legal pattern database tool (133, 230) can store data indicating that in family law litigation, in divorce proceedings, property formed during the marriage is usually divided 50:50, and child support is calculated at 20–30% of the average income depending on the parents' income. Additionally, the pre-trained legal pattern database tool (133, 230) can store data indicating that the spouse responsible for raising the child is likely to receive more property, or that child support may increase depending on the age of the child.

[0211] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can be utilized in various industries in addition to the legal field. For example, in the field of medical consultation, the pre-trained legal pattern database tool (133, 230) can automate the recording of a patient's condition. The pre-trained legal pattern database tool (133, 230) can store specific symptoms, diagnostic results, treatment methods, and related drug information as patterns. For example, if a patient says, "My cough has lasted for three weeks and I have a fever," the pre-trained legal pattern database tool (133, 230) can identify this as a respiratory infection pattern and suggest additional necessary tests or treatment plans. In addition, in the field of customer service, it can automate the problem-solving process to improve the quality of customer service.

[0212] For example, a pre-trained legal pattern database tool (133, 230) can immediately respond to customer inquiries by storing common troubleshooting procedures, such as paper removal methods and device reset procedures related to product usage problems like "paper jam in the printer," as patterns. Accordingly, customer service agents can utilize this data to solve problems more efficiently or, if necessary, automatically receive recommendations for additional support procedures. In addition, in the insurance industry, a pre-trained legal pattern database tool (133, 230) can analyze and patternize insurance claim data. For example, a pre-trained legal pattern database tool (133, 230) can store average claim amounts and processing times in a database according to the type of accident (rear-end collision, side collision, etc.) and the degree of damage in automobile accident insurance claims.

[0213] Search for similar cases

[0214] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can search for cases similar to input data. For example, if a conversation such as "an accident that occurred in Seoul on January 1, 2024" is input, the pre-trained legal pattern database tool (133, 230) can search for similar traffic accident cases and provide reference information regarding the severity of the accident, liability, and the scope of compensation. Additionally, the pre-trained legal pattern database tool (133, 230) can analyze a conversation regarding "breach of contract concluded in 2023" to find similar dispute cases and suggest the feasibility of fulfilling contract conditions and legal solutions. In this case, the pre-trained legal pattern database tool (133, 230) can use Cosine Similarity, BERT-based sentence embedding, etc. Similarity calculations can be used to compare the similarity between input data and cases within the database.

[0215] Verification and supplementation of missing information

[0216] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can identify missing information by comparing essential information by litigation type with extracted data. For example, if a client mentions the location and date of an accident but lacks weather information, the pre-trained legal pattern database tool (133, 230) can recognize the need for a "weather" item and operate to generate a question in an advanced factual analysis and supplementation tool (134, 240). Additionally, in a contract dispute lawsuit, if the client does not mention contract clauses or evidence, the pre-trained legal pattern database tool (133, 230) can determine the need for such information and trigger an additional question.

[0217] Support for creating additional questions

[0218] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can generate additional questions to supplement missing or incomplete information. For example, in a traffic accident damages lawsuit, questions such as "What was the weather like at the time of the accident?" or "Please describe the traffic conditions." Or, in a contract dispute lawsuit, questions such as "What did the other party fail to perform under the contract terms?" or "What is the amount of damages?" can be generated.

[0219] Data-driven reasoning

[0220] According to one embodiment of the present invention, a pre-learned legal pattern database tool (133, 230) can infer additional information based on similar cases and stored patterns. For example, in a traffic accident damages lawsuit, if the location of the accident is "Seoul," the pre-learned legal pattern database tool (133, 230) can suggest information such as "general traffic congestion in Seoul" or "times when accidents are likely to occur" by referring to similar cases. Additionally, in a contract dispute lawsuit, the pre-learned legal pattern database tool (133, 230) can predict the range of compensation amounts generally required in specific types of contract breach cases and provide this to the consultant.

[0221] Advanced factual analysis and supplementary tools (134, 240)

[0222] Identifying missing information

[0223] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of verifying whether all necessary information has been secured in the litigation preparation process based on data analyzed by the natural language processing and data extraction tool (132, 220). Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can identify missing information by comparing information extracted from a conversation with a pre-learned legal pattern database, and generate appropriate questions based thereon to provide the effect of improving the completeness of the consultation. For example, in a traffic accident damages lawsuit, if a client says, "An accident occurred in Seoul on January 1, 2024," the advanced factual relationship analysis and supplementary tool (134, 240) can recognize that the information extracted from the conversation (Date: January 1, 2024, Place: Seoul, Event: Accident) is insufficient.

[0224] According to one embodiment of the present invention, an advanced factual analysis and supplementation tool (134, 240) may recognize that additional data, such as weather conditions at the time of the accident, traffic conditions, and information on the offending vehicle, is needed for litigation preparation. Accordingly, the advanced factual analysis and supplementation tool (134, 240) may request additional information from the client by generating questions such as "What was the weather like at the time of the accident?", "Was the road where the accident occurred congested?", or "Do you know the license plate number of the offending vehicle?" to supplement the missing information.

[0225] According to one embodiment of the present invention, in a contract dispute lawsuit, when a client says, "The other party has violated the contract terms," ​​the advanced factual analysis and supplementary tool (134, 240) can identify that while the violation of the contract terms has been mentioned, important information such as the date the contract was concluded or the specific amount of damages is missing. To verify such missing information, the advanced factual analysis and supplementary tool (134, 240) can generate questions such as "When was the contract concluded?" or "What is the amount of damages caused by the other party's failure to fulfill its obligations?"

[0226] According to one embodiment of the present invention, the advanced factual analysis and supplementation tool (134, 240) can be utilized in various industries in addition to the legal field. For example, in the field of medical consultation, the advanced factual analysis and supplementation tool (134, 240) can identify missing data in information provided by a patient and generate questions to supplement it. If a patient says, "I have had pain in my right shoulder since last week," the advanced factual analysis and supplementation tool (134, 240) can recognize that the patient needs additional information, such as the intensity and frequency of the pain and whether the pain occurs in specific situations. Accordingly, the advanced factual analysis and supplementation tool (134, 240) can generate questions such as, "Is the pain continuous, or does it occur only during specific movements?" and "On a scale of 1 to 10, how intense is the pain?"

[0227] According to one embodiment of the present invention, an advanced factual analysis and supplementation tool (134, 240) in the field of customer service can collect additional information necessary for problem solving by analyzing customer inquiry data. When a customer inquires, "My internet connection keeps dropping," the advanced factual analysis and supplementation tool (134, 240) identifies that specific information necessary for problem solving (e.g., the device being used, whether it occurs only at specific times) is lacking and can generate questions such as, "Does the internet connection problem occur only at specific times or always?" and "What is the model of the device being used?" Additionally, in the field of education, information missing from a student's question or answer can be identified and supplemented. When a student says, "I need a formula to solve this problem," the advanced factual analysis and supplementation tool (134, 240) can generate additional questions such as, "What situation does this problem deal with?" and "Which of the previously learned formulas did you try to apply?" to determine what type of problem it is and whether the conditions for applying the formula are met.

[0228] Analysis of understanding and emotional state

[0229] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of analyzing the client's level of understanding and emotional state during the question generation process and adjusting the expression method and difficulty level of the question. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's emotional state and level of speech in detail to optimize the question so that the client can participate in the consultation more comfortably and naturally.

[0230] According to one embodiment of the present invention, emotional state analysis can be performed based on emotional clues appearing in the client's conversation. For example, if the client says, "It's too hard," the advanced factual analysis and supplementary tool (134, 240) can recognize this as an "anxious state." In this case, the advanced factual analysis and supplementary tool (134, 240) can generate soft and comforting questions to relieve the client's psychological burden and facilitate a smooth conversation. For example, instead of a direct question such as "Can you explain the situation at the time of the accident in detail?", the advanced factual analysis and supplementary tool (134, 240) can present a question using a soft expression such as "Could you explain the situation at the time of the accident a little more within a comfortable range?"

[0231] According to one embodiment of the present invention, the level of comprehension analysis can be performed by adjusting the difficulty of the questions based on the client's speech level and the degree of legal knowledge. The advanced factual analysis and supplementary tool (134, 240) can generate more specific and professional questions if the client is judged to frequently use specialized terminology or to have high legal knowledge. Conversely, if the client's speech contains many non-professional expressions or uses everyday language, the advanced factual analysis and supplementary tool (134, 240) can adjust the questions to be simple and easy to understand. For example, if the client is not familiar with complex legal terminology, the advanced factual analysis and supplementary tool (134, 240) can use expressions that are easier to understand, such as "How much is the amount expected due to the damage?" instead of the question "Have you calculated the amount of damage?"

[0232] Maintaining conversation flow and optimizing question order

[0233] According to one embodiment of the present invention, an advanced factual analysis and supplementary tool (132, 240) can maintain the logical flow of conversation and perform the role of optimizing the order of questions to collect information necessary during the litigation preparation process. Specifically, the advanced factual analysis and supplementary tool (132, 240) analyzes the client's conversation, determines the priority of questions to secure the most important information first, and can dynamically adjust questions when new information is provided during the conversation.

[0234] According to one embodiment of the present invention, the priority determination by the advanced factual analysis and supplementary tool (132, 240) can be performed by first asking for information that is essential for preparing a lawsuit, and then asking for supplementary information. Specifically, the advanced factual analysis and supplementary tool (132, 240) can distinguish between essential information and additional information in each type of lawsuit by utilizing domain knowledge of a pre-learned legal pattern database. For example, in the case of a traffic accident damage compensation lawsuit, the advanced factual analysis and supplementary tool (132, 240) can present a question such as "When did the accident occur?" first, as the date and location of the accident constitute the basic information of the case. On the other hand, information regarding road conditions or weather at the time of the accident is considered supplementary information, so a question such as "What were the road conditions at the time of the accident?" can be presented at a later stage.

[0235] According to one embodiment of the present invention, when a client provides new information during a conversation, the advanced factual analysis and supplementary tool (132, 240) can analyze this information to omit questions that are redundant with already obtained information or are unnecessary, and generate alternative questions. For example, if the client says, "The accident happened in Seoul on January 1, 2024," the advanced factual analysis and supplementary tool (132, 240) can automatically omit planned questions such as "When did the accident happen?" Instead, the advanced factual analysis and supplementary tool (132, 240) can generate new questions such as "Do you know the license plate number of the offending vehicle at the time of the accident?"

[0236] Multi-step question generation

[0237] According to one embodiment of the present invention, an advanced factual analysis and supplementation tool (134, 240) can generate multi-stage questions to collect necessary information step-by-step in complex litigation types or conversational situations. For example, in the case of a traffic accident damages lawsuit, Advanced factual analysis and supplementary tools (134, 240)First, to obtain basic accident information, the tool may present the question, "When did the accident occur?" Accordingly, if the client answers, "The accident happened on January 1, 2024," the advanced factual analysis and supplementary tool (134, 240) can store this information in a database and proceed to the next question. Subsequently, the advanced factual analysis and supplementary tool (134, 240) may present the question, "What was the speed of the vehicle at the time of the accident?" to specifically identify the accident situation. Accordingly, if the client answers, "It was about 50 km / h," this can be recorded so that it can be used for accident cause analysis. Accordingly, the advanced factual analysis and supplementary tool (134, 240) may present the question, "What measures did the perpetrator take after the accident?" to understand the post-accident handling process. Accordingly, if the client answers, "The perpetrator just ran away," this can be determined as a hit-and-run and stored as information to review additional legal measures.

[0238] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) may present the question "Which part of the contract was violated?" to identify violations of the contract in a contract dispute lawsuit. Accordingly, if the client answers "The delivery deadline was not met," this can be recorded as a violation. Additionally, the advanced factual analysis and supplementary tool (134, 240) can verify whether negotiations with the other party took place through the question "Have you discussed the violation with the other party?" Accordingly, if the client answers "We discussed it, but they denied responsibility," this result can be added to the database. Subsequently, to verify the result of the discussion, the advanced factual analysis and supplementary tool (134, 240) may ask the question "What was the result of that discussion?" and if the client answers "We ultimately failed to reach an agreement," this can be recorded and utilized for establishing a litigation strategy.

[0239] Consultation Interface Tool (135, 250)

[0240] According to one embodiment of the present invention, a consultation interface tool (135, 250) can smoothly proceed with the consultation process by collecting the client's response in real time, delivering additional questions, and managing the flow of conversation. Specifically, the client receives questions through the consultation interface tool (135, 250), and the consultation interface tool (135, 250) can provide answers in voice or text. For example, in the case of a traffic accident damage compensation lawsuit, if the client inputs "An accident occurred on January 1, 2024" into the user terminal (100), the consultation interface tool (135, 250) can record this data in real time and generate and deliver additional questions.

[0241] According to one embodiment of the present invention, the consultation interface tool (135, 250) can be utilized in various industries in addition to the legal field. For example, in the field of medical consultation, the consultation interface tool (135, 250) allows a patient to describe their condition and receive additional questions. For example, if a patient inputs, "I have had a cough and a fever since last week," the consultation interface tool (135, 250) records this in real time and can supplement the patient's condition record by generating additional questions such as "When does the cough get worse?" or "What is the temperature?"

[0242] According to one embodiment of the present invention, in the field of customer service, a consultation interface tool (135, 250) can efficiently support the customer problem-solving process. For example, when a customer inputs "My mobile phone screen is not working" into the consultation interface tool (135, 250), the consultation interface tool (135, 250) records this in real time and can collect detailed information necessary for problem-solving by generating additional questions such as "Did the screen suddenly turn off, or is the touch not working?" or "Please describe in detail the situation where the screen recently stopped working." Additionally, in the field of education, the consultation interface tool (135, 250) can be utilized to support interaction between students and teachers. For example, when a student inputs "I don't understand this math problem," the consultation interface tool (135, 250) can generate additional questions such as "Which part of the problem do you not understand?" or "Have you tried applying the formulas you learned previously?"

[0243] Data storage and management tools (136, 260)

[0244] According to one embodiment of the present invention, a data storage and management tool (136, 260) can store data generated during the consultation process. Specifically, the security of the data can be maintained by storing conversation content provided by the client, extracted information, and additional data generated during in-depth consultation in an encrypted form in a database. For example, if the client answers, "An accident occurred in Seoul on January 1, 2024," this can be stored in a database in a structured format such as date, location, and type of incident. In addition, the data storage and management tool (136, 260) can remove duplicate items from the collected data and convert information such as dates or amounts into a consistent format to make it easy to use. For example, the client's answer, "An accident occurred early last year," can be converted into "January 1, 2024" and stored. Additional answers provided by the client can be integrated with existing data to supplement missing information. For example, if the answer "It rained" is provided in response to the question "What was the weather like at the time of the accident?", the weather information can be added to the existing data and stored.

[0245] According to one embodiment of the present invention, a data storage and management tool (136, 260) can provide a search function. Specifically, the data storage and management tool (136, 260) can search for data by specific litigation type or date using keywords or filters, and use this for litigation preparation. For example, if a search condition such as "January 1, 2024 accident" is entered, all information related to accident data for that date can be obtained.

[0246] According to one embodiment of the present invention, a data storage and management tool (136, 260) can apply various security technologies and personal information protection methods to securely store data generated during the consultation process and to protect personal information. Specifically, the data storage and management tool (136, 260) can perform encryption on data stored in a database by utilizing AES-256 encryption technology to maintain the security of the data. In addition, the data storage and management tool (136, 260) can maintain the confidentiality of data during network communication by using the Transport Layer Security (TLS) protocol.

[0247] According to one embodiment of the present invention, a data storage and management tool (136, 260) may utilize a data processing process that complies with privacy protection regulations, such as the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA), to enhance privacy protection. Specifically, the data storage and management tool (136, 260) may anonymize sensitive information, such as names, contact information, and addresses, by utilizing Differential Privacy technology. Additionally, the data storage and management tool (136, 260) may apply Role-Based Access Control (RBAC) to control data access, thereby restricting access to data to only authorized users. Specifically, the data storage and management tool (136, 260) may record system logs to track data access history, thereby detecting abnormal data access attempts or potential security threats in real time.

[0248] FIG. 2 is a diagram illustrating the user conversation collection step of a conversation collection tool according to an embodiment of the present invention.

[0249] Dialogue input data collection step (S100)

[0250] Voice data processing step (S100-1)

[0251] According to one embodiment of the present invention, a conversation collection tool (131, 210) can convert a client's voice data into a processable form in the voice data processing step (S100-1). Specifically, in the voice data processing step (S100-1), the conversation collection tool (131, 210) can convert the voice signal into digital data, convert it into text through deep learning-based speech recognition technology, and perform the role of improving the quality of the data by removing unnecessary elements through preprocessing.

[0252] According to one embodiment of the present invention, a conversation collection tool (131, 210) can digitize a voice signal. Specifically, the client's utterance is collected through a microphone or other input device, and the conversation collection tool (131, 210) can convert it from an analog signal into a digital signal. During this process, sampling and quantization are performed so that the main features of the voice signal (e.g., frequency, amplitude) can be stored as digital data. Additionally, the conversation collection tool (131, 210) can perform a voice recognition task to convert voice data into text. In this case, the conversation collection tool (131, 210) can convert the client's utterance into text in real time using a deep learning-based voice recognition model. Specifically, during the voice recognition process, the sentence structure and speaking speed of the utterance are analyzed, and background noise, intonation, and differences in pronunciation can be corrected to improve the recognition rate.

[0253] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform voice data preprocessing operations. Specifically, the conversation collection tool (131, 210) can remove unnecessary noise or errors from voice signals and improve the quality of the data during the preprocessing stage. For example, the conversation collection tool (131, 210) can remove background noise or filter abnormal data such as broken speech. In addition, in the case of data containing multiple speakers, speaker diarization can be performed. Furthermore, the conversation collection tool (131, 210) can organize converted text information. Specifically, the conversation collection tool (131, 210) can separate sentences, remove unnecessary symbols or duplicate data, and organize the case and spacing of sentences to maintain data consistency.

[0254] According to one embodiment of the present invention, Transformer-based models (e.g., Wav2Vec, Whisper) and Recurrent Neural Network (RNN) or Long Short-Term Memory (LSTM) models are primarily used for speech recognition, and Convolutional Neural Network (CNN) and Speaker Diarization models may be used for noise removal and speaker separation.

[0255] Text information processing step (S100-2)

[0256] According to one embodiment of the present invention, a conversation collection tool (131, 210) can organize text information provided by a client into an analyzable form in a text information processing step (S100-2). Specifically, the text information processing step (S100-2) can collect, organize, and preprocess text information. For example, text entered by a client can be collected in real time through a keyboard, a mobile device, or a web interface. The conversation collection tool (131, 210) manages the flow of data without interruption and can record the text provided by the client. For example, the conversation collection tool (131, 210) can record the input "An accident occurred in Seoul on January 1, 2024." Additionally, the conversation collection tool (131, 210) can organize the collected data by removing unnecessary symbols or meaningless elements. For example, if the sentence “January 1, 2024!! An accident??? occurred in Seoul.” is input, the conversation collection tool (131, 210) can organize it into a neat and consistent sentence such as “An accident occurred in Seoul on January 1, 2024.” In addition, the conversation collection tool (131, 210) can also perform case conversion, spacing cleanup, etc.

[0257] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform data preprocessing. Specifically, the conversation collection tool (131, 210) can remove duplicate sentences or filter abnormal inputs. For example, if a sentence such as "The date of the accident is January 1, 2024" is repeatedly entered, the conversation collection tool (131, 210) can integrate it into a single sentence and remove meaningless inputs ("um," "uh," "......"). Afterward, the conversation collection tool (131, 210) can perform data structuring. Specifically, the conversation collection tool (131, 210) can convert text information into an analyzable format and store it for data structuring. For example, when a conversation collection tool (131, 210) receives the sentence “An accident occurred in Seoul on January 1, 2024.”, it can convert this into JSON format and store it as {"Date": "2024-01-01", "Place": "Seoul", "Event": "Accident occurred"}.

[0258] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform emotion and intention analysis. The conversation collection tool (131, 210) can analyze the client's emotional state (positive, negative, neutral) from text information and identify the main intention of the utterance to be used in the subsequent counseling process. For example, if the utterance "It's too hard. The accident handling is too complicated" is input, the conversation collection tool (131, 210) can classify it as "negative emotion" and analyze the intention as "request for damages."

[0259] According to one embodiment of the present invention, in the text information processing step (S100-2), a Transformer-based model (BERT, GPT, etc.) is used for preprocessing text information and analyzing intent and sentiment, and an RNN and LSTM model can learn the temporal order of sentences to supplement incomplete sentences. Additionally, in the text information processing step (S100-2), a Named Entity Recognition (NER) model identifies entities such as dates, places, and events, and Pre-trained Language Models (T5, RoBERTa, etc.) can be utilized for text quality improvement and structuring tasks.

[0260] Voice data format conversion step (S110)

[0261] Voice data conversion step (S110-1)

[0262] According to one embodiment of the present invention, a conversation collection tool (131, 210) can convert voice data provided by a client into text in the voice data to text conversion step (S110-1). Specifically, the conversation collection tool (131, 210) can perform analysis and preprocessing of the voice signal. That is, after collecting voice data, the conversation collection tool (131, 210) can perform preprocessing work to remove background noise and leave only the utterance signal. For example, when a client utters, "There was an accident in Seoul on January 1, 2024," the conversation collection tool (131, 210) can extract a pure voice signal by removing ambient noise or unnecessary noise. Subsequently, the conversation collection tool (131, 210) can perform the task of converting the voice data into text. In this case, the conversation collection tool (131, 210) can analyze the voice signal using a deep learning-based speech recognition model and convert the utterance into words and sentences. At this time, the voice signal is converted into frequency features (MEL frequency spectrum) and used as input data for the model, and the conversation collection tool (131, 210) can learn the temporal continuity and context of the utterance and convert it into text. For example, if voice data such as "An accident occurred in Seoul on January 1, 2024" is received, the conversation collection tool (131, 210) can convert it into text such as "An accident occurred in Seoul on January 1, 2024".

[0263] According to one embodiment of the present invention, when voice data contains multiple speakers, the conversation collection tool (131, 210) can perform speaker separation and utterance separation operations. Specifically, the conversation collection tool (131, 210) can identify each speaker by analyzing the voice characteristics (spectrogram) of the speaker and process the utterances independently to convert them into text. For example, when voice input of two speakers conversing is received, the conversation collection tool (131, 210) can separate and store the text by speaker, such as "Speaker A: When did the accident happen?" and "Speaker B: It is January 1, 2024." Additionally, the conversation collection tool (131, 210) can perform organization and verification on the converted text information. Specifically, the conversation collection tool (131, 210) can correct typos that may occur during the conversion process and organize the sentence structure and spacing to maintain consistency in the text. Additionally, the conversation collection tool (131, 210) can supplement incomplete sentences. For example, if the conversion result is "January 1, 2024. An accident occurred in Seoul.", the conversation collection tool (131, 210) can organize it into "January 1, 2024, an accident occurred in Seoul."

[0264] According to one embodiment of the present invention, in the step of converting voice data into text (S110-1), a Transformer-based model (e.g., Whisper, Wav2Vec 2.0), RNN and LSTM models, a Sequence-to-Sequence model (e.g., DeepSpeech, Kaldi), and a Speaker Diarization model may be utilized. Whisper and Wav2Vec 2.0 are suitable for high-precision speech recognition and text conversion tasks, and RNN and LSTM can convert voice data into text by understanding temporal context. In addition, the Speaker Diarization model can distinguish speakers and process utterances independently.

[0265] Step of organizing text information into a processable structure (S110-2)

[0266] According to one embodiment of the present invention, in the step of organizing text information into a processable structure (S110-2), the conversation collection tool (131, 210) can perform the task of standardizing the converted text information. Specifically, the conversation collection tool (131, 210) can unify uppercase and lowercase letters to maintain data consistency, or convert formats such as dates into a standard format. For example, the conversation collection tool (131, 210) can convert data such as "January 1, 2024" into a standard format such as "2024-01-01", and can improve the readability of the text by removing duplicate spaces or unnecessary spacing.

[0267] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform the task of removing unnecessary elements from text information. Specifically, the quality of the data can be improved by removing meaningless symbols (e.g., !!, ??, ###), duplicate words, or incomplete sentences. For example, the conversation collection tool (131, 210) can organize the text "January 1, 2024!! An accident?? occurred in Seoul." into "An accident occurred in Seoul on January 1, 2024." Afterwards, the conversation collection tool (131, 210) can organize the data into an analyzable structure. Specifically, the conversation collection tool (131, 210) can extract key information such as dates, places, and events from the text and convert it into a standard format such as JSON or CSV. For example, a conversation collection tool (131, 210) can store the text "An accident occurred in Seoul on January 1, 2024." in a structure such as {"Date": "2024-01-01", "Place": "Seoul", "Event": "Accident occurred"}.

[0268] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform sentence completion and context organization tasks for text information. Specifically, the conversation collection tool (131, 210) can organize spacing and punctuation within a sentence to supplement it into a natural sentence, and can correct sentences that are incomplete or have redundant meanings. For example, the conversation collection tool (131, 210) can correct the text "January 1, 2024 accident in Seoul." to "Accident occurred in Seoul on January 1, 2024." Additionally, the conversation collection tool (131, 210) can identify essential information missing from the text information and correct items that do not conform to standards. For example, the conversation collection tool (131, 210) can supplement incomplete data such as "Date: 2024" to "Date: 2024-01-01."

[0269] According to one embodiment of the present invention, various deep learning models may be utilized in the step (S110-2) of organizing text information into a processable structure. Specifically, a Named Entity Recognition (NER) model is used to structure data by identifying key entities in the text, and a Transformer-based model (BERT, RoBERTa, etc.) can analyze context and supplement incomplete sentences. Additionally, a Sequence-to-Sequence model (T5, GPT, etc.) can reconstruct text information, and a Text Normalization model can be used for data standardization.

[0270] Data quality verification and preprocessing step (S120)

[0271] Noise removal step (S120-1)

[0272] According to one embodiment of the present invention, the noise removal step (S120-1) is a process performed to improve the quality of voice data in a conversation collection tool, and the conversation collection tool (131, 210) can perform the role of extracting pure voice data by removing background noise or unnecessary noise from the voice signal. Specifically, in the noise removal step (S120-1), the noise characteristics of the voice data are analyzed, and the conversation collection tool (131, 210) can distinguish between the voice signal and the noise by dividing the collected voice data into frequency and temporal characteristics. Accordingly, the conversation collection tool (131, 210) can separate background noise (e.g., car sounds, keyboard typing sounds, ambient conversation noise) from the voice signal. In addition, the conversation collection tool (131, 210) can evaluate the initial quality of the voice data by calculating the Signal-to-Noise Ratio (SNR) and determine which noise removal technique to apply.

[0273] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform noise filtering and removal operations. Specifically, the conversation collection tool (131, 210) can distinguish between fixed noise, such as mechanical sounds, and non-fixed noise, such as intermittent background noise, in voice data and apply a suitable filtering technique to each. That is, the conversation collection tool (131, 210) can separate signals in the time-frequency domain and remove background noise from voice signals. For example, the conversation collection tool (131, 210) can process data such as "On January 1, 2024, an accident [wind sound] occurred in Seoul [car noise]" and leave only a pure voice signal such as "On January 1, 2024, an accident occurred in Seoul."

[0274] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform noise filtering and removal operations. Specifically, the conversation collection tool (131, 210) can distinguish between stationary noise, such as machine sounds, and non-stationary noise, such as intermittent background noise, in voice data and apply a suitable filtering technique to each. For example, in the case of stationary noise, the conversation collection tool (131, 210) can remove the noise by utilizing a frequency domain filtering technique, which is a method of attenuating or removing specific frequency bands. Specifically, in the case of stationary noise, the conversation collection tool (131, 210) can utilize a notch filter that removes specific frequency bands, a low-pass filter that removes high-frequency noise and retains only low-frequency components, and a high-pass filter that removes low-frequency noise and retains only high-frequency components. In addition, in the case of fixed noise, the conversation acquisition tool (131, 210) can remove the noise by utilizing a Spectral Subtraction technique that estimates the spectrum of the fixed noise in advance and subtracts it from the signal spectrum. In addition, in the case of fixed noise, the conversation acquisition tool (131, 210) can remove the noise by utilizing a Linear Predictive Coding (LPC) technique that models the speech signal and removes unpredicted fixed noise.

[0275] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform noise filtering and removal operations. Specifically, the conversation collection tool (131, 210) can distinguish between stationary noise, such as mechanical sounds, and non-stationary noise, such as intermittent background noise, in voice data and apply a suitable filtering technique to each. For example, in the case of non-stationary noise, the conversation collection tool (131, 210) can utilize an adaptive filtering technique that adjusts filter coefficients in real time according to environmental changes. Additionally, in the case of non-stationary noise, the conversation collection tool (131, 210) can utilize a wavelet denoising technique that decomposes the signal into the time-frequency domain to remove noise components. In addition, for non-fixed noise, the conversation collection tool (131, 210) can remove the noise by utilizing a human auditory model to acoustically attenuate the noise using Masking-Based Noise Reduction.

[0276] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform a task of restoring a voice signal after noise removal. Specifically, the conversation collection tool (131, 210) can perform a restoration task to compensate for voice data damaged during the filtering process and maintain natural voice quality. In addition, the conversation collection tool (131, 210) can verify the quality of the voice data after noise removal is completed. Specifically, the conversation collection tool (131, 210) can verify whether the noise removal task was performed effectively by measuring the SNR, voice bandwidth, and the clarity of the voice signal.

[0277] According to one embodiment of the present invention, deep learning models that can be utilized in this step include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), Generative Adversarial Networks (GANs), Transformer-based models, and Microsoft’s Deep Noise Suppression (DNS) model. Specifically, CNNs and RNNs learn the frequency and temporal features of speech data to remove noise, and GANs can be useful for restoring damaged signals after noise removal. In addition, real-time speech processing models such as DNS can be used to effectively remove noise occurring in various environments.

[0278] Abnormal input filtering step (S120-2)

[0279] According to one embodiment of the present invention, in the abnormal input filtering step (S120-2), the conversation collection tool (131, 210) can detect, remove, or modify abnormal inputs among the collected voice data and text information. Specifically, the conversation collection tool (131, 210) can perform abnormal input detection. For example, the conversation collection tool (131, 210) detects inputs such as excessively long silences, discontinuous utterances, or meaningless sounds like "um," "uh," or "geu" in the voice data, and incomplete sentences or "asdf," "..." in the text information. Meaningless symbols can be identified. Additionally, the conversation collection tool (131, 210) can detect elements in the data that are not suitable for analysis. For example, the conversation collection tool (131, 210) can detect "..." and "um" as abnormal inputs in text such as "January 1, 2024... an accident in Seoul... um...".

[0280] According to one embodiment of the present invention, a conversation collection tool (131, 210) can refine data by removing detected abnormal inputs. Specifically, the conversation collection tool (131, 210) can remove unnecessary noise and abnormal utterances from voice data, and delete duplicate words, unnecessary symbols, and incomplete sentences from text information. For example, the conversation collection tool (131, 210) can organize the text "January 2024... accident sound... occurred." into "January 2024 accident occurred." Afterward, the conversation collection tool (131, 210) can perform data supplementation. Specifically, the conversation collection tool (131, 210) can generate complete data by supplementing sentences or voices damaged during the filtering process in a contextual manner. That is, the conversation collection tool (131, 210) can contextually supplement missing information in text information and naturally connect disconnected utterances in voice data. For example, the conversation collection tool (131, 210) can supplement the data “January 1, 2024 accident…” with “January 1, 2024 accident occurred.”

[0281] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform a data verification step. Specifically, the conversation collection tool (131, 210) checks whether the filtered and supplemented data is suitable for analysis and can perform additional corrections if essential information is missing or is grammatically incorrect. For example, the conversation collection tool (131, 210) can correct the data "in Seoul in 2024" to "an accident occurred in Seoul in 2024" by supplementing the missing information.

[0282] According to one embodiment of the present invention, various deep learning models may be utilized in the abnormal input filtering step (S120-2). For example, Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) can be used to learn temporal continuity to detect and correct long silences or discontinuous utterances in speech data. Additionally, Transformer-based models (BERT, GPT, etc.) can analyze context in text information and supplement incomplete sentences. Furthermore, Noise Detection and Signal Classification models can detect abnormal signals in speech data, and Anomaly Detection models (Autoencoder, GAN) can detect and remove abnormal patterns in the data. Additionally, Named Entity Recognition (NER) models can identify key entities (date, place, etc.) in text information and supplement missing information.

[0283] Text information cleanup step (S120-3)

[0284] According to one embodiment of the present invention, in the text information organization step (S120-3), the conversation collection tool (131, 210) can organize the collected text information into an analyzable state, remove unnecessary elements, and refine it into a grammatically complete form. Specifically, the conversation collection tool (131, 210) can standardize the text information. For example, the conversation collection tool (131, 210) can unify uppercase and lowercase letters to maintain data consistency, or convert specific data such as dates into a standard format. For example, the conversation collection tool (131, 210) can organize the input text "An accident occurred in Seoul on January 1, 2024!!" into "An accident occurred in Seoul on January 1, 2024". In addition, the conversation collection tool (131, 210) can improve the readability of the text by correcting duplicate spaces or incorrect spacing.

[0285] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform the task of removing meaningless data from text. Specifically, the conversation collection tool (131, 210) can remove unnecessary elements or duplicate words such as symbols like !!, ###, and ??, and can also delete unnecessary utterances such as "um," "uh," and "geu." For example, the conversation collection tool (131, 210) can organize the text "January 2024... accident um... occurred" into "January 2024 accident occurred." After that, the conversation collection tool (131, 210) can perform the task of organizing the context by supplementing incomplete sentences. Specifically, the conversation collection tool (131, 210) can correct the spacing and punctuation of the sentences and, if necessary, supplement incomplete sentences to convert them into natural sentences. For example, the conversation collection tool (131, 210) can supplement the text “January 1, 2024 accident” with “January 1, 2024 accident occurred.”

[0286] According to one embodiment of the present invention, a conversation collection tool (131, 210) can structure data and convert it into a form suitable for analysis and storage. Specifically, the conversation collection tool (131, 210) can identify key entities such as dates, places, and events from text information and convert them into a standard format such as JSON or CSV. For example, the conversation collection tool (131, 210) can store the text "An accident occurred in Seoul on January 1, 2024." as structured data such as {"Date": "2024-01-01", "Place": "Seoul", "Event": "Accident occurred"}. Additionally, the conversation collection tool (131, 210) can perform the task of verifying the organized text information. Specifically, the conversation collection tool (131, 210) can check whether essential information is missing from the text, correct grammatically incorrect sentences, and supplement incomplete data according to the context. For example, the conversation collection tool (131, 210) can supplement the missing information for the input “in Seoul in 2024” to modify it to “an accident occurred in Seoul in 2024.”

[0287] According to one embodiment of the present invention, various deep learning models may be utilized in the text information cleaning step (S120-3). Specifically, Transformer-based models (BERT, GPT, T5) can analyze context and supplement incomplete text, and Named Entity Recognition (NER) models can identify key entities and systematically structure data. Additionally, Text Normalization models can perform data standardization, and Sequence-to-Sequence models can be used to supplement and reconstruct incomplete text. Furthermore, Pre-trained Embedding models (FastText, Word2Vec, etc.) can learn semantic relationships between words.

[0288] Quality verification and data storage step (S120-4)

[0289] According to one embodiment of the present invention, in the quality verification and data storage step (S120-4), the conversation collection tool (131, 210) can finally verify the quality of the organized text and voice data and safely store them. Specifically, the conversation collection tool (131, 210) can verify the data quality and, in this process, review the integrity and consistency of the text information and voice data. For example, the conversation collection tool (131, 210) can verify whether the text information is grammatically correct and review whether there is any missing essential information (e.g., date, place, event). In addition, the conversation collection tool (131, 210) can verify whether the data maintains a standardized format (date format, number notation, etc.). For example, the conversation collection tool (131, 210) can determine that the data "in Seoul in 2024" is missing essential event information and modify it to "An accident occurred in Seoul on January 1, 2024." In addition, for voice data, the conversation collection tool (131, 210) can evaluate whether noise removal and restoration work has been properly performed and check whether the voice signal is clearly heard.

[0290] According to one embodiment of the present invention, a conversation collection tool (131, 210) can review data validity to determine whether it is in a state where it can be utilized in a subsequent step. Specifically, text information can verify whether key entities (date, place, event) are correctly identified and stored, and voice data can evaluate the clarity of speech. In addition, data that has undergone quality verification can be stored in an environment equipped with a security and management system. For example, data can be encrypted using an encryption algorithm such as AES-256 and can be classified and stored according to the type of litigation (e.g., traffic accident, contract dispute).

[0291] According to one embodiment of the present invention, the conversation collection tool (131, 210) can perform a version control function to track changes by recording previous versions when data is updated. In addition, the conversation collection tool (131, 210) can perform regular backups to prepare for data loss and prepare a recovery system. Furthermore, the conversation collection tool (131, 210) can control access to sensitive data by setting user-specific access rights and, if necessary, de-identify the data.

[0292] According to one embodiment of the present invention, various deep learning models may be utilized in the quality verification and data storage step (S120-4). Specifically, Transformer-based models (BERT, GPT, etc.) verify quality by analyzing the context and semantics of text information, and Named Entity Recognition (NER) models can identify key information and supplement missing information. In addition, Anomaly Detection models (Autoencoder, GAN) detect abnormal patterns in text and voice data, and Speech Quality Assessment models evaluate the clarity and quality of voice data. Furthermore, data classification and encryption models may be utilized to automatically classify data and protect sensitive information.

[0293] Real-time transmission step (S130)

[0294] According to one embodiment of the present invention, in the real-time transmission step (S130), the conversation collection tool (131, 210) can transmit data to transmit collected voice and text information to a natural language processing tool (132, 220) without delay. Specifically, the conversation collection tool (131, 210) can perform data packaging operations and convert the collected voice or text information into a transmittable standard format such as JSON, XML, or Protobuf. Additionally, the conversation collection tool (131, 210) can structure the data to minimize the transmission size and add metadata such as timestamps or litigation types to preserve the context of the data. For example, the conversation collection tool (131, 210) can transmit the text "An accident occurred in Seoul on January 1, 2024." in a structured format such as {"Date": "2024-01-01", "Place": "Seoul", "Event": "Accident occurred", "Timestamp": "2024-01-01T14:00:00Z"}.

[0295] According to one embodiment of the present invention, a conversation collection tool (131, 210) can perform data transmission optimization. Specifically, the conversation collection tool (131, 210) can reduce the transmission size by compressing data (Gzip, Zstandard, etc.) to efficiently utilize network bandwidth, and can use efficient protocols such as HTTP / 2, WebSocket, or gRPC. In addition, when processing large amounts of data, the conversation collection tool (131, 210) can improve the transmission speed by dividing the data into small chunks and transmitting them in parallel. For example, when 100MB of data is input, the conversation collection tool (131, 210) can divide it into 10MB chunks and perform parallel transmission.

[0296] According to one embodiment of the present invention, the conversation collection tool (131, 210) can monitor the transmission status in real time. In this case, the conversation collection tool (131, 210) can check for packet loss and, if a problem occurs, automatically request retransmission. That is, the conversation collection tool (131, 210) can check the network status by measuring the transmission speed and latency in real time and verify whether the data has arrived accurately without loss. In addition, the conversation collection tool (131, 210) can verify whether the transmitted data has been successfully processed at the receiving end. Specifically, when the data arrives at the destination, the conversation collection tool (131, 210) can send an ACK (Acknowledgment) signal to the sending end to notify the reception status. In addition, if the ACK is not received, the conversation collection tool (131, 210) can automatically retransmit the data to prevent transmission failure.

[0297] According to one embodiment of the present invention, various deep learning models may be utilized in the real-time transmission step (S130). Specifically, an Anomaly Detection model (Autoencoder, GAN) can detect abnormal conditions (delay, loss) that may occur during network transmission, and a Compression Optimization model (Variational Autoencoder) can efficiently compress data. In addition, a Network Traffic Prediction model (RNN, LSTM) can predict network conditions and dynamically adjust the transmission strategy. Furthermore, a Transformer-based model (BERT, T5, etc.) can analyze the structure of transmitted data and perform automatic packaging, and a Data Integrity Verification model can verify the integrity of the data in real time.

[0298] FIG. 3 is a diagram illustrating the natural language processing and data extraction steps of a natural language processing and data extraction tool according to one embodiment of the present invention.

[0299] Data preprocessing step (S200)

[0300] Text cleanup step (S200-1)

[0301] According to one embodiment of the present invention, in the text organization step (S200-1), the natural language processing and data extraction tool (132, 220) can organize the collected text information into a form suitable for analysis and processing. Specifically, the natural language processing and data extraction tool (132, 220) can standardize the collected text information. For example, the natural language processing and data extraction tool (132, 220) can convert items such as dates, times, and amounts within the text into a consistent format and unify uppercase and lowercase letters. For example, the natural language processing and data extraction tool (132, 220) can organize text such as "Accident occurred in Seoul on January 1, 2024!!" into "Accident occurred in Seoul on January 1, 2024". In addition, the natural language processing and data extraction tool (132, 220) can make the data neat and concise by correcting duplicate spaces and incorrect spacing.

[0302] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform the task of removing unnecessary data. Specifically, the natural language processing and data extraction tool (132, 220) can delete unnecessary elements or duplicate words from text that interfere with analysis, such as symbols (e.g., !!, ###, ???). In addition, the natural language processing and data extraction tool (132, 220) can improve the clarity and conciseness of the data by removing meaningless words and incomplete sentences such as "um," "uh," and "geu." For example, the natural language processing and data extraction tool (132, 220) can organize the text "January 2024... accident um... occurred" into "January 2024 accident occurred." After that, the natural language processing and data extraction tool (132, 220) can supplement the text information into a grammatically complete sentence.

[0303] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can convert incomplete or interrupted sentences into natural sentences by taking context into account. Specifically, the natural language processing and data extraction tool (132, 220) can adjust spacing and punctuation. For example, the natural language processing and data extraction tool (132, 220) can complete the text "January 1, 2024 accident" into "January 1, 2024 accident occurred." Additionally, the natural language processing and data extraction tool (132, 220) can convert text information into a structured format such as JSON or XML. For example, a natural language processing and data extraction tool (132, 220) can convert the text “An accident occurred in Seoul on January 1, 2024.” into structured data such as {“Date”: “2024-01-01”, “Place”: “Seoul”, “Event”: “Accident occurred”}.

[0304] According to one embodiment of the present invention, various deep learning models may be utilized in the text cleaning step (S200-1). Specifically, Transformer-based models (BERT, GPT, T5, etc.) can understand context, clean up text, and generate grammatically complete sentences. Additionally, Named Entity Recognition (NER) models can be used to clean up data by identifying key information such as dates, places, and events. Furthermore, Text Normalization models perform case conversion and date format standardization, and Sequence-to-Sequence models can generate natural sentences by reconstructing incomplete sentences. Additionally, Anomaly Detection models can detect and remove abnormal input data.

[0305] Language standardization stage (S200-2)

[0306] According to one embodiment of the present invention, in the language standardization step (S200-2), the natural language processing and data extraction tool (132, 220) can convert collected text information into a standardized format. Specifically, the natural language processing and data extraction tool (132, 220) can unify the case of the text information. For example, in the case of a language such as English, the natural language processing and data extraction tool (132, 220) can convert all characters into lowercase or uppercase. In addition, the natural language processing and data extraction tool (132, 220) can handle proper nouns where case is important exceptionally so that their meaning is preserved. For example, the natural language processing and data extraction tool (132, 220) can convert the input "The ACCIDENT happened in Seoul." into "the accident happened in seoul."

[0307] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can standardize date and time formats. Specifically, the natural language processing and data extraction tool (132, 220) can convert date data of various expression modes into an ISO standard (e.g., YYYY-MM-DD) to increase the accuracy of analysis, and can unify time into a 24-hour format (e.g., 14:00). For example, the natural language processing and data extraction tool (132, 220) can convert an input such as "January 1, 2024" into "2024-01-01". Additionally, the natural language processing and data extraction tool (132, 220) can convert numeric formats included in text. Furthermore, the natural language processing and data extraction tool (132, 220) converts numbers expressed as text into numeric data, and if a unit is included, such as "5,000 won", it can convert it into a standardized numeric format. For example, a natural language processing and data extraction tool (132, 220) can convert an input of “five million won” into “5,000,000 won”.

[0308] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform the task of standardizing domain-specific terms. Specifically, the natural language processing and data extraction tool (132, 220) can maintain consistency between data by converting various expressions with the same meaning into a single unified term. For example, the natural language processing and data extraction tool (132, 220) can integrate the expressions "traffic accident" and "traffic disaster" into "traffic accident." In addition, the natural language processing and data extraction tool (132, 220) can remove unnecessary symbols and sentences from text. Furthermore, the natural language processing and data extraction tool (132, 220) can delete unnecessary special characters such as "!!!" and "???" or duplicate spaces, and leave only contextually appropriate symbols. For example, a natural language processing and data extraction tool (132, 220) can organize the text “Accident in Seoul!!!” into “Accident in Seoul”.

[0309] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can convert abbreviations or shortened expressions into their original forms. Specifically, the natural language processing and data extraction tool (132, 220) can convert abbreviations into predefined standard extensions and convert shortened expressions into their original complete sentences. For example, the natural language processing and data extraction tool (132, 220) can convert the input "January 1, 2024, the accident occurred as expected." into "January 1, 2024, the accident occurred as expected."

[0310] According to one embodiment of the present invention, various deep learning models can be utilized in the language standardization step (S200-2). Specifically, Transformer-based models (BERT, GPT, etc.) can understand context and convert text into a consistent expression. Additionally, Sequence-to-Sequence models (T5, etc.) are suitable for reconstructing text such as dates, times, and numbers into a standardized format, and Pre-trained Embedding models (FastText, Word2Vec, etc.) can learn relationships between words and convert synonyms into a standardized expression. Furthermore, Named Entity Recognition (NER) models can identify and standardize key information such as dates, places, and events, and Text Normalization models can remove unnecessary symbols within the text and maintain a standard format.

[0311] Additional work step (S200-3)

[0312] According to one embodiment of the present invention, in an additional work step (S200-3), the natural language processing and data extraction tool (132, 220) can refine text information more precisely and generate or reinforce additional information for a specific purpose. Specifically, if the text information contains contextually incomplete or missing information, the natural language processing and data extraction tool (132, 220) can analyze the context of the sentence to complete the incomplete sentence or infer and add the missing information. In addition, the natural language processing and data extraction tool (132, 220) can connect data by referring to the context, such as previous conversation content. For example, the natural language processing and data extraction tool (132, 220) can supplement the text "The accident occurred in Seoul." into a complete sentence such as "An accident occurred in Seoul on January 1, 2024."

[0313] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can identify and structure additional information required in a specific domain, such as law or medicine, through domain-specific information extraction. For example, the natural language processing and data extraction tool (132, 220) can extract whether there is a breach of contract and the amount of damage from the text "100 million won in damages caused by breach of contract." and perform conversion into structured data such as {"Breach of contract": True, "Amount of damage": 100,000,000 won}. In addition, the natural language processing and data extraction tool (132, 220) can convert simple text information into data in a hierarchical structure or network format that can be analyzed and utilized. For example, a natural language processing and data extraction tool (132, 220) can convert the text “January 1, 2024, A rear-ended B and caused an accident.” into structured data such as {“Event”: “Rear-end collision accident”, “Date”: “2024-01-01”, “Perpetrator”: “A”, “Victim”: “B”}.

[0314] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can define relationships between entities within a text by performing relationship analysis between data. Specifically, the natural language processing and data extraction tool (132, 220) extracts relationships between subjects and objects in a sentence and can separate and define relationships when they are complexly intertwined. For example, the natural language processing and data extraction tool (132, 220) can convert the text "Kim Cheol-su compensated the disaster victims on January 1, 2024" into data such as {"Subject": "Kim Cheol-su", "Action": "Compensation for damages", "Object": "Disaster victims", "Date": "2024-01-01"}. Additionally, the natural language processing and data extraction tool (132, 220) can add metadata to text information through data enhancement. For example, a natural language processing and data extraction tool (132, 220) can generate text "An accident occurred in Seoul." into augmented data such as {"content": "An accident occurred in Seoul", "timestamp": "2024-01-01T10:00:00Z"}.

[0315] According to one embodiment of the present invention, various deep learning models may be utilized in the additional work step (S200-3). Specifically, Transformer-based models (BERT, GPT, T5, etc.) can understand context and supplement incomplete text, and Sequence-to-Sequence models (T5, Bart) can convert text into structured data. In addition, Named Entity Recognition (NER) models can extract domain-specific information and key entities, and Relation Extraction models (OpenNRE, BERT-based) can analyze relationships between entities. Furthermore, Contextual Embedding models (FastText, Word2Vec) can be used to understand domain-specific expressions by learning semantic relationships between words.

[0316] Quality verification and storage step (S200-4)

[0317] According to one embodiment of the present invention, in the quality verification and storage step (S200-4), the natural language processing and data extraction tool (132, 220) can verify and store the quality of the finally organized text information. Specifically, the natural language processing and data extraction tool (132, 220) can perform data quality verification work. That is, the natural language processing and data extraction tool (132, 220) can verify whether the collected data is grammatically correct and can perform a check to ensure that essential information (e.g., date, place, event) is not missing. In addition, the natural language processing and data extraction tool (132, 220) can verify whether the data follows a standardized format (e.g., ISO date format, unified number representation). That is, the natural language processing and data extraction tool (132, 220) can supplement semantically incomplete sentences by analyzing the context. For example, a natural language processing and data extraction tool (132, 220) can supplement the missing event information in the text “in Seoul in January 2024” to correct it to “an accident occurred in Seoul on January 1, 2024.”

[0318] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform data validity checks. That is, the natural language processing and data extraction tool (132, 220) can remove duplicate items so that duplicate data is not stored multiple times, and can detect and process abnormal inputs such as meaningless text and error data. In addition, the natural language processing and data extraction tool (132, 220) can check the correlation between data contextually. For example, the natural language processing and data extraction tool (132, 220) can combine duplicate sentences such as "The accident occurred in Seoul." and "An accident occurred in Seoul." into one.

[0319] According to one embodiment of the present invention, various deep learning models may be utilized in the quality verification and storage step (S200-4). Specifically, Transformer-based models (BERT, GPT, etc.) verify the quality of text information by analyzing context and semantics, and Named Entity Recognition (NER) models can identify essential information in text and detect missing information. Additionally, Anomaly Detection models (Autoencoder, GAN) refine data by detecting abnormal data patterns, and Data Validation models (Variational Autoencoder, VAE) can check the structure and consistency of the data. Furthermore, security models may be utilized for data encryption and integrity verification.

[0320] Sentence analysis step (S210)

[0321] Sentence Tokenization Step (S210-1)

[0322] According to one embodiment of the present invention, in the sentence tokenization step (S210-1), the natural language processing and data extraction tool (132, 220) can perform the task of separating text information into sentence units. Specifically, the natural language processing and data extraction tool (132, 220) can perform the task of detecting sentence boundaries. That is, the natural language processing and data extraction tool (132, 220) can separate sentences by detecting punctuation marks indicating the end of a sentence, such as periods, question marks, and exclamation marks, within the text. In this case, the natural language processing and data extraction tool (132, 220) can identify the end of a sentence based on punctuation marks, divide the text, and determine sentence boundaries by considering the context. For example, a natural language processing and data extraction tool (132, 220) can separate the text "January 1, 2024. An accident occurred in Seoul. The victim is Kim Cheol-su." into sentences such as ["January 1, 2024.", "An accident occurred in Seoul.", "The victim is Kim Cheol-su."].

[0323] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform sentence boundary ambiguity resolution. Specifically, the natural language processing and data extraction tool (132, 220) can perform processing so that a period is not incorrectly recognized as the end of a sentence even if it is used in abbreviations or numeric formats. For example, the natural language processing and data extraction tool (132, 220) can correctly separate the sentence "He holds a Ph.D. degree" by setting the period after "Ph.D." so that it is not recognized as the end of the sentence. As a result, the natural language processing and data extraction tool (132, 220) can separate "Dr. Kim Cheol-su arrived at 10:30 AM. He holds a Ph.D. degree." into ["Dr. Kim Cheol-su arrived at 10:30 AM.", "He holds a Ph.D. degree."].

[0324] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform separation of long sentences that are continuous without punctuation based on contextual meaning or conjunctions through the separation of long sentences. For example, the natural language processing and data extraction tool (132, 220) can perform processing by dividing "An accident occurred in Seoul and the victim was transported to the hospital." into ["An accident occurred in Seoul.", "And the victim was transported to the hospital."]. In addition, the natural language processing and data extraction tool (132, 220) can perform context-based sentence boundary recognition. For example, the natural language processing and data extraction tool (132, 220) can separate contextually continuous text into semantic units and can perform separation of the input "An accident occurred on January 1, 2024 in Seoul and the victim is Kim Cheol-su." into ["An accident occurred on January 1, 2024.", "The victim in Seoul is Kim Cheol-su."].

[0325] According to one embodiment of the present invention, various deep learning models may be utilized in the sentence tokenization step (S210-1). Specifically, Transformer-based models (BERT, GPT, T5, etc.) can understand context and learn sentence boundaries to separate sentences, and Sequence Tagging models (BiLSTM-CRF) can tag sentence boundaries based on relationships between words. In addition, Rule-Based Hybrid models can process simple sentence boundaries by combining regular expressions and rule-based approaches with deep learning models, and Contextual Embedding models (FastText, Word2Vec) can detect the boundaries of long or ambiguous sentences by learning contextual meaning. Furthermore, Attention-based models can recognize boundaries by focusing on important words and punctuation marks at sentence boundaries.

[0326] Part-of-speech tagging step (S210-2)

[0327] According to one embodiment of the present invention, in the part-of-speech tagging step (S210-2), the natural language processing and data extraction tool (132, 220) can perform the task of identifying the part of speech of each word in text information and assigning tags. Specifically, the natural language processing and data extraction tool (132, 220) can separate text information into word units and tag the part of speech of each word. Specifically, the natural language processing and data extraction tool (132, 220) determines which part of speech a word belongs to, such as a noun, verb, adjective, adverb, or preposition, and can predict the appropriate part of speech by analyzing the form of the word and its position within the sentence. For example, a natural language processing and data extraction tool (132, 220) can tag parts of speech for the sentence “Kim Cheol-su witnessed an accident” as [("Kim Cheol-su", "NNP"), ("ga", "JKS"), ("sago", "NNG"), ("reul", "JKO"), ("witnessed", "VV"), ("did", "EP+EF")].

[0328] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can select an appropriate part of speech based on context for words that can be used as multiple parts of speech, such as polysemous words. Specifically, it determines the correct part of speech by analyzing the role and meaning of a word in a sentence and can improve the accuracy of tagging by utilizing the relationship with surrounding words. For example, the natural language processing and data extraction tool (132, 220) tags "bright" as an adjective (VA+ETM) in the sentence "The bright light was dazzling," but can tag "bright" as a verb (VV+ETM) in the sentence "The problem was solved in a bright state." Additionally, the natural language processing and data extraction tool (132, 220) can assign an appropriate part of speech by comparing it with training data in the case of special words such as compound words, abbreviations, and proper nouns. For example, a natural language processing and data extraction tool (132, 220) tags "Seoul Metropolitan City" as a proper noun (NNP) in the sentence "An accident occurred in Seoul Metropolitan City," and can separate compound words as needed or process them as a single unit.

[0329] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can improve the accuracy of part-of-speech tagging by utilizing contextual information. That is, the natural language processing and data extraction tool (132, 220) determines the part of speech by referring to the part of speech of the previous or next word, and can perform accurate tagging according to context by analyzing the overall structure of the sentence. For example, the natural language processing and data extraction tool (132, 220) can tag "quickly" as an adverb (MAG) in the sentence "He ran quickly." Additionally, the natural language processing and data extraction tool (132, 220) can identify predefined domain-specific terms and perform customized part-of-speech tagging. For example, the natural language processing and data extraction tool (132, 220) can tag "defendant" as a proper noun (NNP) in the sentence "The defendant claimed breach of contract."

[0330] According to one embodiment of the present invention, various deep learning models may be utilized in the part-of-speech tagging step (S210-2). Specifically, a BiLSTM-CRF model can perform part-of-speech tagging based on relationships between words by learning contextual information bidirectionally. Additionally, a Transformer-based model (BERT, RoBERTa, etc.) can be utilized for polysemous word processing and context-based tagging by deeply learning the context of a sentence. Furthermore, a word vector model such as FastText analyzes lexical characteristics to predict parts of speech based on similarity between words, and a Pre-trained Embedding model (Word2Vec, GloVe, etc.) can process compound words and abbreviations by learning the semantic relationships between words.

[0331] Parsing step (S210-3)

[0332] According to one embodiment of the present invention, in the parsing step (S210-3), the natural language processing and data extraction tool (132, 220) can analyze the relationships between words in a sentence to identify the structure of the sentence and represent it in a tree format. Specifically, the natural language processing and data extraction tool (132, 220) can perform dependency parsing. That is, the natural language processing and data extraction tool (132, 220) can analyze the grammatical dependency relationships between words in a sentence to identify the relationships between the subject and the predicate, and between the object and the modifier. Accordingly, the natural language processing and data extraction tool (132, 220) can represent the structure of the sentence as a dependency tree. For example, the natural language processing and data extraction tool (132, 220) can represent the sentence "Kim Cheol-su witnessed the accident." with dependency relationships such as "witnessed -> Kim Cheol-su (subject)" and "witnessed -> the accident (object)".

[0333] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can generate a tree by dividing phrases and clauses of a sentence according to grammatical rules and can express the hierarchical structure of the sentence. For example, the natural language processing and data extraction tool (132, 220) can convert the sentence "An accident occurred in Seoul" into a tree structure such as "VP -> PP (in Seoul)" and "NP (an accident occurred)". In addition, the natural language processing and data extraction tool (132, 220) can identify semantic relationships between words based on syntactic structure. Specifically, the natural language processing and data extraction tool (132, 220) can identify the agent (subject) and object (object) of an action related to a verb and understand the meaning of the sentence. For example, a natural language processing and data extraction tool (132, 220) can define the semantic relationship of an action by connecting the verb "helped" in the sentence "the lawyer helped the client" with the subject "lawyer" and the object "client".

[0334] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can identify and separate major phrases (noun phrases, verb phrases, prepositional phrases) and clauses (noun clauses, adverbial clauses, etc.) in a sentence during the phrase and clause separation task. For example, the natural language processing and data extraction tool (132, 220) can separate the sentence "The accident that occurred in Seoul was serious." into "NP (the accident that occurred in Seoul)" and "VP (was serious)." In addition, the natural language processing and data extraction tool (132, 220) can analyze domain terms and syntactic structures by learning syntactic patterns frequently used within a domain. For example, the natural language processing and data extraction tool (132, 220) can analyze the sentence "Claim for damages due to breach of contract." into "Action: Claim," "Reason: Breach of contract," and "Subject: Damages."

[0335] According to one embodiment of the present invention, various deep learning models may be utilized in the parsing step (S210-3). Specifically, a BiLSTM-CRF model can generate a dependency tree by learning grammatical dependency relationships between words. Additionally, a Transformer-based model (BERT, T5, etc.) can analyze complex sentence structures by learning contextual information and perform polysemous word processing and semantic relationship analysis. Furthermore, Graph Neural Networks (GNNs) can represent dependency trees in the form of graphs and learn complex relationships between words. Additionally, a Seq2Seq model can analyze input text and generate a hierarchical output (tree structure). Furthermore, a Context-Free Grammar (CFG)-based model can generate a syntax tree based on rules.

[0336] Key information extraction step (S220)

[0337] NER Step (S220-1)

[0338] According to one embodiment of the present invention, in the NER step (S220-1), that is, the Named Entity Recognition step, a natural language processing and data extraction tool (132, 220) can identify specific types of information, such as proper nouns, dates, numbers, places, organization names, and event names, in text information and perform tagging. Specifically, the natural language processing and data extraction tool (132, 220) can perform a named entity identification task. That is, the natural language processing and data extraction tool (132, 220) can detect specific information, such as dates, times, places, people, and events, within the text, and identify the boundaries of each entity to assign tags. For example, a natural language processing and data extraction tool (132, 220) can identify and tag entities in the form of [("Kim Cheol-su", "PERSON"), ("January 1, 2024", "DATE"), ("Seoul", "LOCATION"), ("Accident", "EVENT")] for the sentence "Kim Cheol-su witnessed an accident in Seoul on January 1, 2024."

[0339] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform entity type classification to classify which type each entity belongs to, such as a person, place, or event. For example, the natural language processing and data extraction tool (132, 220) can tag the sentence "The defendant claimed 100 million won in damages due to breach of contract" as [("Defendant", "LEGAL_ROLE"), ("Contract", "CONTRACT"), ("Violation", "VIOLATION"), ("100 million won", "MONEY")]. Additionally, the natural language processing and data extraction tool (132, 220) can perform a task to improve the precision of entities based on context. For example, a natural language processing and data extraction tool (132, 220) can tag "accident" as "EVENT" in the sentence "An accident occurred in Seoul." but tag "accident occurred" as "EVENT" and "press conference" as a separate "EVENT" in the sentence "A press conference was held in Seoul after the accident occurred."

[0340] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can integrate and concisely express the same entity when it appears repeatedly or is expressed consecutively in text. For example, the natural language processing and data extraction tool (132, 220) can generate a result in which duplicate entities are removed from the sentence "Kim Cheol-su witnessed an accident in Seoul, in Seoul Metropolitan City." as [("Kim Cheol-su", "PERSON"), ("Seoul", "LOCATION"), ("Accident", "EVENT")]. Additionally, the natural language processing and data extraction tool (132, 220) can learn terms and patterns frequently used in a domain to identify them as entities and assign appropriate tags. For example, a natural language processing and data extraction tool (132, 220) can tag the sentence “The plaintiff filed a breach of contract lawsuit on March 10” as [("Plaintiff", "LEGAL_ROLE"), ("March 10", "DATE"), ("Breach of contract", "VIOLATION"), ("Lawsuit", "LEGAL_ACTION")].

[0341] According to one embodiment of the present invention, various deep learning models may be utilized in the NER step (S220-1). Specifically, Transformer-based models (BERT, RoBERTa, XLM-RoBERTa, etc.) can learn context to perform entity identification within text. Additionally, BiLSTM-CRF models can learn relationships between words and reflect dependencies between tags to distinguish entity boundaries and types. Furthermore, Sequence-to-Sequence models (T5, GPT, etc.) can analyze text and perform entity tagging tasks in the form of sequence output.

[0342] Relationship extraction step (S220-2)

[0343] According to one embodiment of the present invention, in the relationship extraction step (S220-2), the natural language processing and data extraction tool (132, 220) can analyze the relationships between entities identified in the text and define the logical or semantic connections between them. Specifically, the natural language processing and data extraction tool (132, 220) can perform the task of identifying relationships between entities and can analyze the interactions between entities detected in the sentence. For example, in the sentence "Kim Cheol-su witnessed an accident in Seoul," the natural language processing and data extraction tool (132, 220) defines the relationships as "Kim Cheol-su" as the "subject," "accident" as the "object," and "Seoul" as the "place," and the connection between them can be expressed by the verb "witness."

[0344] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can classify detected relationships into predefined categories through a relationship type classification task. Specifically, relationships can be classified into general types such as "cause-effect," "affiliation," and "location," and specialized relationship types such as "plaintiff-defendant" and "breach of contract" can be additionally defined in specific domains such as law and medicine. For example, the natural language processing and data extraction tool (132, 220) can classify the relationship between "defendant" and "breach of contract" as "lawsuit" for the sentence "The defendant was sued for breach of contract."

[0345] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can analyze the context to infer indirect relationships or define ambiguous relationships when a relationship is not explicitly mentioned in a sentence. For example, the natural language processing and data extraction tool (132, 220) can define the relationship between "Seoul" and "accident" as "occurrence" in the sentence "The accident that occurred in Seoul was serious." Additionally, when a single sentence contains multiple entities and relationships, the natural language processing and data extraction tool (132, 220) can perform multiple relationship extraction. For example, the natural language processing and data extraction tool (132, 220) can define two relationships, "witnessing" and "reporting," in the sentence "Kim Cheol-su witnessed the accident in Seoul on January 1, 2024, and reported it to the police."

[0346] According to one embodiment of the present invention, various deep learning models may be utilized in the relationship extraction step (S220-2). Specifically, Transformer-based models (BERT, T5, etc.) can detect complex relationships in sentences by learning contextual information. Additionally, a Relation Classification model (BiLSTM-CRF) can learn relationships between entities, define relationships through a tagging method, and provide high accuracy. Furthermore, Graph Neural Networks (GNNs) are suitable for processing complex multiple relationships by representing entities and relationships in a graph structure, and a Seq2Seq model can process relationship extraction in the form of sequence output.

[0347] Keyword extraction step (S220-3)

[0348] According to one embodiment of the present invention, in the keyword extraction step (S220-3), the natural language processing and data extraction tool (132, 220) can identify important words or phrases in text information to summarize the content of the document or identify the main topic. Specifically, the natural language processing and data extraction tool (132, 220) can perform preprocessing work before analyzing the text information. That is, the natural language processing and data extraction tool (132, 220) can remove unnecessary punctuation and special characters, standardize the case of the text, and remove stop words such as "and" and "but". For example, the natural language processing and data extraction tool (132, 220) can organize the sentence "An accident occurred in Seoul on January 1, 2024." into ["2024", "January", "Seoul", "Accident", "Occurrence"].

[0349] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can generate keyword candidates from text. That is, the natural language processing and data extraction tool (132, 220) extracts meaningful words such as nouns, verbs, and adjectives from a sentence, and can analyze word combinations by generating N-grams such as 2-grams and 3-grams as needed. For example, the natural language processing and data extraction tool (132, 220) can generate ["Seoul", "traffic accident", "damage", "occurrence"] as keyword candidates from the sentence "The traffic accident that occurred in Seoul left serious damage." In addition, the natural language processing and data extraction tool (132, 220) can calculate the importance of keywords to select important words among the generated keyword candidates. Here, natural language processing and data extraction tools (132, 220) can extract words that appear frequently in text but rarely in other documents using the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm. Additionally, the RAKE (Rapid Automatic Keyword Extraction) algorithm selects keywords based on the coexistence relationship between words, and TextRank can rank words by analyzing the connectivity between words using a graph-based algorithm.

[0350] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can perform domain-specific keyword filtering in specific domains (law, medical, etc.). That is, in the case of legal documents, the natural language processing and data extraction tool (132, 220) can learn specialized terms such as "breach of contract," "damages," and "litigation" in advance and select these terms as keywords.

[0351] According to one embodiment of the present invention, various deep learning models may be utilized in the keyword extraction step (S220-3). Specifically, Transformer-based models (BERT, GPT, T5, etc.) can identify important words within a text by analyzing the context and select keywords by reflecting contextual importance. Additionally, Attention-based models can calculate the importance of keywords by focusing on specific words and phrases. Furthermore, TextRank analyzes connectivity between words using a graph-based algorithm, and word embedding models such as FastText and Word2Vec can group related keywords by analyzing similarity between words.

[0352] Information supplementation stage (S230)

[0353] Missing Information Verification Step (S230-1)

[0354] According to one embodiment of the present invention, in the missing information verification step (S230-1), the natural language processing and data extraction tool (132, 220) can check whether essential information is missing from the text information. Specifically, the natural language processing and data extraction tool (132, 220) can define information that must be included in the document or conversation to be analyzed by domain. For example, in the case of the legal domain, the natural language processing and data extraction tool (132, 220) can set the date, place, and related person of an event as essential information. Subsequently, the natural language processing and data extraction tool (132, 220) can check if there are any missing items by comparing with the information extracted from the text. For example, the natural language processing and data extraction tool (132, 220) can identify that "place" and "time" information are missing in the sentence "Kim Cheol-su witnessed the accident."

[0355] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can analyze the context to infer or verify that specific information is implicitly omitted or not explicitly mentioned in a sentence. For example, the natural language processing and data extraction tool (132, 220) can identify that essential information such as "date," "perpetrator," and "victim" is missing in the sentence "The accident occurred in Seoul." Additionally, the natural language processing and data extraction tool (132, 220) can analyze information frequently required in a specific domain to determine whether it is missing. For example, the natural language processing and data extraction tool (132, 220) can identify that "specific details of the breach of contract" and "evidence" are missing in the sentence "The defendant lost to the plaintiff in the breach of contract lawsuit."

[0356] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can classify missing information into time, place, event details, person, etc. through a missing information type classification task. For example, the natural language processing and data extraction tool (132, 220) can classify the sentence "The contract was concluded but not performed" as having missing information such as "date of contract conclusion," "place of contract performance," and "reason for non-performance." Additionally, the natural language processing and data extraction tool (132, 220) can comprehensively identify missing information by comparing data provided across multiple sentences through information linkage between conversations and documents. That is, the natural language processing and data extraction tool (132, 220) checks the data completeness of a new sentence by utilizing data provided in previous conversations or sentences, and can supplement missing items based on the connectivity between related data. For example, a natural language processing and data extraction tool (132, 220) can identify the “location of the accident” as missing information when given the sentence “an accident occurred in January 2024” and the sentence “the victim of the accident is Kim Cheol-su”.

[0357] According to one embodiment of the present invention, various deep learning models may be utilized in the missing information verification step (S230-1). Specifically, Transformer-based models (BERT, GPT, T5, etc.) can analyze context to identify missing parts among essential items within the text. Additionally, Question-Answering models (BERT-QA, RoBERTa-QA) can be utilized to generate questions regarding essential information and to identify cases where answers are missing. Furthermore, Sequence Classification models can evaluate the completeness of the text and classify missing information by type. Additionally, Relation Extraction models identify necessary information based on relationships between entities, and Graph Neural Networks (GNNs) can detect missing information by learning relationships between multiple sentences.

[0358] Additional question generation step (S230-2)

[0359] According to one embodiment of the present invention, in the additional question generation step (S230-2), the natural language processing and data extraction tool (132, 220) can automatically generate appropriate questions to supplement information missing from text information. Specifically, the natural language processing and data extraction tool (132, 220) can analyze the context related to the missing information and set the topic of the question. For example, if the missing item is "accident date," the natural language processing and data extraction tool (132, 220) can generate a question such as "When did the accident occur?". In addition, the natural language processing and data extraction tool (132, 220) can perform a question type classification task. Specifically, questions are classified into open questions and closed questions, open questions allow the user to answer freely, and closed questions can elicit specific answers. For example, natural language processing and data extraction tools (132, 220) can classify questions such as “Please describe the situation at the time of the accident” as open questions and questions such as “Was the weather clear or cloudy at the time of the accident?” as closed questions.

[0360] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can adjust the expression method and difficulty of a question through contextual question expression adjustment work. Specifically, the natural language processing and data extraction tool (132, 220) analyzes the user's understanding and emotional state to adjust the question so that it is conveyed appropriately, and can convert complex questions into simple expressions. For example, if a user says, "The situation after the accident is too difficult," the natural language processing and data extraction tool (132, 220) can express the question gently as, "Could you explain the situation at the time of the accident a little more within a comfortable range?"

[0361] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can collect information step by step through a multi-stage question generation task in complex situations. Specifically, the natural language processing and data extraction tool (132, 220) can generate a main question first and then generate additional detailed questions based on the answer to gradually accumulate information. For example, if the missing information is "accident circumstances," the natural language processing and data extraction tool (132, 220) can generate the first question "How was the vehicle moving at the time of the accident?", then generate the second question "How far was it from the perpetrator's vehicle?", and then generate the third question "What damage occurred to the victim's vehicle?", thereby developing the questions step by step.

[0362] According to one embodiment of the present invention, various deep learning models may be utilized in the additional question generation step (S230-2). Specifically, Sequence-to-Sequence models (T5, GPT, etc.) can generate natural and appropriate questions based on input data. Additionally, Transformer-based models (BERT, RoBERTa) can draft questions by analyzing context and may be suitable for both open and closed questions. Furthermore, Attention-based models can adjust the topic of the question by highlighting important words or phrases in the context. Additionally, Pre-trained Language Models (GPT-3, ChatGPT) can generate natural questions suitable for the conversation flow and reflect emotional states or context. Furthermore, Reinforcement Learning-based models can continuously improve the quality of questions through user feedback and optimize question difficulty and expression methods.

[0363] Data structuring step (S240)

[0364] According to one embodiment of the present invention, in the data structuring step (S240), the natural language processing and data extraction tool (132, 220) can convert information extracted from text information into a systematic and structured format. Specifically, the natural language processing and data extraction tool (132, 220) can perform entity alignment and standardization operations. That is, the natural language processing and data extraction tool (132, 220) can classify extracted entities such as people, dates, and places, and organize them into a consistent format. Accordingly, duplicate data can be removed and identical entities can be integrated. For example, if date-related data is expressed as "January 1, 2024," "2024.01.01," or "January 1," the natural language processing and data extraction tool (132, 220) can convert it into a unified format of "2024-01-01."

[0365] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can identify the relationships between elements such as subject, object, place, and time in a sentence and express them in a database or graph format. For example, the natural language processing and data extraction tool (132, 220) can convert the sentence “Kim Cheol-su witnessed an accident in Seoul on January 1, 2024” into structured data such as {“Subject”: “Kim Cheol-su”, “Action”: “Witness”, “Object”: “Accident”, “Place”: “Seoul”, “Date”: “2024-01-01”}.

[0366] According to one embodiment of the present invention, a natural language processing and data extraction tool (132, 220) can define the relationship between upper items and lower items to perform data storage in a tree structure or a hierarchical database. For example, the natural language processing and data extraction tool (132, 220) can represent data such as "traffic accident that occurred in Seoul" in a hierarchical structure such as {"Event": {"Type": "Traffic Accident", "Location": "Seoul", "Details": {"Date": "2024-01-01", "Severity": "Severe"}}}.

[0367] According to one embodiment of the present invention, various deep learning models may be utilized in the data structuring step (S240). Specifically, Transformer-based models (BERT, T5) are useful for analyzing context and converting data into a structured format, and Graph Neural Networks (GNNs) can represent relationships between data in the form of a graph. Additionally, Sequence-to-Sequence models (GPT, T5) can perform conversion of text information into structured data formats such as JSON and XML. Furthermore, Relation Extraction models identify relationships between entities and reflect them in the data structure, and Autoencoder-based models can perform conversion of unstructured data into a structured format.

[0368] FIG. 4 is a diagram illustrating the step of identifying the need for additional information of a pre-learned legal pattern database tool (133, 230) according to one embodiment of the present invention.

[0369] Data storage and management step (S300)

[0370] Data classification step by litigation type (S300-1)

[0371] According to one embodiment of the present invention, in the data classification step (S300-1) by litigation type, a pre-trained legal pattern database tool (133, 230) can perform the role of classifying which litigation type the data belongs to by analyzing collected voice information or text information input from a consultation interface tool (135, 250). Specifically, the pre-trained legal pattern database tool (133, 230) operates based on a database that systematically organizes necessary information and characteristics for each litigation type based on data learned in the past, and can perform appropriate type classification by comparing the input data with the patterns of each litigation type stored in the database. For example, in the case of a traffic accident litigation type, essential information such as the date, time, place, perpetrator information, and details of damage may be included, and in the case of a contract dispute litigation, the contract details, counterparty obligations, violations, and amount of damage may be included.

[0372] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) may utilize a deep learning model. Specifically, the deep learning model operates based on natural language processing technology and can analyze contextual meaning by embedding text information collected through a pre-trained model (BERT, RoBERTa, etc.). Subsequently, the text information can be converted into structured information and classification tasks can be performed through an RNN, LSTM, or CNN-based model. In addition, Transformer-based models demonstrate high performance in contextual analysis and keyword extraction, and can improve classification accuracy by highlighting important parts within collected voice information or text information through an attention mechanism. Furthermore, by utilizing a multi-task learning model, tasks such as classifying litigation types, identifying missing information, and generating additional questions can be processed simultaneously.

[0373] Important information storage step (S300-2)

[0374] According to one embodiment of the present invention, in the important information storage step (S300-2), the pre-trained legal pattern database tool (133, 230) can store important information and patterns pre-defined by type of litigation and perform the role of analyzing whether essential information is missing based on input data. Specifically, the pre-trained legal pattern database tool (133, 230) can analyze collected voice information or text information input through the consultation interface tool (135, 250) using natural language processing technology, and then compare this data with the pre-trained legal pattern database to check whether essential information necessary for the progress of litigation is included. In this case, if the essential information is missing or insufficient, the pre-trained legal pattern database tool (133, 230) can identify this and generate additional questions to supplement the information. For example, if essential information such as the date or location of the accident is missing in a traffic accident lawsuit, the pre-trained legal pattern database tool (133, 230) can detect this and generate an appropriate question such as "Would you please tell me the date the accident occurred?"

[0375] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) may use a pre-trained NLP model (BERT, RoBERTa, etc.) to analyze input data, thereby embedding collected voice information or text information and extracting important keywords and information. Specifically, the extracted information is compared with essential information patterns in the database, and a dense layer may be utilized to identify missing information. Additionally, when missing information is identified, a Seq2Seq (Sequence-to-Sequence) model may generate additional questions. In this case, collected voice information or text information from various litigation cases and essential information patterns by litigation type are used as training data, and the model may be trained using evaluation metrics such as Cross Entropy Loss and BLEU Score to evaluate the accuracy of identifying missing information and the quality of the generated questions.

[0376] Database update step (S300-3)

[0377] According to one embodiment of the present invention, in the database update step (S300-3), the pre-trained legal pattern database tool (133, 230) can integrate new information into and update the existing database. Specifically, the pre-trained legal pattern database tool (133, 230) can determine whether the input new information is information already included in the existing database or whether it needs to be added as new information. For example, when a pattern stored in the existing database as a "traffic accident" type includes information such as the date and location of the accident and the perpetrator, if the newly input data includes weather information at the time of the accident, such information can be added to the existing pattern or supplemented to the existing information. In addition, the pre-trained legal pattern database tool (133, 230) can detect if there is a lack of consistency or inconsistency between the existing data and the input data, and update the database.

[0378] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can use a pre-trained NLP model (BERT, GPT, etc.) to embed new information and perform a comparison with existing patterns in the database. In this case, the comparison result is analyzed through semantic similarity (Cosine Similarity), and based on this, it can be determined whether to "update," "add," or "exclude" the new information from the existing data. In particular, if the new information is similar to existing patterns but has some differences, the pre-trained legal pattern database tool (133, 230) can use an LSTM or Transformer-based deep learning model to update the patterns in the database by reflecting the temporal context.

[0379] Data search and lookup step (S310)

[0380] Case search step (S310-1)

[0381] According to one embodiment of the present invention, in the case search step (S310-1), the pre-learned legal pattern database tool (133, 230) can perform the role of searching for similar cases stored in the database based on case information provided by the client and providing key information of the case. Additionally, the pre-learned legal pattern database tool (133, 230) includes case data organized by litigation type and can select the most suitable case by comparing it with the input case information. Furthermore, the pre-learned legal pattern database tool (133, 230) can evaluate similarity with cases stored in the database after analyzing input data (e.g., keywords, date, location, information of related parties, etc.) during the case search process. Additionally, the search scope can be narrowed by filtering cases by litigation type. For example, if the input data is related to a traffic accident, the pre-learned legal pattern database tool (133, 230) searches only for cases corresponding to the traffic accident litigation type in the database and can provide a summary of key information of the case (case overview, case law results, relevant legal grounds, etc.).

[0382] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can convert text into a high-dimensional vector by utilizing a pre-trained NLP model (BERT, Sentence-BERT, etc.) to analyze input data, and can evaluate the similarity with cases stored in the database through cosine similarity.

[0383] Step to look up Pill information by type (S310-2)

[0384] According to the present invention, in the type-specific essential information lookup step (S310-2), a pre-trained legal pattern database tool (133, 230) can analyze case information provided by the client and perform the role of verifying whether the essential information required for each type of litigation is satisfied. Specifically, the database stores pre-defined essential information templates for each type of litigation, and these templates may include information items required according to the type of litigation (e.g., in the case of a traffic accident, the date of the accident, location, perpetrator information, etc.). In addition, the pre-trained legal pattern database tool (133, 230) can effectively identify missing information by matching the input data with the essential information template. For example, in the case of a traffic accident, if accident vehicle information is missing, the pre-trained legal pattern database tool (133, 230) can detect this and return the information so that it can be requested.

[0385] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can convert input case data into text embeddings by utilizing a pre-trained NLP model (BERT, GPT, etc.) and identify missing information by comparing it with essential information templates stored in the database. Specifically, during the comparison process, similarity between data can be evaluated using semantic similarity calculation techniques such as cosine similarity. Additionally, the pre-trained legal pattern database tool (133, 230) can update database templates by utilizing a Sequence-to-Sequence model.

[0386] Pattern learning and utilization stage (S320)

[0387] Pattern saving step (S320-1)

[0388] According to one embodiment of the present invention, in the pattern storage step (S320-1), a pre-trained legal pattern database tool (133, 230) can analyze input case data to learn repetitive and important features by litigation type as patterns and store them in a database. Specifically, the database can generate patterns based on unique features appearing in the case data according to various litigation types (e.g., major keywords, flow of the case, relevant laws, etc.). Additionally, when case data is input, the pre-trained legal pattern database tool (133, 230) can analyze the data to extract key information and perform normalization. Furthermore, the pre-trained legal pattern database tool (133, 230) can identify unique patterns by litigation type by analyzing keywords or sentence structures that appear repeatedly in the case text and the temporal flow of the case. For example, in the case of a traffic accident, information such as the date and location of the accident, the relationship between the perpetrator and the victim, and the cause of the accident may mainly be included in the pattern.

[0389] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can convert event text information into embeddings using a pre-trained NLP model (BERT, RoBERTa, etc.) and then extract key features through a CNN or Transformer model. Here, the extracted features can generate patterns by learning the temporal flow and sequential characteristics of the event using an LSTM or GRU model. Accordingly, the generated patterns can be stored in a database and can be updated or expanded by comparing them with new data.

[0390] Pattern matching step (S320-2)

[0391] According to the present invention, in the pattern matching step (S320-2), a pre-trained legal pattern database tool (133, 230) can perform the role of identifying which litigation type the input data belongs to by comparing the input case data with litigation type-specific patterns stored in the database, and providing a matching result based on similarity. Specifically, the pre-trained legal pattern database tool (133, 230) can compare the input case data with litigation type-specific patterns stored in the database to determine whether the input data satisfies the essential elements of the corresponding type or if there is missing information. For example, a pattern for a traffic accident type includes the accident date, location, vehicle information, etc., and the pre-trained legal pattern database tool (133, 230) can identify and return the accident date if it is missing from the input data. In addition, the pre-trained legal pattern database tool (133, 230) can determine which litigation type the input data is most suitable for by calculating the similarity between the input data and the pattern. Accordingly, a pre-trained legal pattern database tool (133, 230) can classify input data into specific litigation types (e.g., traffic accidents, contract disputes, etc.).

[0392] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can embed input case data using a pre-trained NLP model (BERT, Sentence-BERT, etc.) and calculate cosine similarity with a pattern stored in a database. Such similarity indicates the degree of agreement between the input data and the pattern, and the pre-trained legal pattern database tool (133, 230) can select the pattern showing the highest similarity.

[0393] Database-based additional information generation step (S330)

[0394] Missing Information Inference Step (S330-1)

[0395] According to one embodiment of the present invention, in the missing information inference step (S330-1), a pre-trained legal pattern database tool (133, 230) can analyze input case data to identify missing items and, to supplement them, utilize patterns by litigation type and past case data stored in the database to infer missing information. Specifically, the pre-trained legal pattern database tool (133, 230) can compare the input data with a template of essential information required for each type of litigation. The template defines information that must be included in each type, and for example, in a traffic accident case, the date of the accident, location, vehicle information, etc., may be included as essential items.

[0396] That is, the pre-trained legal pattern database tool (133, 230) can check whether such essential information is missing from the input case data and identify the missing items. In this case, if the missing information is identified, the pre-trained legal pattern database tool (133, 230) can infer the information by utilizing past similar case data stored in the database. For example, if the accident location is missing in a traffic accident type of case, the pre-trained legal pattern database tool (133, 230) can infer a common location related to the accident type (e.g., intersection, highway, etc.) based on past similar cases.

[0397] Additional question generation step (S330-2)

[0398] According to one embodiment of the present invention, in the additional question generation step (S330-2), the pre-trained legal pattern database tool (133, 230) can perform the role of generating appropriate additional questions to supplement the missing information identified in the previous step. Specifically, if the missing essential information is identified through comparison with the database, the pre-trained legal pattern database tool (133, 230) can generate questions based thereon. For example, if the accident date is missing in a traffic accident case, the pre-trained legal pattern database tool (133, 230) can generate specific questions such as "When was the date the accident occurred?". In this case, the pre-trained legal pattern database tool (133, 230) can determine the priority of the questions. Specifically, the pre-trained legal pattern database tool (133, 230) can prioritize asking about the most important items among the missing information for litigation preparation. For example, in a traffic accident case, essential information such as the location or date of the accident can be prioritized in the question.

[0399] Performance optimization step (S340)

[0400] According to one embodiment of the present invention, in the performance optimization step (S340), the pre-trained legal pattern database tool (133, 230) can perform continuous improvement and optimization to increase the processing efficiency and analysis accuracy of the database. Specifically, the pre-trained legal pattern database tool (133, 230) can optimize the database structure to remove duplicate or unnecessary data and systematically classify and manage information by litigation type. In this case, the pre-trained legal pattern database tool (133, 230) can improve the speed of data search and analysis by introducing data normalization, efficient indexing, caching techniques, etc. Additionally, the pre-trained legal pattern database tool (133, 230) can maintain data reliability by reviewing data quality for performance optimization, correcting incomplete or erroneous data, and removing duplicate case data.

[0401] Model training support stage (S350)

[0402] Data provision step (S350-1)

[0403] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can select data suitable for model training from data by litigation type stored in a database. Specifically, the pre-trained legal pattern database tool (133, 230) can remove noise data (errors, duplicates, etc.) and unify the data format through normalization. In addition, the pre-trained legal pattern database tool (133, 230) can select data such as traffic accidents and contract disputes that the model needs to learn. Data can be filtered and provided to match specific litigation types. Accordingly, the pre-trained legal pattern database tool (133, 230) can increase data efficiency by including only information that matches the model's learning objectives. In addition, the pre-trained legal pattern database tool (133, 230) can provide information such as case description -> litigation type required for model training. Input data and correct answer labels can be linked and provided.

[0404] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can generate balanced data so that the training dataset is not biased toward a specific type of litigation. Additionally, the pre-trained legal pattern database tool (133, 230) can store in the database records of when the provided data was used for model training.

[0405] Evaluation data provision step (S350-2)

[0406] According to one embodiment of the present invention, a pre-trained legal pattern database tool (133, 230) can prepare test data by separating case data stored in a database so as not to overlap with training data. Additionally, the pre-trained legal pattern database tool (133, 230) can generate a test set containing various types of data through random sampling from the database. Furthermore, the pre-trained legal pattern database tool (133, 230) can perform verification to ensure that evaluation data does not overlap with data used in the training process, and together with the evaluation data, metrics such as Precision, Recall, and F1 Score for performance measurement It can provide standards. In addition, the pre-trained legal pattern database tool (133, 230) can manage the training and evaluation history by storing the results of the model's performance evaluation in the database.

[0407] Security and management steps (S360)

[0408] According to one embodiment of the present invention, in the security and management step (S360), the pre-trained legal pattern database tool (133, 230) can apply role-based access control (RBAC) so that only authenticated users can access the database through data access control. Specifically, the pre-trained legal pattern database tool (133, 230) can verify the identity of the user and block unauthorized access to sensitive data. In addition, the pre-trained legal pattern database tool (133, 230) records all operations, such as updating or deleting data, so that they can be traced, thereby preventing incorrect data modification or damage to integrity.

[0409] FIG. 5 is a diagram illustrating the advanced factual relationship analysis and supplementation steps of an advanced factual relationship analysis and supplementation tool according to one embodiment of the present invention.

[0410] Missing Information Identification Step (S400)

[0411] Essential Information Comparison Step (S400-1)

[0412] According to one embodiment of the present invention, in the essential information comparison step (S400-1), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of identifying missing information by comparing the data provided by the client with the essential information templates for each type of lawsuit stored in the database. At this time, the essential information templates in the database consist of information that must be included in each type of lawsuit (e.g., in the case of a traffic accident, the date of the accident, the location, information on the perpetrator, etc.), and the advanced factual relationship analysis and supplementary tool (134, 240) can identify missing information by comparing the provided data with these templates. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can list the missing information based on the comparison results and provide data so that it can be used to generate additional questions. For example, if the date of the accident is not provided, the advanced factual relationship analysis and supplementary tool (134, 240) can identify this as missing information and transmit it to the question generation step (S410). Additionally, the advanced factual analysis and supplementary tool (134, 240) can verify the accuracy and consistency of the provided data to additionally identify data that is inconsistent with or incomplete with essential information. Here, the advanced factual analysis and supplementary tool (134, 240) can evaluate the importance and priority of the missing information.

[0413] Data uncertainty verification step (S400-2)

[0414] According to one embodiment of the present invention, in the data uncertainty verification step (S400-2), the advanced factual relationship analysis and supplementary tool (134, 240) can identify incomplete or ambiguous information among the data provided by the client. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can detect ambiguous or incomplete information in the provided data by utilizing text analysis and natural language processing (NLP) technologies. For example, if the date of the accident is not specific, such as "early 2024," or if the description of the incident is brief and lacks the necessary context, the advanced factual relationship analysis and supplementary tool (134, 240) can recognize this as uncertain data. Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze uncertainty by comparing the provided data with the essential information templates and case patterns for each litigation type in the database. For example, if the date of the accident, which is important information in a lawsuit, is unclear, an advanced factual analysis and supplementary tool (134, 240) can classify it as data that needs to be resolved first.

[0415] Appropriate question generation step (S410)

[0416] Missing Information-Based Question Generation Step (S410-1)

[0417] According to one embodiment of the present invention, in the missing information-based question generation step (S410-1) Advanced factual analysis and supplementary tools (134, 240) It can perform the role of automatically generating necessary questions to supplement missing information identified in the previous step. Specifically, the advanced factual analysis and supplementation tool (134, 240) can perform the role of supplementing information missing in the previous step, such as the date and location of the accident. The identified items can be analyzed, and questions can be generated based on them. That is, the advanced factual analysis and supplementation tool (134, 240) can automatically generate questions suitable for missing information by utilizing question templates for each litigation type stored in the database. For example, if the date of the accident is missing, a question such as "What is the exact date the accident occurred?" can be generated. In this case, the advanced factual analysis and supplementation tool (134, 240) can structure the question in a specific and clear form, customize the question to reflect the context of the case, and generate open-ended or multiple-choice questions as needed. For example, the advanced factual analysis and supplementation tool (134, 240) can generate open-ended questions such as "Could you please explain the circumstances of the accident?" or multiple-choice questions such as "Which of A, B, or C is the location of the accident?" In addition, the advanced factual analysis and supplementary tool (134, 240) can evaluate the importance of missing information and generate questions starting with essential information.

[0418] Custom question generation step (S410-2)

[0419] According to one embodiment of the present invention, in the customized question generation step (S410-2), the advanced factual analysis and supplementary tool (134, 240) can perform the role of generating customized questions suitable for the client by analyzing the client's situation, provided data, and the context of the case. Specifically, the advanced factual analysis and supplementary tool (134, 240) can create questions that reflect the context of the case by analyzing information provided by the client. For example, in a traffic accident case, if the client mentions that the vehicle was stopped at the time of the accident, the advanced factual analysis and supplementary tool (134, 240) can generate specific and situation-appropriate questions such as "Did a collision occur with the stopped vehicle?" In addition, the advanced factual analysis and supplementary tool (134, 240) creates questions that can grasp the essence of the case by reflecting the context specialized for each type of litigation, and can generate questions suitable for each type of case, such as traffic accidents and contract disputes.

[0420] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) can compose questions in clear and easy language, taking into account the client's understanding. In addition, the advanced factual analysis and supplementary tool (134, 240) can minimize legal terms or complex expressions and write questions concisely so that the client can answer comfortably. Furthermore, the advanced factual analysis and supplementary tool (134, 240) can also analyze the emotional state to generate questions in a soft and empathetic tone according to the client's emotional state. For example, if the client feels anxious about the accident situation, the advanced factual analysis and supplementary tool (134, 240) can use expressions such as, "Could you please explain the situation at the time of the accident slowly?"

[0421] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can generate questions that are simple and stepwise, and can divide complex questions into multiple parts to make it easier for the client to answer. For example, the advanced factual relationship analysis and supplementary tool (134, 240) can generate questions in a flexible form depending on the situation by combining open questions and multiple-choice questions. For example, the advanced factual relationship analysis and supplementary tool (134, 240) can generate questions in a flexible form depending on the situation by combining open questions such as "Can you explain the situation at the time of the accident?" and multiple-choice questions such as "Which of A, B, or C is the location of the accident?"

[0422] According to one embodiment of the present invention, an advanced factual analysis and supplementation tool (134, 240) may utilize an emotion analysis algorithm to generate customized questions by taking into account the client's understanding and emotional state. Specifically, the advanced factual analysis and supplementation tool (134, 240) analyzes text and voice data to classify the client's emotions and determines an appropriate tone and question content based on this. The following technologies may be utilized for emotion analysis. For example, the advanced factual analysis and supplementation tool (134, 240) may utilize a BERT-based emotion classification model or a Transformer architecture to analyze input text and classify emotional states such as positive, negative, and neutral. For example, if the client says, "It is a very difficult situation," the advanced factual analysis and supplementation tool (134, 240) may classify this as a negative emotion and generate a softer question reflecting it. In addition, the advanced factual analysis and supplementary tool (134, 240) uses voice feature extraction techniques such as Mel-Frequency Cepstral Coefficients (MFCCs) to analyze voice data and combines this with a Long Short-Term Memory (LSTM) network to identify the emotional state based on the client's voice tone, speed, intonation, etc. For example, if the client's voice trembles or is slow, it can be analyzed as an anxious state.

[0423] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) can more precisely identify the client's emotions by integrating text and voice data to perform multimodal emotion analysis. Through this, even if negative words are found in the text, if a stable tone is detected in the voice, the advanced factual analysis and supplementary tool (134, 240) can determine the emotion neutrally. Through this emotion analysis method, the advanced factual analysis and supplementary tool (134, 240) can generate questions suitable for the client's emotional state. For example, if the client feels anxious, it can generate questions with an empathetic tone such as "Could you please explain the situation at the time of the accident slowly?", and if the client is in a neutral state, it can generate specific questions such as "Could you tell me the exact location of the accident?"

[0424] Natural language question generation step (S410-3)

[0425] According to one embodiment of the present invention, in the natural language question generation step (S410-3), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of generating natural questions that the client can easily understand and answer. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) analyzes the data and conversation records provided by the client to understand the context of the event and necessary additional information, and can write questions in a natural conversational format instead of stiff or formalized expressions.

[0426] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) can generate natural questions based on the client's conversation content and incident data by utilizing data analysis and natural language processing (NLP) technology. For example, if the date of the accident is provided unclear, the advanced factual analysis and supplementary tool (134, 240) can generate soft, conversational questions such as, "Could you tell me the exact date the accident occurred?" Additionally, the advanced factual analysis and supplementary tool (134, 240) can reflect the flow and context of the conversation and continue the questioning based on previous questions or answers, thereby reducing the sense of disconnection the client may feel and maintaining the continuity of the conversation.

[0427] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) adjusts the tone of the question by taking into account the client's emotional state. For example, if the client expresses anxiety about the accident situation, the advanced factual analysis and supplementary tool (134, 240) can generate empathetic and gentle questions such as, "Please take your time thinking and could you explain the situation at the time of the accident?" Specifically, when drafting questions, the advanced factual analysis and supplementary tool (134, 240) can exclude legal terms or complex expressions as much as possible and compose questions in concise and easy language. In addition, the advanced factual analysis and supplementary tool (134, 240) can use an appropriate mix of open questions and closed questions depending on the situation. For example, the advanced factual analysis and supplementary tool (134, 240) can use an appropriate mix of open questions such as, "Please tell me the situation at the time of the accident," and closed questions such as, "Was the road wet at the time of the accident?" depending on the situation.

[0428] Question order optimization step (S410-4)

[0429] According to one embodiment of the present invention, in the question order optimization step (S410-4), the advanced factual relationship analysis and supplementary tool (134, 240) can arrange the generated questions in an efficient order to smooth the flow of consultation and effectively collect necessary information. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can perform optimization in various aspects, such as logical arrangement of questions, setting priorities, removing duplicates, and maintaining the continuity of conversation. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can logically arrange the generated questions to match the context of the case and the flow of data provided by the client. For example, in the case of a traffic accident, the advanced factual relationship analysis and supplementary tool (134, 240) can perform arrangement in the order of first asking for basic information such as the date, location, and type of accident, and then asking for additional information such as road conditions or weather.

[0430] According to one embodiment of the present invention, an advanced factual analysis and supplementary tool (134, 240) can be designed to evaluate the priority of questions and collect information essential for litigation preparation first. Specifically, the advanced factual analysis and supplementary tool (134, 240) can collect important information in the initial stage and place complex or time-consuming questions in a later stage, thereby providing the effect of reducing the burden on the client. In addition, the advanced factual analysis and supplementary tool (134, 240) can remove or modify questions that overlap with previously answered content, thereby increasing the efficiency of the conversation and preventing inconvenience caused by repetitive questions.

[0431] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) may prioritize presenting questions linked to the client's previous answers to maintain the flow and continuity of the conversation. For example, the advanced factual analysis and supplementary tool (134, 240) may induce a natural conversation by following up with related questions, such as "Where was the other vehicle located?" after the client explains the location of the vehicle at the time of the accident. Additionally, the advanced factual analysis and supplementary tool (134, 240) may analyze the client's response data to place simple questions that can be answered within a short time at the beginning, and more complex questions at the end, thereby designing the consultation to proceed gradually.

[0432] User understanding and emotional state analysis step (S420)

[0433] User Understanding Analysis Step (S420-1)

[0434] According to one embodiment of the present invention, in the user comprehension analysis step (S420-1), the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's conversation content and answers to evaluate how well the client understands the presented questions, and based on this, perform the role of adjusting the questions or optimizing the conversation method. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) evaluates the understanding of the questions by analyzing the words and sentence structure used in the client's answers and the context of the answers, and can adjust the difficulty of the questions or change the explanation method as necessary. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze whether legal terms or case-related specialized terms are used in the client's answers and perform an evaluation of whether the client sufficiently understands the said terms. At this time, if the client does not accurately understand the specialized terms or is confused, the advanced factual relationship analysis and supplementary tool (134, 240) can explain them easily or present questions with simplified terms.

[0435] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can verify the consistency of the answer and perform an analysis to see if there are any parts that conflict or contradict with previously provided answers. In this case, if there is a lack of consistency, the advanced factual relationship analysis and supplementary tool (134, 240) can provide additional explanations or simplified questions. For example, if a complex question is not understood by the client, the advanced factual relationship analysis and supplementary tool (134, 240) can simplify the question or convert it into a multiple-choice question. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can adjust the speed and difficulty of the conversation according to the client's level of understanding, and can record the client's answer patterns and level of understanding and store them in a database.

[0436] Emotional state analysis step (S420-2)

[0437] According to one embodiment of the present invention, in the emotional state analysis step (S420-2), the advanced factual relationship analysis and supplementary tool (134, 240) can identify the emotional state by analyzing the client's conversation content and expression style, and accordingly, perform the role of adjusting the conversation style or optimizing questions. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can understand various emotional states of the client, such as anxiety, anger, confusion, and calmness. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can classify the emotional state by analyzing vocabulary, sentence structure, and emotional expressions used in the client's text conversation, or by analyzing the tone, speed, and emphasis of the voice if voice data is provided. For example, the advanced factual relationship analysis and supplementary tool (134, 240) can classify expressions such as "I'm worried" as an anxiety state and "I'm really angry" as an anger state.

[0438] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) uses a soft and stable tone when the client is in an anxious state, and can provide the effect of reducing psychological burden through expressions such as, for example, "Please take your time to think and tell me." When the client is in an angry state, the advanced factual analysis and supplementary tool (134, 240) can soothe emotions and continue the conversation by using empathetic and stable expressions such as, "You must have had a very hard time in that situation. I would appreciate it if you could tell me what I can help you with."

[0439] According to one embodiment of the present invention, the advanced factual analysis and supplementary tool (134, 240) can adjust the difficulty of the question according to the emotional state or convert the question into a simple and specific form. For example, the advanced factual analysis and supplementary tool (134, 240) can present a simple question such as "Where was the location of the accident?" instead of the question "Can you explain the situation at the time of the accident?" In addition, the advanced factual analysis and supplementary tool (134, 240) can adjust the speed and content of the conversation to provide the client with sufficient time to think and answer. Specifically, the advanced factual analysis and supplementary tool (134, 240) can provide time to understand the situation through additional explanations or examples to a client who is in a confused state.

[0440] Counseling conversation management step (S430)

[0441] Conversation record management step (S430-1)

[0442] According to one embodiment of the present invention, in the conversation record management step (S430-1), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of recording the content of the conversation with the client in real time. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can store the conversation content, including questions and answers that occur during the consultation, in various formats such as text and voice data. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can manage the stored collected voice information or text information in a structured form by classifying it into event type, chronological order, question category, etc. For example, in the case of a traffic accident, the advanced factual relationship analysis and supplementary tool (134, 240) can organize the conversation content into items such as the date of the accident, location, and type of accident. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) may provide a keyword search function.

[0443] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) can link the current ongoing consultation with previous consultation records to maintain the continuity of the consultation. Accordingly, the advanced factual relationship analysis and supplementary tool (134, 240) can prevent repetitive questions and effectively refer to the client's case progress and past answers. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can manage data security by encrypting the client's sensitive information. Furthermore, the advanced factual relationship analysis and supplementary tool (134, 240) can visualize recorded conversation content in the form of a summary report. For example, the advanced factual relationship analysis and supplementary tool (134, 240) can organize important information, such as the date and location of an accident, in a table format. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze conversation record data.

[0444] Conversation flow maintenance step (S430-2)

[0445] According to one embodiment of the present invention, in the conversation flow maintenance step (S430-2), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of ensuring the logical continuity and natural progression of the conversation during the consultation. Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze the client's previous answers and the context of the conversation to appropriately adjust questions or generate new questions. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can dynamically adjust the speed of the conversation according to the client's response pattern and speed. For example, if the client has difficulty answering or takes a long time, the advanced factual relationship analysis and supplementary tool (134, 240) can switch to simpler and more specific questions or increase the interval between questions to provide the client with sufficient time to think. Furthermore, the advanced factual relationship analysis and supplementary tool (134, 240) can maintain the logical order of the conversation and arrange questions so that the collection of case information can be carried out step by step. For example, an advanced factual analysis and supplementary tool (134, 240) can first ask about the date of the accident and then ask about the location or situation of the accident.

[0446] According to one embodiment of the present invention, if a client has difficulty answering a specific question during a conversation, the advanced factual analysis and supplementary tool (134, 240) can guide the conversation by providing hints or examples. For example, if a client has difficulty explaining the location of the vehicle at the time of the accident, the advanced factual analysis and supplementary tool (134, 240) can present specific options such as, "Did the other vehicle come from the left or from the right?" Additionally, the advanced factual analysis and supplementary tool (134, 240) can monitor the flow of the conversation by analyzing the client's response data in real time during the conversation, and collect feedback to record it so that it can be reflected in the subsequent consultation process.

[0447] Result data cleaning step (S440)

[0448] Information supplementation stage (S440-1)

[0449] According to one embodiment of the present invention, in the information supplementation step (S440-1), the advanced factual relationship analysis and supplementation tool (134, 240) can perform the role of reviewing data collected during the consultation process to identify missing or incomplete information and systematically supplement it. In addition, the advanced factual relationship analysis and supplementation tool (134, 240) can compare the data provided by the client with the essential information templates for each type of litigation stored in the database to secure additional missing data or supplement existing data. Specifically, the advanced factual relationship analysis and supplementation tool (134, 240) can identify missing information by comparing the data collected during the consultation with the essential information templates. For example, the advanced factual relationship analysis and supplementation tool (134, 240) can verify whether all essential information, such as the date, location, and type of accident in a traffic accident case, is included. In this case, if missing information is found, the advanced factual relationship analysis and supplementation tool (134, 240) can list it and use it as a basis for supplementation work.

[0450] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementation tool (134, 240) can review ambiguous or uncertain parts of the data provided by the client. For example, the advanced factual relationship analysis and supplementation tool (134, 240) can supplement incomplete data, such as "the date of the accident is early 2024," with more specific information.

[0451] According to one embodiment of the present invention, the advanced factual analysis and supplementation tool (134, 240) may generate additional questions to supplement missing or incomplete information. For example, the advanced factual analysis and supplementation tool (134, 240) may refine data through questions such as "When exactly is the date of the accident?" or "Can you tell me the location of the accident a little more specifically?" Additionally, the advanced factual analysis and supplementation tool (134, 240) may correct conflicting or inconsistent data by verifying consistency between the collected data and existing data. For example, if different dates are provided for the same event, the advanced factual analysis and supplementation tool (134, 240) may correct the data through an additional verification process.

[0452] Final consultation data organization stage (S440-2)

[0453] According to one embodiment of the present invention, in the final consultation data organization step (S440-2), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of organizing data collected and supplemented during the consultation process to complete the final consultation result. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can integrate all data collected during the consultation and classify it by case. For example, the advanced factual relationship analysis and supplementary tool (134, 240) can organize information such as the date of the accident, the location, and the scale of damage in the case of a traffic accident, and organize the date of contract signing, details of the violation, and the amount of damages in the case of a contract dispute.

[0454] Model training support stage (S450)

[0455] Question Generation Model Training Data Preparation Step (S450-1)

[0456] According to one embodiment of the present invention, in the question generation model training data preparation step (S450-1), the advanced factual relationship analysis and supplementary tool (134, 240) can perform the role of preparing a dataset necessary for training the question generation model. Additionally, the advanced factual relationship analysis and supplementary tool (134, 240) can analyze collected voice information or text information collected during the consultation process and organize questions and answers into pairs to process them into a form suitable for model training. Specifically, the advanced factual relationship analysis and supplementary tool (134, 240) can collect collected voice information or text information recorded during the consultation and generate meaningful question-answer pairs by linking the client's answer with the question regarding it. For example, if the client's answer is "The accident happened in Gangnam," the advanced factual relationship analysis and supplementary tool (134, 240) can connect the related question "Where did the accident happen?" to form training data.

[0457] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementation tool (134, 240) can improve the quality of training data by refining and preprocessing data. Specifically, the advanced factual relationship analysis and supplementation tool (134, 240) can remove duplicate data or incomplete information and add necessary contextual information to the data. For example, the advanced factual relationship analysis and supplementation tool (134, 240) can include contextual data, such as consultation topics (traffic accidents, contract disputes, etc.) or the client's emotional state, in the training data. In addition, the advanced factual relationship analysis and supplementation tool (134, 240) can label and categorize data to improve the efficiency of the training data. At this time, the advanced factual relationship analysis and supplementation tool (134, 240) can classify questions and answers by litigation type or distinguish them by question type.

[0458] Emotion-based question model learning phase (S450-2)

[0459] According to one embodiment of the present invention, in the emotion-based question model learning step (S450-2) Advanced factual analysis and supplementary tools (134, 240)The tool can combine emotional state data and question-and-answer data collected during the counseling process. Specifically, the advanced factual analysis and supplementary tool (134, 240) can perform training data organization by linking emotional states (e.g., anxiety, anger, confusion, etc.) recorded during counseling with question-and-answer data. Additionally, the advanced factual analysis and supplementary tool (134, 240) can configure the data by adding the client's emotional state as a label to each question-and-answer pair so that the model can reflect the emotional state when generating questions. Furthermore, the advanced factual analysis and supplementary tool (134, 240) can configure different question types depending on the emotional state. For example, the advanced factual analysis and supplementary tool (134, 240) can configure simple multiple-choice questions for an anxious state and open-ended questions that provide hints for a confused state. Additionally, the advanced factual analysis and supplementary tool (134, 240) can refine and preprocess the emotional data to remove duplicate or erroneous data.

[0460] According to one embodiment of the present invention, the advanced factual relationship analysis and supplementary tool (134, 240) in the model learning process may include various emotional state data to maintain the balance of the learning data and not be biased toward a specific emotional state. In addition, the advanced factual relationship analysis and supplementary tool (134, 240) may evaluate the performance of the learned model to determine whether it can generate appropriate questions according to the emotional state, and if necessary, improve the model performance by adding or modifying data.

[0461] FIG. 6 is a diagram illustrating the counseling progress and information supplementation steps of a counseling interface tool according to one embodiment of the present invention.

[0462] Question delivery stage (S500)

[0463] Question format setting step (S500-1)

[0464] According to the present invention, in the question format setting step (S500-1) of the question delivery step, the consultation interface tool (135, 250) can perform the role of analyzing the client's input data using a deep learning model and automatically generating a suitable question format. To this end, the consultation interface tool (135, 250) can design and utilize a natural language processing-based model, a sentiment analysis model, a user intent classification model, a conversation flow maintenance model, etc. Specifically, the natural language processing model can understand the conversation context using Transformer-based technology (BERT, GPT, etc.), generate suitable questions according to the client's litigation purpose, and summarize long sentences or reconstruct them into specific questions using a Seq2Seq model. In addition, the sentiment analysis model analyzes the client's emotional state in real time to adjust the tone of the questions, and can present soft and trustworthy questions to, for example, a client who is in a tense state.

[0465] According to the present invention, a consultation interface tool (135, 250) can utilize a user intent classification model to analyze a client's utterance to classify the intent of the question and suggest a predefined question format for each type of litigation. Additionally, the consultation interface tool (135, 250) can use a Reinforcement Learning-based conversation management model to maintain the flow of the conversation naturally and generate additional questions to supplement missing information. Furthermore, the consultation interface tool (135, 250) can analyze the client's voice or text input in real time to construct appropriate questions, extract necessary information step-by-step during the conversation, and provide additional questions to supplement missing information.

[0466] Channel selection step (S500-2)

[0467] According to the present invention, in the channel selection step (S500-2) of the question transmission step (S500), the consultation interface tool (135, 250) can automatically determine or suggest an appropriate channel to the client by considering the client's preference, accessibility, and sensitivity of the consultation topic. Specifically, the consultation interface tool (135, 250) can select one of the channels, such as voice, text, or video consultation, by analyzing data entered by the client (e.g., options selected upon initial connection or information set in advance). In this case, the channel capable of the smoothest connection can be set by evaluating the status of devices such as smartphones and PCs currently available to the client and the network status in real time. For example, if the network status is unstable, the consultation interface tool (135, 250) can preferentially recommend a text-based channel instead of a video channel. Additionally, the consultation interface tool (135, 250) can provide flexibility regarding channel changes and provide a function to switch channels during the consultation. For example, the consultation interface tool (135, 250) can immediately switch to the corresponding channel when a client requests to switch to voice consultation during text consultation, and can connect the existing conversation content to the new channel so that the flow of conversation is not interrupted.

[0468] According to one embodiment of the present invention, a consultation interface tool (135, 250) can optimize a channel according to the subject of the consultation. For example, if the subject includes sensitive personal information or legal disputes, the consultation interface tool (135, 250) can recommend and set a channel such as a highly secure encrypted video conferencing platform.

[0469] Question Priority Management Step (S500-3)

[0470] According to one embodiment of the present invention, in the question priority management step (S500-3) of the question delivery step (S500), the consultation interface tool (135, 250) can perform the role of evaluating the importance of questions and setting priorities according to the client's consultation purpose and situation. Specifically, the consultation interface tool (135, 250) can identify the topic of consultation by analyzing initial input data provided by the client. For example, in the case of a traffic accident damage compensation lawsuit, the consultation interface tool (135, 250) must first verify essential information such as the date, time, location, and damage situation of the accident. Here, the consultation interface tool (135, 250) can classify questions of high importance by lawsuit type by utilizing a pre-learned legal pattern database.

[0471] According to one embodiment of the present invention, a counseling interface tool (135, 250) can update the priority of questions based on the client's response. Specifically, the counseling interface tool (135, 250) can analyze the response to previous questions to determine additional information needed and present new questions in a timely manner. For example, if the client answers, "I do not know the contact information of the perpetrator at the time of the accident," the counseling interface tool (135, 250) can prioritize presenting alternative questions, such as vehicle license plate numbers or witness information, to identify the perpetrator. Additionally, the counseling interface tool (135, 250) can also consider the client's time constraints or emotional state. If the client requests counseling in an urgent situation, the counseling interface tool (135, 250) can shorten the counseling time by setting the priority of questions based on essential information. Conversely, if the client is anxious or nervous, the counseling interface tool (135, 250) can adjust the difficulty of the questions or start with less important questions to build trust and then switch to key questions.

[0472] Answer collection stage (S510)

[0473] Answer input support step (S510-1)

[0474] According to one embodiment of the present invention, in the answer input support step (S510-1) of the answer collection step (S510), the consultation interface tool (135, 250) can perform the role of supporting the client in various ways so that the client can smoothly answer the question. Specifically, the consultation interface tool (135, 250) can support answer input by utilizing a user interface so that the client can perform at least one of voice input, text input, and a choice-type response.

[0475] According to one embodiment of the present invention, in the response input support step (S510-1), the consultation interface tool (135, 250) can suggest a suitable method according to the client's situation and preference. Additionally, the consultation interface tool (135, 250) can provide an interface that induces voice input when driving or when hands are not free, and allows for a simple response via a selectable button when writing is difficult.

[0476] Client Input Assistance

[0477] According to one embodiment of the present invention, the answer data input support step may further include the step of providing at least one of a text auto-completion function, a keyword recommendation function, a dynamic question generation function, a real-time spelling correction function, and a real-time grammar correction function when the client performs the text input method.

[0478] According to one embodiment of the present invention, a consultation interface tool (135, 250) can assist the client in the input process. Specifically, the consultation interface tool (135, 250) supports the client in easily writing answers through an auto-completion function or a keyword recommendation function when inputting text, and when inputting voice, it converts speech into text through real-time voice recognition and provides correction options to reduce recognition errors. In addition, the consultation interface tool (135, 250) can provide clear guidance on questions. Specifically, in the case of complex or abstract questions, the consultation interface tool (135, 250) can provide additional specific examples or explanations. For example, regarding the question "Please describe the situation at the time of the accident," the consultation interface tool (135, 250) can add a specific explanation such as "Please briefly describe the date, time, place, and surrounding environment at the time of the accident."

[0479] Input format validation step (S510-2)

[0480] According to one embodiment of the present invention, in the input format verification step (S510-2) of the answer collection step (S510), the consultation interface tool (135, 250) can perform the role of verifying in real time whether the answer provided by the client conforms to the required input format and criteria. Specifically, the consultation interface tool (135, 250) can pre-define the format of the answer according to the requirements of the question. For example, the consultation interface tool (135, 250) can set criteria for the input format, such as requiring the "YYYY-MM-DD" format for a question requiring the input of a date, or requiring an answer containing only numbers for a question requiring the input of an amount.

[0481] According to one embodiment of the present invention, when a client enters an answer, the consultation interface tool (135, 250) can verify the input data. Specifically, the consultation interface tool (135, 250) can provide feedback to the client if the input data does not conform to the required format. For example, if a date is entered in an incorrect format, the consultation interface tool (135, 250) can output a message saying "Please enter the date in YYYY-MM-DD format," or if an answer containing only numbers is entered for a question that requires text input, it can notify the client and provide an opportunity to correct it.

[0482] According to one embodiment of the present invention, a consultation interface tool (135, 250) can verify the logical consistency of input data. For example, the consultation interface tool (135, 250) can detect if an accident date is entered as a future date or if an amount is entered as an unrealistically large value, and request a review from the client. Accordingly, the consultation interface tool (135, 250) can block incorrect information in advance during the consultation process and minimize the process of additional information verification. In addition, the consultation interface tool (135, 250) can also provide automated recommendation and correction functions to enhance the convenience of the client during the input verification process. For example, the consultation interface tool (135, 250) can provide a function to automatically clean up spaces or hyphens when entering a phone number, or to correct misspelled parts in an address input.

[0483] Input data cleaning step (S510-3)

[0484] According to one embodiment of the present invention, in the input data organization step (S510-3) of the answer collection step (S510), the counseling interface tool (135, 250) can perform the role of systematically organizing the answer data provided by the client and preparing it for use in the counseling process. Specifically, the counseling interface tool (135, 250) can convert the input data provided by the client into a structured format. For example, the counseling interface tool (135, 250) can automatically extract key information such as dates, times, people, and events from answers entered in text form and store them in a predefined database field.

[0485] According to one embodiment of the present invention, the consultation interface tool (135, 250) can remove duplicate information or clean unnecessary data during the process of organizing input data. For example, the consultation interface tool (135, 250) can consolidate the same answer if it is entered multiple times, or correct typos and unnecessary spaces. In addition, the consultation interface tool (135, 250) can classify and sort input data according to the topic and purpose of the consultation. For example, in the case of a traffic accident damage compensation lawsuit, the consultation interface tool (135, 250) can prioritize organizing information essential to the progress of the lawsuit, such as the date, location, and extent of damage of the accident, while storing incidental information in a separate field. In addition, the consultation interface tool (135, 250) can organize input data, analyze the correlation between data, and store it in a form that can be visually represented or track the flow of data. For example, the counseling interface tool (135, 250) can arrange the order of events along a time axis or diagram the relationships between the people involved in the events.

[0486] Real-time conversation management step (S520)

[0487] Conversation flow maintenance step (S520-1)

[0488] According to one embodiment of the present invention, in the conversation flow maintenance step (S520-1) of the real-time conversation management step (S520), the counseling interface tool (135, 250) can perform the role of maintaining the flow of conversation so that the counseling conversation with the client continues smoothly, and managing it so that unnecessary confusion or interruption does not occur. Specifically, the counseling interface tool (135, 250) can analyze the context of the conversation in real time and present follow-up questions or guidance suitable for the client's response or question. For example, if the client provides specific information such as the date of the accident, the counseling interface tool (135, 250) can subsequently present a question such as the location of the accident that is logically connected to that information. In addition, the counseling interface tool (135, 250) can monitor the client's response speed or the progress of the conversation in real time and intervene appropriately if the conversation is interrupted or slows down. For example, if a client hesitates to respond or does not type for a long time, the consultation interface tool (135, 250) may provide guidance messages such as “It is okay to take your time answering” or “Please let me know if you need additional information.”

[0489] According to one embodiment of the present invention, the counseling interface tool (135, 250) uses clear and concise sentences to minimize confusion that may occur during conversation, and may provide additional explanations or examples if the client does not understand the intent of the question. For example, if the question "Please tell me about the accident situation" feels ambiguous to the client, the counseling interface tool (135, 250) may provide specific examples such as "Please tell me the date, time, and place where the accident occurred." In addition, the counseling interface tool (135, 250) analyzes the client's emotional state to maintain the flow of conversation and may adjust the speed and tone of the conversation according to the client's state. For example, if the client is anxious or tense, the counseling interface tool (135, 250) may proceed with the conversation in a soft and encouraging tone and provide messages that can stabilize the client's emotions.

[0490] Conversation state saving step (S520-2)

[0491] According to one embodiment of the present invention, in the conversation state saving step (S520-2) of the real-time conversation management step (S520), the consultation interface tool (135, 250) can store the conversation state and related information that occurs during the consultation process. Specifically, if the client stops the consultation or the conversation is interrupted due to a system failure, the consultation interface tool (135, 250) can retrieve the last step and context of the previous conversation to continue the conversation. In addition, if the information provided by the client is changed or supplemented, the consultation interface tool (135, 250) can immediately update the existing data and synchronize the conversation state.

[0492] Information supplementation stage (S530)

[0493] Missing Information Supplementation Step (S530-1)

[0494] According to one embodiment of the present invention, in the omitted information supplementation step (S530-1) of the information supplementation step (S530), the consultation interface tool (135, 250) can automatically identify parts that are missing from the information provided by the client and perform the role of generating and presenting additional questions to supplement them. Specifically, the consultation interface tool (135, 250) can analyze the answer data provided by the client to review whether essential information is missing. For example, the consultation interface tool (135, 250) can detect cases where the date, location, scale of damage, etc., of an accident are missing in a traffic accident damage compensation lawsuit.

[0495] According to one embodiment of the present invention, a counseling interface tool (135, 250) can generate additional questions suitable for the client's situation and context to supplement missing information. Specifically, the counseling interface tool (135, 250) can provide questions in a format that is easy for the client to understand and can be structured clearly and concisely. For example, the counseling interface tool (135, 250) can request necessary information from the client through specific questions such as "Please tell me the date of the accident" or "What is the amount of damage?"

[0496] According to one embodiment of the present invention, the consultation interface tool (135, 250) can set the priority of additional questions. Specifically, the consultation interface tool (135, 250) can prioritize questions starting with the most important information and, if the client finds it difficult to answer, can provide examples or additional explanations to elicit an answer. For example, the consultation interface tool (135, 250) can reduce the burden on the client by saying, "If you do not know the location of the accident, please tell me the approximate area or the name of a nearby building." In addition, the consultation interface tool (135, 250) can be designed to integrate supplementary information provided by the client into existing data and maintain data consistency. That is, if new information conflicts with or duplicates existing information, the consultation interface tool (135, 250) can identify this and provide an appropriate warning message or clean up the duplicate data.

[0497] Uncertain information verification step (S530-2)

[0498] According to one embodiment of the present invention, in the uncertain information verification step (S530-2) of the information supplementation step (S530), the counseling interface tool (135, 250) can perform the role of identifying ambiguous or uncertain information among the answers provided by the client and presenting additional questions to clarify it. Specifically, when uncertain information is found, the counseling interface tool (135, 250) can generate verification questions that fit the client's conversation context and present them in a concise and specific form. For example, the counseling interface tool (135, 250) can clarify unclear information through questions such as "Please confirm the date of the accident again" or "Can you tell me the amount of damage a little more specifically?" In addition, if the client finds it difficult to understand the intent of the question or has difficulty answering, the counseling interface tool (135, 250) can provide additional examples or explanations to elicit an answer. For example, the consultation interface tool (135, 250) can provide guidance such as, "Please tell us the street address or a well-known nearby location of the accident."

[0499] Emotion and Intention Analysis Step (S540)

[0500] User intent analysis step (S540-1)

[0501] According to one embodiment of the present invention, in the user intent analysis step (S540-1) of the emotion and intent analysis step (S540), the counseling interface tool (135, 250) can perform the role of identifying the user's intent in real time from the client's conversation, setting the direction of counseling based thereon, and deriving an appropriate response strategy. To this end, the counseling interface tool (135, 250) analyzes the client's input data (text, voice, etc.) using natural language processing (NLP) technology and can extract the purpose of counseling based on the context and keywords of the conversation. For example, the counseling interface tool (135, 250) can identify that 'claiming compensation for damages' is the client's main intent from the utterance "I want to claim compensation for damages due to a traffic accident."

[0502] According to one embodiment of the present invention, a consultation interface tool (135, 250) can classify a client's intentions by utilizing an intention classification model, such as a pre-trained legal pattern database and a Transformer-based model, and tag them in the consultation process. For example, the consultation interface tool (135, 250) can identify two intentions, 'claim for damages' and 'refusal of insurance processing,' from the utterance "I want to receive compensation for damages from a traffic accident, but the other party refused to process the insurance claim," and suggest appropriate response strategies for each.

[0503] According to one embodiment of the present invention, when the client's intention is identified, the counseling interface tool (135, 250) can set the direction of counseling based on this and present questions or provide relevant information to effectively achieve the client's purpose. For example, if the intention is to claim damages, the counseling interface tool (135, 250) can proceed with counseling through specific questions such as "Please tell me the date and location of the accident." If the client's intention is unclear or there is room for misunderstanding, the counseling interface tool (135, 250) can reconfirm the intention through additional questions and, if necessary, adjust the content of the conversation.

[0504] Step to identify user's emotional state (S540-2)

[0505] According to one embodiment of the present invention, in the step of identifying the user's emotional state (S540-2) of the emotion and intention analysis step (S540), the counseling interface tool (135, 250) can perform the role of analyzing the emotional state in the client's conversation and adjusting the method of proceeding with counseling and response strategies based on this. Specifically, the counseling interface tool (135, 250) can identify the emotional state by analyzing the client's text input or voice data in real time. In addition, the counseling interface tool (135, 250) can analyze the emotional implications of sentence structure and vocabulary using natural language processing (NLP) technology in the case of text input, and extract the emotional state based on the tone, speed, volume, intonation, etc. of the voice signal in the case of voice input. For example, if expressions such as "I feel so frustrated" are repeated in the client's text, the counseling interface tool (135, 250) can identify frustration, and if a high pitch and fast speech rate are detected in the voice data, anger or anxiety can be identified.

[0506] According to one embodiment of the present invention, when an emotional state is analyzed, the counseling interface tool (135, 250) can adjust the method of proceeding with counseling based on this. Specifically, if the client is in an anxious or angry state, the counseling interface tool (135, 250) can proceed with the conversation in a soft and trustworthy tone and request necessary information calmly without rushing. Conversely, if the client is in a calm or cooperative state, the counseling interface tool (135, 250) can proceed with counseling quickly by presenting more direct and efficient questions. For example, the counseling interface tool (135, 250) can provide guidance messages such as "It is okay to speak slowly" or add messages that acknowledge and empathize with the client's emotions.

[0507] According to one embodiment of the present invention, a counseling interface tool (135, 250) can continuously monitor the emotional state and manage the overall emotional flow of the counseling. Specifically, if the counseling interface tool (135, 250) shows a tendency for the client's emotional state to worsen, it can reduce the psychological burden on the client by adjusting the pace of the counseling or switching the topic. For example, when dealing with a sensitive topic, if the client becomes emotionally agitated, the counseling interface tool (135, 250) can pause the conversation and present a message such as, "I will give you some time to take a break and think."

[0508] Visual feedback and summary provision step (S550)

[0509] Real-time summary provision step (S550-1)

[0510] According to one embodiment of the present invention, in the real-time summary provision step (S550-1) of the visual feedback and summary provision step (S550), the counseling interface tool (135, 250) can summarize the content of the conversation with the client in real time during the consultation. Specifically, the counseling interface tool (135, 250) can process the collected voice information or text information with the client in real time to automatically extract important information and keywords. For example, if there is a statement in the conversation such as "traffic accident on May 12, 2023 at 3:00 PM," the counseling interface tool (135, 250) can perform a summary by organizing it into "Accident Date: May 12, 2023" and "Accident Time: 3:00 PM." In addition, the counseling interface tool (135, 250) can provide the real-time updated summary information in a visually readable format. For example, the consultation interface tool (135, 250) can organize summarized data into items or tables so that the client can easily understand it.

[0511] According to one embodiment of the present invention, a consultation interface tool (135, 250) can continuously update summary content according to the progress of the conversation. Specifically, if a client provides additional information or modifies an existing response, the tool can update the summary data to reflect this. For example, if a client corrects the date of the accident, the consultation interface tool (135, 250) can automatically modify the existing summary content to maintain the latest state.

[0512] Progress Visualization Step (S550-2)

[0513] According to one embodiment of the present invention, in the progress visualization step (S550-2) of the visual feedback and summary provision step (S550), the counseling interface tool (135, 250) can visually display the progress status of the counseling process. Specifically, the counseling interface tool (135, 250) can visually structure the main steps of the counseling. For example, the counseling interface tool (135, 250) can divide the counseling into steps such as "information collection," "information supplementation," "emotion and intention analysis," and "visual feedback and summary provision," and can distinguish and display the steps currently in progress and the completed steps.

[0514] According to one embodiment of the present invention, a counseling interface tool (135, 250) may utilize various visual elements to intuitively convey the progress status. For example, the counseling interface tool (135, 250) may display the progress status of counseling using a bar graph, a pie chart, a checklist, or a timeline format indicating the progress rate. At this time, the status of each stage is distinguished by color or icon, and can be easily distinguished through visual differences, such as displaying completed stages in green, stages in progress in blue, and waiting stages in gray.

[0515] Final result delivery step (S560)

[0516] Organized data provision step (S560-1)

[0517] According to one embodiment of the present invention, in the organized data provision step (S560-1) of the final result delivery step (S560), the consultation interface tool (135, 250) can organize data collected and supplemented during the consultation process. Specifically, the consultation interface tool (135, 250) can analyze all data accumulated during the conversation process and reorganize it according to importance and context. For example, in the case of a traffic accident damage compensation lawsuit, the consultation interface tool (135, 250) can organize and provide important information such as the date, location, amount of damage, and related evidence by item. In addition, the organized data can be provided in a visual and systematic format so that the user can easily understand and utilize it. Specifically, the consultation interface tool (135, 250) can visually represent the data using various formats such as tables, graphs, and diagrams, and can provide it along with a text summary. For example, the consultation interface tool (135, 250) can provide the flow of the accident occurrence visualized in a timeline form along with a table summarizing the incident information.

[0518] Consultation Summary Saving Step (S560-2)

[0519] According to one embodiment of the present invention, in the consultation summary storage step (S560-2) of the final result delivery step (S560), the consultation interface tool (135, 250) can store summary data generated during the consultation process. Specifically, the consultation interface tool (135, 250) can organize key information extracted during the consultation and results discussed, and manage the stored content through data structuring. For example, in the case of a traffic accident damage compensation lawsuit, the consultation interface tool (135, 250) can organize and store the date of the accident, information on the person responsible, amount of damages, list of necessary documents, etc. Additionally, the consultation interface tool (135, 250) can store data in various formats such as text files, PDFs, and spreadsheets.

[0520] FIG. 7 is a diagram illustrating the data storage and provision steps of a data storage and management tool according to an embodiment of the present invention.

[0521] Data storage step (S600)

[0522] According to one embodiment of the present invention, a data storage and management tool (136, 260) can store data generated during a consultation process in a structured form. Specifically, the data storage and management tool (136, 260) can classify important information such as dates, times, and event details extracted from conversations, as well as conversation records, and organize them by field. In addition, the data storage and management tool (136, 260) enhances security functions by considering the sensitivity of the data. It can protect the stored data by applying encryption technology and perform access control to prevent unauthorized users from accessing it.

[0523] Data management step (S610)

[0524] According to one embodiment of the present invention, a data storage and management tool (136, 260) can systematically classify and organize stored data according to purpose and type. For example, consultation data can be categorized and managed according to criteria such as case type (e.g., compensation for damages, contract disputes), date, importance, and person in charge. Additionally, the data storage and management tool (136, 260) can increase the efficiency of the database by detecting and removing duplicate data so that identical or similar data is not stored. For example, if a client enters the same information multiple times, the data storage and management tool (136, 260) can consolidate and store it, and remove unnecessary data. Furthermore, the data storage and management tool (136, 260) can process changes or updates to existing data and record change history to ensure the traceability of the data.

[0525] Data retrieval and provision step (S620)

[0526] According to one embodiment of the present invention, a data storage and management tool (136, 260) may provide a search function to search for stored data. For example, the data storage and management tool (136, 260) may utilize various options such as keyword-based search, date filters, event type filters, and search by data classification. Additionally, the data storage and management tool (136, 260) may provide an intuitive and easy-to-use search interface. Furthermore, the data storage and management tool (136, 260) may recommend additional information related to the searched data. For example, when searching for "traffic accident in May 2023," the data storage and management tool (136, 260) may present related evidence or records of similar incidents.

[0527] Data security step (S630)

[0528] According to one embodiment of the present invention, a data storage and management tool (136, 260) can protect data by applying encryption technology when data is stored or transmitted. Specifically, when stored, data is stored on a disk ...

Claims

Claim 1 A client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; and a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information. Claim 2 In claim 1, the conversation collection tool is configured to convert the collected voice information into text information using voice recognition technology, a client conversation collection and analysis system for consultation. Claim 3 A client conversation collection and analysis system for consultation, wherein, in claim 1 or 2, the conversation collection tool is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data with respect to the collected text information or the converted text information. Claim 4 In claim 1, the conversation collection tool is configured to perform an analysis of the client's emotional state with respect to the collected voice information or the collected text information, a client conversation collection and analysis system for counseling. Claim 5 A client conversation collection and analysis system for consultation, wherein, in claim 1, the system further comprises an advanced factual relationship analysis and supplementary tool, and the conversation collection tool can perform a determination on whether there is missing information regarding the collected voice information or the collected text information, and if it is determined by the judgment performed that there is missing information, the system is configured to transmit the determined missing information to the advanced factual relationship analysis and supplementary tool in order to cause the tool to generate additional questions for the client. Claim 6 In claim 1, the natural language processing and data extraction tool is configured to perform syntactic analysis on the collected voice information or text information, and is configured to extract factual relationship information by the performed syntactic analysis, a client conversation collection and analysis system for consultation. Claim 7 A client conversation collection and analysis system for consultation, wherein the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, and event information using Named Entity Recognition (NER) technology on the collected voice information or text information, and to extract factual relationship information based on the extracted entity. Claim 8 A client conversation collection and analysis system for consultation, wherein the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, and, if there is one or more extracted entities, to identify the relationship between the one or more entities and to extract factual relationship information based on the identified relationship. Claim 9 A client conversation collection and analysis system for consultation, wherein the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, compare the extracted entity with a pre-trained legal pattern database, and identify missing information based on the comparison. Claim 10 In claim 1, the natural language processing and data extraction tool is configured to perform an analysis to determine context by associating currently collected voice information or text information with voice information or text information previously provided by the client, a client conversation collection and analysis system for consultation. Claim 11 In claim 1, the natural language processing and data extraction tool is configured to learn the structural patterns of the collected text information by utilizing GPT (Generative Pre-trained Transformer) technology, a client conversation collection and analysis system for consultation. Claim 12 In claim 1, the natural language processing and data extraction tool is configured to understand the context of the collected text information bidirectionally by utilizing BERT (Bidirectional Encoder Representations from Transformers) technology, a client conversation collection and analysis system for consultation. Claim 13 A client conversation collection and analysis system for consultation, wherein the natural language processing and data extraction tool is configured to extract at least one entity among subject information, date information, place information, weather information, and event information from the collected voice information or text information, and to store the relationships between the extracted entities by representing them as nodes and edges using graph database technology. Claim 14 A client conversation collection and analysis system for consultation, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and a pre-trained legal pattern database tool for generating additional questions to supplement missing or incomplete information based on said extracted factual relationship information. Claim 15 In claim 14, the pre-learned legal pattern database tool is configured to learn and store at least one of the following data: case law data by litigation type, judgment result statistical data, standard data by damages item, damage amount adjustment data by negligence ratio, negligence data by legal type, evidence standard data by damages item, litigation procedure and evidence submission data, legal time and deadline data, probability of winning data by case type, medical dispute data, real estate litigation data, labor law related data, case law-based damages standard data, negligence compensation data by litigation type, legal negligence and compensation mapping data, legal accident analysis and compensation data, and case law-based litigation outcome prediction data, a client conversation collection and analysis system for consultation. Claim 16 In claim 14, the pre-learned legal pattern database tool is configured to perform automation of the client's condition recording based on the extracted factual information when the client's request is in the field of medical consultation, a client conversation collection and analysis system for consultation. Claim 17 In claim 16, the pre-learned legal pattern database tool stores at least one of the following: diagnostic result information related to the specific symptom according to the specific symptom information, treatment information related to the specific symptom, and drug information related to the specific symptom in advance; and is configured to integrate and store at least one of the following: the client's specific symptom information, diagnostic result information related to the client's specific symptom, treatment information related to the client's specific symptom, and drug information related to the client's specific symptom, based on the stored information, a client conversation collection and analysis system for consultation. Claim 18 In claim 14, the pre-learned legal pattern database tool is configured to search for cases similar to the collected voice information or text information based on the extracted factual information, a client conversation collection and analysis system for consultation. Claim 19 In claim 18, the pre-learned legal pattern database tool is configured to compare the similarity between the extracted factual information and the searched similar cases using similarity calculations, a client conversation collection and analysis system for consultation. Claim 20 In claim 19, the pre-trained legal pattern database tool is configured to perform the similarity calculation by utilizing at least one of a vector-based similarity calculation technique, a text-based similarity calculation technique, an image data similarity technique, a deep learning-based similarity technique, a Cosine Similarity technique, and a BERT-based sentence embedding technique, a client conversation collection and analysis system for consultation. Claim 21 In claim 14, the pre-learned legal pattern database tool is configured to identify missing information by comparing the extracted factual information with pre-stored essential information by litigation type, a client conversation collection and analysis system for consultation. Claim 22 In claim 14, the conversation collection tool is a client conversation collection and analysis system for consultation configured to perform conversion of the collected voice information into text information using voice recognition technology. Claim 23 A client conversation collection and analysis system for consultation, wherein, in claim 14 or 22, the conversation collection tool is configured to perform at least one of removing unnecessary symbols, removing unnecessary spaces, and removing duplicate data with respect to the collected text information or the converted text information. Claim 24 In claim 14, the conversation collection tool is configured to perform an analysis of the client's emotional state with respect to the collected voice information or the collected text information, a client conversation collection and analysis system for counseling. Claim 25 In claim 15, the pre-learned legal pattern database tool is configured to infer additional information from the extracted factual information based on the learned and stored information, a client conversation collection and analysis system for consultation. Claim 26 A client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and an advanced factual relationship analysis and supplementation tool for performing factual relationship analysis based on said extracted factual relationship information. Claim 27 In claim 26, the advanced factual relationship analysis and supplementary tool is configured to pre-learn a legal pattern database and is configured to identify missing information by comparing the pre-learned legal pattern database with the extracted factual relationship information, a client conversation collection and analysis system for consultation. Claim 28 In claim 27, the advanced factual analysis and supplementary tool is a client conversation collection and analysis system for consultation configured to generate additional questions based on the identified missing information and provide them to the client. Claim 29 In claim 28, the advanced factual relationship analysis and supplementary tool is configured to analyze the emotional state of the client based on the collected voice information or text information, and, when generating the additional questions, to adjust the expression method of the additional questions based on the analyzed emotional state, a client conversation collection and analysis system for counseling. Claim 30 In claim 28, the advanced factual relationship analysis and supplementary tool is configured to analyze the client's understanding of the content of the consultation based on the collected voice information or text information, and, when generating the additional questions, to adjust the difficulty of the additional questions based on the analyzed understanding, a client conversation collection and analysis system for consultation. Claim 31 In claim 28, the advanced factual analysis and supplementary tool is configured to analyze the level of speech of the client based on the collected voice information or text information, and, when generating the additional questions, to adjust the difficulty of the additional questions based on the analyzed understanding, a client conversation collection and analysis system for counseling. Claim 32 In claim 28, the advanced factual analysis and supplementary tool is configured to determine the priority of the generated additional questions when generating the additional questions, and is configured to dynamically adjust the determined priority when new information is provided during the consultation with the client, a client conversation collection and analysis system for consultation. Claim 33 In claim 28, the above-described advanced factual analysis and supplementary tool is a client conversation collection and analysis system for consultation configured to generate multi-stage questions to collect necessary information step by step when generating the above-described additional questions. Claim 34 A client conversation collection and analysis system for counseling, comprising: a conversation collection tool for collecting voice information or text information from said client; a natural language processing and data extraction tool for analyzing said collected voice information or text information using a natural language processing-based artificial intelligence model and extracting factual relationship information; and a counseling interface tool for delivering additional questions or answers based on said extracted factual relationship information. Claim 35 In claim 34, the counseling interface tool is configured to provide the additional questions or answers in voice or text, a client conversation collection and analysis system for counseling. Claim 36 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; and a step of extracting factual relationship information based on the separated sentences. Claim 37 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises a filtering step for removing noise or noise from the collected voice data. Claim 38 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: a step of distinguishing fixed noise and non-fixed noise in the collected voice data; a step of removing the fixed noise by utilizing at least one filtering technique among a frequency domain filtering technique, a spectrum attenuation technique, and a linear predictive coding technique for the fixed noise; and a step of removing the non-fixed noise by utilizing at least one filtering technique among an adaptive filtering technique, a waveform-based noise removal technique, and a masking-based noise removal technique for the non-fixed noise. Claim 39 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises a restoration step for supplementing voice data damaged in the filtering step. Claim 40 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises the step of removing noise or noise from the collected voice data by utilizing at least one model among Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) models, Generative Adversarial Networks (GANs), Transformer-based models, and Microsoft’s Deep Noise Suppression (DNS) models. Claim 41 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: an abnormal input detection step for detecting abnormal inputs in the collected voice data or text information; and a step for removing or modifying the detected abnormal inputs. Claim 42 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein, in claim 41, the abnormal input detection step; further comprises a step performed by at least one model among a Recurrent Neural Networks (RNNs) model, a Long Short-Term Memory (LSTM) model, a Transformer-based model, a Noise Detection model, a Signal Classification model, an Anomaly Detection model, and a Named Entity Recognition (NER) model. Claim 43 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of correcting duplicate spaces, correcting incorrect spacing, unifying case, and converting dates, times, and amounts within the text into a consistent format for the converted text information and the collected text information. Claim 44 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of deleting unnecessary elements, deleting duplicate words, deleting meaningless words, removing incomplete sentences, deleting symbols that interfere with analysis, adjusting punctuation, and adjusting spacing on the converted text information and the collected text information. Claim 45 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of performing at least one of converting date data within the text into an ISO standard, unifying the time within the text into a 24-hour system, converting numbers within the text into numerical forms if they are expressed as characters, and converting units within the text into a standardized form for the converted text information and the collected text information. Claim 46 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the preprocessing step further comprises: a step of converting the collected voice data into text information; and a step of converting various expressions having the same meaning within the text into a single unified term for the converted text information and the collected text information. Claim 47 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the sentence tokenization step further comprises: a step of detecting punctuation including at least one of a period, a question mark, and an exclamation mark within the text of the generated standardized text information; a sentence end identification step of identifying the end of a sentence based on the detected punctuation; and a step of dividing the identified end of a sentence to separate the text into sentence units. Claim 48 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the sentence end identification step further comprises a sentence boundary ambiguity resolution step that processes not to identify the end of a sentence based on the detected punctuation when the detected punctuation is not a punctuation that signifies the end of a sentence. Claim 49 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the sentence tokenization step further comprises: a step of identifying the end of a sentence based on the contextual meaning or conjunction of the text of the generated standardized text information; and a step of dividing the identified end of a sentence to separate the text into sentence units. Claim 50 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a part-of-speech tagging step of identifying the part of speech of each word in the generated standardized text information and assigning tags; and a step of extracting factual relationship information based on the tagged parts of speech. Claim 51 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the part-of-speech tagging step further comprises: a step of separating the generated standardized text information into words; and a step of determining which part of speech the separated words belong to among nouns, verbs, adjectives, adverbs, and prepositions. Claim 52 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the part-of-speech tagging step further comprises the step of identifying and tagging the part of speech of each word by utilizing at least one model among a BiLSTM-CRF model, a Transformer-based model, a word vector model, and a Pre-trained Embedding model. Claim 53 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, comprising: a step of collecting voice data or text information of the client; a preprocessing step of preprocessing the collected voice data or text information to generate standardized text information; a sentence tokenization step of separating the generated standardized text information into sentences; a factual relationship information extraction step of extracting factual relationship information based on the separated sentences; and a feedback provision step of providing feedback to the client based on the extracted factual relationship information. Claim 54 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the step of extracting factual information further comprises the step of identifying missing information by comparing the collected voice data or text information with an essential information template by litigation type stored in a database. Claim 55 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the feedback providing step further comprises: a step of generating additional questions related to the identified missing information when missing information is identified in the factual information extraction step; and a step of providing the generated additional questions. Claim 56 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the feedback providing step further comprises: a step of generating customized questions suitable for the client based on the extracted factual information; and a step of providing additional questions generated. Claim 57 A method for collecting and analyzing client conversations for counseling using a natural language processing-based artificial intelligence model, wherein the feedback providing step further comprises: a step of analyzing the emotional state of the client based on the collected voice data or text information of the client; a step of generating additional questions based on the analyzed emotional state of the client; and a step of providing the generated additional questions. Claim 58 A method for collecting and analyzing client conversations for counseling using a natural language processing-based artificial intelligence model, wherein the feedback providing step further comprises: a step of analyzing the client's understanding of the content of the counseling based on the collected client's voice data or text information; a step of generating additional questions based on the analyzed client's understanding of the content of the counseling; and a step of providing the generated additional questions. Claim 59 A method for collecting and analyzing client conversations for consultation using a natural language processing-based artificial intelligence model, wherein the feedback providing step further comprises: a step of generating additional questions based on previous questions or previous answers by utilizing natural language processing (NLP) technology to reflect the flow and context of the conversation of the collected client's voice data or text information; and a step of providing the generated additional questions. Claim 60 A method for providing an answer to a client's question using a consultation interface tool, comprising: a step of supporting the input of answer data for collecting voice information or text information of the client; a step of converting the input voice information or text information into a structured format; a preprocessing step of performing preprocessing on the converted voice information or text information; and a response providing step of providing response information regarding the preprocessed voice information or text information using a natural language processing-based artificial intelligence model. Claim 61 A method for answering a client's question using a consultation interface tool, wherein the answer data input support step further comprises the step of supporting input using a user interface so that the client can perform at least one of voice input, text input, and a selectable response. Claim 62 A method for answering a client's question using a consultation interface tool, wherein, in claim 61, the answer data input support step further comprises the step of providing at least one of a text auto-completion function, a keyword recommendation function, a dynamic question generation function, a real-time spelling correction function, and a real-time grammar correction function when the client performs the text input method. Claim 63 A method for performing a response to a client's question using a consultation interface tool, wherein the response providing step further comprises the step of dynamically adjusting the speed of the response providing based on the client's response pattern or the client's response speed by utilizing the natural language processing-based artificial intelligence model in claim 60. Claim 64 A method for answering a client's question using a consultation interface tool, wherein, in claim 60, the response providing step further comprises the step of providing a response that presents specific options to the client when the preprocessed voice information or text information is insufficient by utilizing the natural language processing-based artificial intelligence model. Claim 65 A method for answering a client's question using a consultation interface tool, wherein, in claim 60, the response providing step further comprises the step of providing additional questions to the client regarding the ambiguous or uncertain information when the preprocessed voice information or text information is ambiguous or uncertain by utilizing the natural language processing-based artificial intelligence model. Claim 66 In claim 60, the response providing step further comprises the step of providing an additional question to the client regarding the conflicting information when the preprocessed voice information or text information conflicts with the previously collected voice information or text information of the client using the natural language processing-based artificial intelligence model; a method for performing a response to a client's question using a consultation interface tool. Claim 67 A method for answering a client's question using a consultation interface tool, further comprising: a step of generating a pair of the collected voice information or text information and the provided response information in claim 60; and a step of storing the generated pair.

Images