Semantic association analysis model generation method and semantic association analysis method

Abstract

The invention discloses a semantic association analysis model generation method and a semantic association analysis method. The semantic association analysis method comprises the following steps: acquiring historical target data, and preprocessing the historical target data; establishing a comprehensive text semantic analysis model based on a deep neural network; training a comprehensive text semantic analysis model according to the preprocessed historical target data; inputting real-time target data, preprocessing the real-time target data, and inputting the preprocessed real-time target data into the trained comprehensive text semantic analysis model to obtain an output result; and performing semantic association analysis on an output result. Precise prediction of various types of target data is realized through the comprehensive semantic prediction model, the model performance is continuously optimized through a dynamic updating mechanism, and the accuracy and timeliness of a prediction result are ensured. The unique semantic association analysis module can deeply mine internal relationships among different types of data such as texts, pictures, audios and videos, and more comprehensive and deep insight is provided for users.

Description

Technical Field

[0001] The present invention relates to the technical field of text semantic association analysis, and in particular to a method for generating a semantic association analysis model and a semantic association analysis method. Background Art

[0002] With the rapid development of science and technology and the comprehensive digital transformation of society, the status and role of electronic data in case investigation have become indispensable. However, the rapid advancement of information technology has also brought about an extremely complex data environment, resulting in an explosive growth in the amount of data involved in cases. In this process, not only has the amount of data increased dramatically, but the types and formats of data have also become more diverse. Among them, a large amount of unstructured data such as audio and video, pictures, documents, and data from social media, mobile devices, and cloud services have brought unprecedented challenges to the traditional case investigation data governance and research and analysis work.

[0003] Faced with such a complex data environment, traditional analysis methods seem to be unable to cope with it. Traditional analysis methods usually rely on manual screening, processing and analysis of data, which is not only labor-intensive and inefficient, but also easily restricted by human factors, resulting in incomplete and inaccurate analysis results. At the same time, the process of manually combing data is often only applicable to structured data, such as tables and records, and the processing ability for unstructured data such as audio, video and pictures is very limited. In addition, faced with new types of complex scenario cases, traditional analysis methods lack pertinence and flexibility, and cannot quickly adapt to and respond to changing data and information, resulting in a time-consuming and slow processing process. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method for generating a semantic association analysis model and a semantic association analysis method, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for generating a semantic association analysis model, wherein the semantic association analysis model comprises: a data acquisition module, a data preprocessing module, a comprehensive semantic prediction model module, a comprehensive semantic analysis module, and a visualization module; The data acquisition module is used to acquire first target data and transmit the first target data to the data preprocessing module; The data preprocessing module is used to perform a first classification on the first target data transmitted by the data acquisition module to obtain second target data, and perform a first preprocessing on the second target data to obtain third target data, and transmit the third target data to the comprehensive semantic prediction model module; The comprehensive semantic prediction model module is used to perform comprehensive semantic prediction on the third target data transmitted by the data preprocessing module through the comprehensive semantic prediction model obtained based on deep neural network pre-training, and transmit the prediction result of the comprehensive semantic prediction model to the comprehensive semantic analysis module; The comprehensive semantic analysis module is used to analyze the prediction results transmitted by the comprehensive semantic prediction model module, and transmit the prediction results after the analysis to the visualization module; The visualization module is used to perform visualization processing according to the prediction results after the analysis transmitted by the comprehensive semantic analysis module.

[0008] As a preferred solution of the method for generating the semantic association analysis model described in the present invention, the comprehensive semantic prediction model module includes: a data processing and input unit, a comprehensive semantic prediction model unit, a comprehensive semantic prediction model update unit and a data storage unit; The data processing and input unit is used to perform a second preprocessing on the third target data transmitted by the data preprocessing module to obtain fourth target data, and perform a second classification on the fourth target data to obtain fifth target data, and transmit the fifth target data to the comprehensive semantic prediction model unit and the data storage unit; The comprehensive semantic prediction model unit performs model prediction according to the fifth target data transmitted to it, obtains prediction results of different models, and transmits the prediction results to the comprehensive semantic analysis module and the data storage unit; The data storage unit is used to store the fifth target data generated by the data processing and input unit and the prediction results generated by the comprehensive semantic prediction model unit for the fifth target data.

[0009] As a preferred solution of the method for generating the semantic association analysis model of the present invention, the comprehensive semantic prediction model updating unit includes: The comprehensive semantic prediction model updating unit presets an update cycle, and when the update cycle is reached and the model needs to be updated, the first correct data stored in the data storage unit during the cycle is retrieved to update the model, and the updated model is transmitted to the comprehensive semantic prediction model unit to replace the original model; The first correct data is the fifth target data generated by the data processing and input unit whose prediction result within the period is accurate, and the prediction result generated by the comprehensive semantic prediction model unit for the fifth target data.

[0010] As a preferred solution of the method for generating the semantic association analysis model of the present invention, wherein: the comprehensive semantic prediction model unit includes: a first text prediction model part, a second text prediction model part and a third text prediction model part; The first text prediction model unit is used to predict the first text data in the fifth target data, the second text prediction model unit is used to predict the second text data in the fifth target data, and the third text prediction model unit is used to predict the third text data in the fifth target data.

[0011] As a preferred solution of the method for generating the semantic association analysis model of the present invention, wherein: the comprehensive semantic analysis module includes: a first analysis unit, a second analysis unit, a third analysis unit, and a fourth analysis unit; The first analysis unit is used to receive the prediction result transmitted by the first text prediction model unit, and perform a first text analysis on the prediction result; The second analysis unit is used to receive the prediction result transmitted by the second text prediction model unit, and perform a second text analysis on the prediction result; The third analysis unit is used to receive the prediction result transmitted by the third text prediction model unit, and perform a third text analysis on the prediction result; The fourth analysis unit is used to perform a first correlation analysis on the first text analysis result, the second text analysis result and the third text analysis result, and transmit the first correlation analysis to the visualization module for visualization processing.

[0012] As a preferred solution of the method for generating the semantic association analysis model of the present invention, the data preprocessing module includes: a first classification unit, a first preprocessing unit, a first data output unit and a first preprocessing adjustment unit; The first classification unit is used to perform a first classification on the first target data according to a preset classification rule; The first preprocessing unit is used to perform a first preprocessing on the second target data after the first classification, and generate third target data; The first data output unit transmits the preprocessed third target data to the comprehensive semantic prediction model module; The first preprocessing adjustment unit is used to obtain the first error data stored in the data storage unit in the comprehensive semantic prediction model module, and adjust the preprocessing parameters of the first preprocessing unit according to the first error data; The first error data is the fifth target data generated by the data processing and input unit whose prediction result within the period is erroneous and the prediction result generated by the comprehensive semantic prediction model unit for the fifth target data.

[0013] In a second aspect, the present invention provides a semantic association analysis method for generating a semantic association analysis model, comprising: Acquiring historical target data, and preprocessing the historical target data; Establish a comprehensive text semantic analysis model based on deep neural network; Training the comprehensive text semantic analysis model based on the preprocessed historical target data; Input real-time target data, pre-process the real-time target data, and input the pre-processed real-time target data into the trained comprehensive text semantic analysis model to obtain output results; Perform semantic association analysis on the output results.

[0014] As a preferred embodiment of the semantic association analysis method of the method for generating the semantic association analysis model described in the present invention, the comprehensive text semantic analysis model includes: a first text prediction model, a second text prediction model and a third text prediction model; The first text prediction model is used to predict text-type target data; The second text prediction model is used to predict the target data which is a picture type; The third text prediction model is used to predict the target data which is audio or video type data.

[0015] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method described above when executed by a processor.

[0017] Compared with the prior art, the present invention has the following beneficial effects: the present invention proposes a method for generating a semantic association analysis model and a semantic association analysis method, obtains historical target data, and preprocesses the historical target data; establishes a comprehensive text semantic analysis model based on a deep neural network; trains the comprehensive text semantic analysis model according to the preprocessed historical target data; inputs real-time target data, and preprocesses the real-time target data, and at the same time inputs the preprocessed real-time target data into the comprehensive text semantic analysis model after training to obtain an output result; and performs semantic association analysis on the output result. The semantic association analysis model and method not only realizes accurate prediction of various types of target data through a comprehensive semantic prediction model, but also continuously optimizes the model performance through a dynamic update mechanism to ensure the accuracy and timeliness of the prediction results. In addition, its unique semantic association analysis module can deeply explore the intrinsic connections between different types of data such as text, pictures, audio and video, and provide users with more comprehensive and in-depth insights. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them: Figure 1 A schematic diagram of a system structure of a method for generating a semantic association analysis model and a semantic association analysis method provided by an embodiment of the present invention; Figure 2 A method flow chart of a method for generating a semantic association analysis model and a semantic association analysis method provided by one embodiment of the present invention; Figure 3 An internal structure diagram of a computer device for a method for generating a semantic association analysis model and a semantic association analysis method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0020] Example 1 Reference Figure 1-Figure 3, which is the first embodiment of the present invention, and provides a method for generating a semantic association analysis model and a semantic association analysis method, comprising: Before describing the embodiments of the present application in detail, some related concepts are first explained for the sake of clarity.

[0021] Semantic association analysis: This is a way of processing information that focuses on identifying, understanding, and extracting meaningful relationships between different data sets. These relationships may exist in multiple types of data, such as text, images, audio and video. Through semantic association analysis, useful information can be extracted from unstructured data and the inherent connections between them can be revealed.

[0022] Deep neural network: This is a subfield of machine learning that builds multi-layer artificial neural networks by imitating the working principle of human brain neurons. Deep neural networks handle complex learning tasks through multi-level abstract representations. They are applicable to a variety of fields such as image recognition, speech recognition, and natural language processing, and have strong pattern recognition and self-learning capabilities.

[0023] Association analysis: It is a data mining technique that aims to find interesting associations or frequently occurring patterns between variables in a large data set. For example, market basket analysis is an application of association analysis, which helps retailers understand the relationship between customer purchasing behavior, that is, after customers buy a product, what other products do they often buy.

[0024] Long-context learning: This term usually refers to the ability of an algorithm to take into account information in longer time spans or text paragraphs when processing sequential data (such as text or time series). In natural language processing, long-context learning means that the model has the ability to understand information that spans multiple sentences or paragraphs, which is very important for tasks that require contextual understanding (such as question answering systems and text summarization).

[0025] Natural Language Processing: It is a branch of artificial intelligence that focuses on enabling computers to understand, interpret, and generate human language. NLP combines linguistic knowledge, computer science, and mathematical algorithms to enable machines to process and understand large amounts of natural language data, and then perform tasks such as automatic translation, sentiment analysis, and question-answering systems.

[0026] In related technologies, with the deep popularization and widespread application of the Internet, Internet-related criminal cases are showing a rapid growth trend. Such cases usually have the following characteristics: large case scale, many people involved, and various types of inspection materials involved. These characteristics lead to a sharp increase in the amount and complexity of data when processing and analyzing the data involved. Investigators need to face massive amounts of social media content, communication data, etc., even tens of millions of levels. Traditional processing methods mainly rely on manual review, comparison and analysis, which is not only time-consuming and labor-intensive, but also difficult to guarantee accuracy and comprehensiveness. With the increase in the amount and complexity of the data involved, traditional methods can no longer meet the needs of modern case investigation.

[0027] The present application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to generate the semantic association analysis model. Figure 1 A method for generating a semantic association analysis model and a system structure diagram of the semantic association analysis method are shown, including: a data acquisition module 100, a data preprocessing module 200, a comprehensive semantic prediction model module 300, a comprehensive semantic analysis module 400 and a visualization module 500; In the embodiment of the present application, the data acquisition module 100 is used to acquire the first target data and transmit the first target data to the data preprocessing module 200; The data preprocessing module 200 is used to perform a first classification on the first target data transmitted by the data acquisition module 100 to obtain second target data, and perform a first preprocessing on the second target data to obtain third target data, and transmit the third target data to the comprehensive semantic prediction model module 300; The comprehensive semantic prediction model module 300 is used to perform comprehensive semantic prediction on the third target data transmitted by the data preprocessing module 200 through the comprehensive semantic prediction model obtained based on deep neural network pre-training, and transmit the prediction result of the comprehensive semantic prediction model to the comprehensive semantic analysis module 400; The comprehensive semantic analysis module 400 is used to analyze the prediction results transmitted by the comprehensive semantic prediction model module 300, and transmit the prediction results after the analysis to the visualization module 500; The visualization module 500 is used to perform visualization processing according to the prediction results after the analysis transmitted by the comprehensive semantic analysis module 400.

[0028] It should be noted that if it is necessary to generate a semantic association analysis model of the type required by the customer, it is only necessary to introduce the corresponding type of data source (i.e., the first target data) in the data acquisition module 100, and change the input and output of the comprehensive semantic prediction model obtained based on the deep neural network pre-training in the comprehensive semantic prediction model module 300. For example, when a semantic association analysis model of a case investigation type is required, data sources related to case investigation can be introduced, such as social media content, communication records, and information about persons involved in the case. At the same time, the comprehensive semantic prediction model in the comprehensive semantic prediction model module 300 is adjusted so that it can perform effective semantic prediction and analysis on these specific types of data. For example, the input of the comprehensive semantic prediction model module 300 is changed to text data related to case investigation, such as the suspect's communication records, social media content, etc., and the output of the model is adjusted to the semantic association analysis result of the content involved in the case. Such an adjustment can make the model more focused on the specific needs of the case investigation field and improve the accuracy and efficiency of the analysis.

[0029] In an optional embodiment, a flexible data source interface can be introduced in the data collection module 100 to support the capture of information from a variety of data sources, including but not limited to social media platforms, instant messaging applications, email systems, and public security databases, etc. These data sources cover a wide range of information that may be involved in the case investigation process, ensuring the comprehensiveness and timeliness of the data.

[0030] In the embodiment of the present application, the data preprocessing module 200 includes: a first classification unit 201, a first preprocessing unit 202, a first data output unit 203 and a first preprocessing adjustment unit 204; The first classification unit 201 is used to perform a first classification on the first target data according to a preset classification rule; The first preprocessing unit 202 is used to perform a first preprocessing on the second target data after the first classification, and generate third target data; The first data output unit 203 transmits the preprocessed third target data to the comprehensive semantic prediction model module 300; The first preprocessing adjustment unit 204 is used to obtain the first error data stored in the data storage unit 304 in the comprehensive semantic prediction model module 300, and adjust the preprocessing parameters of the first preprocessing unit 202 according to the first error data; The first error data is the fifth target data generated by the data processing and input unit 301 whose prediction result within the period is an error, and the prediction result generated by the comprehensive semantic prediction model unit 302 for the fifth target data.

[0031] It should be noted that the first classification is to classify the first target data into three types: text data, picture data, and audio and video data.

[0032] In an optional embodiment, the first preprocessing includes, but is not limited to, processing the text data such as removing stop words, word segmentation, part-of-speech tagging, named entity recognition, etc., to improve the accuracy of subsequent semantic prediction; processing the picture data such as image recognition, feature extraction, etc., so as to convert the key information in the image into text information that can be processed by the model; processing the audio-visual data such as speech recognition, transcription, keyword extraction, etc., so as to extract useful text information from the audio-visual.

[0033] In the embodiment of the present application, when a semantic association analysis model of a case investigation type needs to be generated, the processing process of the first preprocessing for the text data is as follows: First, remove the common words in the text, such as "de", "le", "shi", which have little impact on understanding the meaning of the text. Second, split the continuous text string into individual words or phrases for subsequent processing. Third, perform part-of-speech tagging on each word, such as noun (N), verb (V), adjective (J), etc. Finally, identify the proper nouns in the text, such as person names, place names, organization names, etc., and classify them.

[0034] Exemplarily, assume there is a text: "Zhang San was arrested in the bank theft case in City M." First, remove "de", "zai", "zhong", "bei".

[0035] Second, cut the text into "Zhang San", "City M", "bank", "theft case", "arrested".

[0036] Third, they are respectively "person name", "place name", "noun", "event", "verb".

[0037] Finally, identify that "Zhang San" is a person name and "City M" is a place name.

[0038] In the embodiment of the present application, when a semantic association analysis model of a case investigation type needs to be generated, the processing process of the first preprocessing for the image data is as follows: First, use computer vision technology to identify the objects or scenes in the picture. Second, extract the key features from the picture, such as edges, color distribution, texture, etc. Finally, convert the information in the image into a text description for subsequent processing.

[0039] Exemplarily, assume there is a picture containing a car and a license plate number.

[0040] First, identify "car" and "license plate number".

[0041] Second, extract features such as the color and model of the car.

[0042] Finally, describe the image as "a red BMW car with the license plate number XX".

[0043] In the embodiment of the present application, when a semantic association analysis model of a case investigation type needs to be generated, the first preprocessing process for the audio and video data is: first, convert the voice in the audio into text. Secondly, convert the recording content into written form. Finally, extract key information from the transcribed text.

[0044] For example, suppose there is a recording: "The suspect admitted that he was at the scene around 10 o'clock on the evening of the crime." Speech Recognition: Convert audio recordings to text.

[0045] Transcribe: Get text "The suspect admitted that he was at the scene around 10 pm on the day of the crime." Keyword extraction: extract "suspect", "around 10 o'clock on the evening of the incident", and "appeared at the scene".

[0046] It should be noted that the above-mentioned first preprocessing can perfectly preprocess three different types of data, namely text data, image data and audio and video data, and provides a solid foundation for subsequent semantic prediction analysis.

[0047] It should also be noted that the first error data is that the prediction result within the period is wrong. Whether the prediction result is correct or wrong can be judged manually or by a preset threshold. When making a manual judgment, it is only necessary to manually randomly check whether the generated prediction result is satisfactory. If there is an unsatisfactory part, the model prediction result is directly marked as wrong or inaccurate. Although manual judgment can be used at this position, the time and energy required for manual judgment are less, and compared with the traditional human comparison process, it is even close to zero, which will not lead to excessive manpower consumption. The preset threshold judgment is to set a threshold based on the deviation between the model's prediction result and the actual result. When the deviation exceeds this threshold, the model's prediction result is considered to be wrong. This method is relatively objective, but a suitable threshold needs to be set in advance to ensure the accuracy of the judgment.

[0048] It should be noted that the generation method of the above model and the operation process of the system are highly modular and flexible. This design not only facilitates the maintenance and upgrade of the system, but also enables the system to be quickly adapted according to different application scenarios and needs. For example, when responding to emergencies or analyzing specific industries, the system can be quickly reconstructed to meet new analysis needs by simply adjusting the corresponding module parameters or replacing specific modules.

[0049] In the embodiment of the present application, the comprehensive semantic prediction model module 300 includes: a data processing and input unit 301, a comprehensive semantic prediction model unit 302, a comprehensive semantic prediction model update unit 303 and a data storage unit 304; The data processing and input unit 301 is used to perform a second preprocessing on the third target data transmitted by the data preprocessing module 200 to obtain fourth target data, and perform a second classification on the fourth target data to obtain fifth target data, and transmit the fifth target data to the comprehensive semantic prediction model unit 302 and the data storage unit 304; The comprehensive semantic prediction model unit 302 performs model prediction according to the fifth target data transmitted thereto, obtains prediction results of different models, and transmits the prediction results to the comprehensive semantic analysis module 400 and the data storage unit 304; The data storage unit 304 is used to store the fifth target data generated by the data processing and input unit 301 and the prediction results generated by the comprehensive semantic prediction model unit 302 for the fifth target data.

[0050] In an optional embodiment, the second preprocessing includes but is not limited to operations such as data cleaning, format unification, and standardization processing, aiming to further improve the accuracy and consistency of the data and provide more reliable data input for the comprehensive semantic prediction model.

[0051] In the embodiment of the present application, the second preprocessing is to ensure that the data input to the model can better meet the expectations and requirements of the model, reduce model errors, and thus improve the accuracy of model predictions. Specifically, data cleaning removes missing, abnormal, or duplicate data to ensure the quality of input data; format unification converts all data into a format that the model can recognize to avoid errors caused by inconsistent formats; standardization scales the data according to a certain ratio so that data of different magnitudes can be fairly compared and calculated in the model.

[0052] It should be noted that the second classification is a specific classification operation. For example, for a specific type of case investigation, the second classification can classify the second pre-processed data into multiple subcategories according to the nature of the case, the field involved, or the purpose of the analysis. For example, in economic crime cases, the data can be divided into multiple subcategories such as financial transaction records, communication records, and electronic evidence; in criminal cases, it can be divided into subcategories such as on-site investigation records, suspect confessions, and witness testimonies. This classification not only helps to analyze the case more finely, but also improves the pertinence and accuracy of the model.

[0053] In the embodiment of the present application, when performing the second classification, text classification algorithms in natural language processing (NLP) technology are used, such as naive Bayes, support vector machine (SVM), deep learning model, etc. These algorithms can automatically extract features from text data and assign text to predefined categories based on the features. At the same time, combined with the specific knowledge base and rule set of case investigation, the accuracy and efficiency of classification can be further improved.

[0054] For example, suppose you are dealing with a financial fraud case involving a large number of bank transaction records and communication records. First, these records need to be preprocessed, including data cleaning, format unification, and standardization. In the data cleaning stage, invalid, duplicate, or abnormal transaction records will be eliminated, such as abnormal amounts, timestamp errors, etc. At the same time, communication records will also be screened to remove irrelevant conversations and noise. In the format unification stage, all transaction records and communication records will be converted into a unified format, such as JSON or XML, for subsequent processing. In the standardization stage, numerical data such as transaction amounts and times will be scaled to ensure that they are compared at the same level.

[0055] Next, the second classification is carried out. According to the characteristics of financial fraud cases, transaction records can be divided into two subcategories: normal transactions and suspicious transactions. Normal transactions usually have reasonable amounts, times, and frequencies, while suspicious transactions may show abnormal characteristics, such as large transactions, nighttime transactions, or frequent fund transfers. At the same time, communication records can also be divided into critical communications and non-critical communications. Critical communications may contain clues or evidence related to fraudulent behavior, such as conspiracy conversations between suspects or fraudulent instructions.

[0056] For the second classification, text classification algorithms in NLP technology can be used. First, key features such as transaction amount, time, location, conversation content, etc. are extracted from transaction records and communication records. Then, these features are used to train a classification model that can automatically assign new transaction records and communication records to corresponding subcategories. During the training process, specific knowledge bases and rule sets for financial fraud cases can be combined to improve the accuracy and efficiency of classification.

[0057] Once the second classification is completed, the classified data can be input into the comprehensive semantic prediction model for predictive analysis. The comprehensive semantic prediction model generates prediction results of different models based on the input data and predefined rule sets. These prediction results may include information such as the probability of fraudulent behavior, the identity of the suspect, and the fraud method. These prediction results can then be transferred to the comprehensive semantic analysis module for further analysis and processing to support case detection and investigation.

[0058] In the embodiment of the present application, the comprehensive semantic prediction model updating unit 303 includes: The comprehensive semantic prediction model updating unit 303 presets an update cycle. When the update cycle is reached and the model needs to be updated, the first correct data stored in the data storage unit 304 in the cycle is retrieved to update the model, and the updated model is transmitted to the comprehensive semantic prediction model unit 302 to replace the original model. The first correct data is the fifth target data generated by the data processing and input unit 301 whose prediction result within the period is accurate, and the prediction result generated by the comprehensive semantic prediction model unit 302 for the fifth target data.

[0059] It should be noted that, with the rapid development of science and technology and the comprehensive digital transformation of society, the amount of information data has shown explosive growth, and when it comes to case investigation, criminals' criminal methods are endless, so the model also needs to be updated in real time. Therefore, a comprehensive semantic prediction model update unit 303 is designed to achieve periodic update of the model through a preset update cycle.

[0060] In the embodiment of the present application, the comprehensive semantic prediction model unit 302 includes: a first text prediction model part 302a, a second text prediction model part 302b and a third text prediction model part 302c; The first text prediction model unit 302a is used to predict the first text data in the fifth target data, the second text prediction model unit 302b is used to predict the second text data in the fifth target data, and the third text prediction model unit 302c is used to predict the third text data in the fifth target data.

[0061] It should be noted that the first text data is the corresponding data in the fifth target data obtained through the plain text data in the first target data, that is, the data in the first target data itself is plain text data, and the fifth target data is obtained after the first classification, the first preprocessing, the second preprocessing, and the second classification; It should be noted that the second text data is the corresponding data in the fifth target data obtained through the image data in the first target data, that is, the data in the first target data which is image data itself is the fifth target data obtained after the first classification, the first preprocessing, the second preprocessing and the second classification; It should be noted that the third text data is the corresponding data in the fifth target data obtained through the audio and video data in the first target data, that is, the data in the first target data which is audio and video data itself is the fifth target data obtained after the first classification, the first preprocessing, the second preprocessing, and the second classification; It should also be noted that, because the accuracy of pure text data, image data, and audio and video data after being converted into text information is different, this application designs three text prediction model units to achieve accurate recognition.

[0062] In the embodiment of the present application, the prediction models in the first text prediction model unit 302a, the second text prediction model unit 302b, and the third text prediction model unit 302c are all designed using a deep learning model based on the Transformer architecture, and a deep learning model based on the Transformer architecture, such as GPT (Generative Pre-trained Transformer) or BERT (Bidirectional Encoder Representation Transformation), is used to capture the semantics and contextual relationships of the text. Through extensive training on large-scale case-related data, the model can generate high-quality embedding vectors or directly output prediction results, thereby improving the ability to understand complex language (text). And the designed prediction model can process long sequence texts, thereby understanding the long-distance dependencies between events, solving the problem of information loss that occurs in traditional RNN or LSTM models when processing long texts.

[0063] Exemplarily, the following architecture may be designed for the prediction model in the first text prediction model unit 302a: Input layer: An embedding layer of text data that maps individual words of the fifth target data first text data into vectors of fixed dimension.

[0064] Position encoding: Position encoding is added to preserve the position information of the input sequence.

[0065] Multi-Head Self Attention (MHA): 6 layers with 8 heads per layer to capture long-range dependencies in text.

[0066] Feed-Forward Network (FFN): Each layer contains two fully connected layers, with an activation function (such as ReLU) in between.

[0067] Residual connections and layer normalization: Residual connections are added after MHA and FFN, and layer normalization is added before and after each sub-layer to stabilize the training process.

[0068] It should be noted that the connection method of this architecture is: Step 1: Input text -> Embedding Layer -> Positional Encoding; Step 2: Position encoding -> Multi-head self-attention layer (MHALayer1) -> Residual connection (ResidualConnection) -> Layer normalization (LayerNormalization) -> Feedforward neural network (FFNLayer1) -> Residual connection -> Layer normalization; Step 3: Repeat step 2 6 times; Step 4: The output layer (OutputLayer) is used to generate the corresponding prediction results.

[0069] The following architecture can be designed for the prediction model in the second text prediction model unit 302b: Input layer: an embedding layer of image data, mapping the fifth target data and the second text data into a text representation.

[0070] Multi-head self-attention layer: 4 layers, 8 heads per layer, used to handle image-to-text conversion.

[0071] Feedforward neural network: Each layer contains two fully connected layers, with an activation function (such as ReLU) in the middle.

[0072] Residual connections and layer normalization: Residual connections are added after MHA and FFN, and layer normalization is added before and after each sub-layer to stabilize the training process.

[0073] It should be noted that the connection method of this architecture is: Step 1: Input the processed text data of image data -> embedding layer -> multi-head self-attention layer (MHALayer1) -> residual connection -> layer normalization -> feed-forward neural network (FFNLayer1) -> residual connection -> layer normalization; Step 2: Repeat step 1 4 times; Step 3: The output layer (OutputLayer) is used to generate prediction results.

[0074] The following architecture can be designed for the prediction model in the third text prediction model unit 302c: Input layer: an embedding layer of audio and video data, mapping the fifth target data and the third text data into a text representation.

[0075] Multi-head self-attention layer: 8 layers, 8 heads per layer, used to process audio and video to text conversion.

[0076] Feedforward neural network: Each layer contains two fully connected layers, with an activation function (such as ReLU) in the middle.

[0077] Residual connections and layer normalization: Residual connections are added after MHA and FFN, and layer normalization is added before and after each sub-layer to stabilize the training process.

[0078] It should be noted that the connection method of this architecture is: Step 1: Input the text data after audio and video data processing -> embedding layer -> multi-head self-attention layer (MHALayer1) -> residual connection -> layer normalization -> feedforward neural network (FFNLayer1) -> residual connection -> layer normalization; Step 2: Repeat step 1 8 times; Step 3: The output layer (OutputLayer) is used to generate prediction results.

[0079] It should be noted that the above example is only one possibility of the model framework. Relevant technical personnel can modify or redesign the framework according to specific circumstances or needs. However, no matter which design is used, if the design process of this application is used, it should be within the scope of protection of this application.

[0080] In an embodiment of the present application, the output of the model may include whether a person or an account in the input first target data is involved in the case, whether there is a relationship, or whether there is suspicious behavior, that is, the probability of being involved in the case, the probability of having a relationship, and the probability of having suspicious behavior.

[0081] In the embodiment of the present application, the comprehensive semantic analysis module 400 includes: a first analysis unit 401, a second analysis unit 402, a third analysis unit 403, and a fourth analysis unit 404; The first analysis unit 401 is used to receive the prediction result transmitted by the first text prediction model part 302a, and perform a first text analysis on the prediction result; It should be noted that the first text analysis uses natural language processing (NLP) technology, such as named entity recognition (NER), to specifically identify key people, places, time and other elements in the case; at the same time, sentiment analysis technology is used to evaluate the emotional tendencies in the text to assist in judging the nature of the case or the possible trend of the judgment result (that is, the texts involved in the case are analyzed and extracted according to the output probability of the model).

[0082] In the embodiment of the present application, the second analysis unit 402 is used to receive the prediction result transmitted by the second text prediction model part 302b, and perform a second text analysis on the prediction result; It should be noted that the second text analysis is responsible for processing the image text prediction results output by the second text prediction model unit 302b. Since image data often contains rich visual information and implicit contextual relationships, the second analysis unit needs to pay special attention to the correlation between key scenes, objects and their converted texts in the image when parsing the image text (i.e., analyze and extract all the image data involved in the case according to the output probability of the model).

[0083] In the embodiment of the present application, the third analysis unit 403 is used to receive the prediction result transmitted by the third text prediction model part 302c, and perform a third text analysis on the prediction result; It should be noted that the third text analysis can recognize and understand the voice content and verbal clues in the audio and video (currently this application only considers the processing of the text model, and converts the audio and video or pictures into text information for processing). Since the accuracy of audio and video data after conversion to text data is relatively low, the third text analysis is based on the lowest weight (that is, the audio and video involved in the case are analyzed and extracted according to the output probability of the model).

[0084] In the embodiment of the present application, the fourth analysis unit 404 is used to perform a first correlation analysis on the first text analysis result, the second text analysis result, and the third text analysis result, and transmit the first correlation analysis to the visualization module 500 for visualization processing.

[0085] For example, suppose there is an actual scenario example, a social account posted a content: "The suspect admitted that he was at the scene at around 10 o'clock on the night of the crime." And with a photo and a video, the first analysis unit 401 will first perform detailed natural language processing on this text. Through named entity recognition (NER), the system can accurately identify the key person "suspect" and the key time "around 10 o'clock on the night of the crime", as well as the key location "scene" in the text. This key information is crucial for subsequent case analysis and prediction of judgment results. At the same time, sentiment analysis technology will be used to evaluate the emotional tendency in the text. Although in this scenario, the text content is relatively objective and the emotional tendency may not be obvious, this step still helps to fully understand the background and possible direction of the case.

[0086] Next, the second analysis unit 402 will take over the image text prediction results. Since image data often contains rich visual information and implicit contextual relationships, the second analysis unit needs to carefully analyze the key scenes and objects in the image, as well as their relationship with the text. For example, the image may show the suspect's facial features, clothing, or the layout of the scene, etc. This information can be verified with the text description to enhance the credibility of the case analysis.

[0087] The third analysis unit 403 is responsible for processing the text information converted from the audio and video data. Although the accuracy of the audio and video data converted into text may not be as good as directly processing the text data, the third analysis unit will still try its best to identify and understand the voice content and verbal clues therein. They can all provide valuable references for case analysis. Of course, in terms of weight distribution, the third analysis unit will give a relatively low weight to the audio and video data according to the output probability of the model to ensure the accuracy and fairness of the overall analysis.

[0088] Finally, the fourth analysis unit 404 will conduct a first correlation analysis based on the results of the first, second and third text analyses. In this step, the system will deeply explore the internal connections and potential rules between the various analysis units, such as the consistency between the text description and the image content, the complementarity between the audio and video information and the text description, etc. Through comprehensive correlation analysis, the system can generate a more complete and accurate case analysis report. This report contains the basic facts of the case, key evidence and suspect information, as well as the suspect's current dynamics, location, possible current appearance, dress, etc., and the probability of being determined as the suspect.

[0089] Finally, this case report that integrates multi-source data and multi-dimensional analysis will be transmitted to the visualization module 500 for visualization processing. By displaying the key information and analysis results of the case through intuitive and easy-to-understand charts and images, users can more easily understand and grasp the overall picture and details of the case. At the same time, the visualization module also supports interactive operations and data filtering functions, so that users can further explore and mine the hidden information and value in the case data according to their needs and interests.

[0090] This embodiment also provides a semantic association analysis method of a semantic association analysis model, including: S101, acquiring historical target data and preprocessing the historical target data; S102, building a comprehensive text semantic analysis model based on deep neural network; S103, training a comprehensive text semantic analysis model according to the preprocessed historical target data; S104, inputting real-time target data, preprocessing the real-time target data, and inputting the preprocessed real-time target data into the trained comprehensive text semantic analysis model to obtain an output result; S105, performing semantic association analysis on the output results.

[0091] In the embodiment of the present application, the comprehensive text semantic analysis model includes: a first text prediction model, a second text prediction model and a third text prediction model; The first text prediction model is used to predict the target data which is text type; The second text prediction model is used to predict the target data which is a picture type; The third text prediction model is used to predict the target data which is audio or video type.

[0092] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0093] This embodiment also provides a computer device, which may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 3 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a semantic association analysis method of a semantic association analysis model is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0094] This embodiment further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: Acquire historical target data and preprocess the historical target data; Establish a comprehensive text semantic analysis model based on deep neural network; Train a comprehensive text semantic analysis model based on the preprocessed historical target data; Input real-time target data, pre-process the real-time target data, and input the pre-processed real-time target data into the trained comprehensive text semantic analysis model to obtain output results; Perform semantic association analysis on the output results.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0096] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0097] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0098] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0100] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for generating a semantic association analysis model, characterized in that: The semantic association analysis model comprises: a data collection module (100), a data preprocessing module (200), a comprehensive semantic prediction model module (300), a comprehensive semantic analysis module (400) and a visualization module (500); The data acquisition module (100) is used to acquire first target data, and transmit the first target data to the data preprocessing module (200); The data preprocessing module (200) is used to perform a first classification on the first target data transmitted by the data acquisition module (100) to obtain second target data, and to perform a first preprocessing on the second target data to obtain third target data, and to transmit the third target data to the comprehensive semantic prediction model module (300); The comprehensive semantic prediction model module (300) is used to perform comprehensive semantic prediction on the third target data transmitted by the data preprocessing module (200) through a comprehensive semantic prediction model obtained through pre-training of a deep neural network, and transmit the prediction result of the comprehensive semantic prediction model to the comprehensive semantic analysis module (400); The comprehensive semantic analysis module (400) is used to analyze the prediction results transmitted by the comprehensive semantic prediction model module (300), and transmit the prediction results after the analysis to the visualization module (500); The visualization module (500) is used to perform visualization processing based on the prediction results after the analysis transmitted by the comprehensive semantic analysis module (400).

2. The method for generating a semantic association analysis model according to claim 1, characterized in that: The comprehensive semantic prediction model module (300) comprises: a data processing and input unit (301), a comprehensive semantic prediction model unit (302), a comprehensive semantic prediction model updating unit (303) and a data storage unit (304); The data processing and input unit (301) is used to perform a second preprocessing on the third target data transmitted by the data preprocessing module (200) to obtain fourth target data, and to perform a second classification on the fourth target data to obtain fifth target data, and to transmit the fifth target data to the comprehensive semantic prediction model unit (302) and the data storage unit (304); The comprehensive semantic prediction model unit (302) performs model prediction according to the fifth target data transmitted thereto, obtains prediction results of different models, and transmits the prediction results to the comprehensive semantic analysis module (400) and the data storage unit (304); The data storage unit (304) is used to store the fifth target data generated by the data processing and input unit (301) and the prediction results generated by the comprehensive semantic prediction model unit (302) for the fifth target data.

3. The method for generating a semantic association analysis model according to claim 2, characterized in that: The comprehensive semantic prediction model updating unit (303) comprises: The comprehensive semantic prediction model updating unit (303) presets an update cycle, and when the update cycle is reached and the model needs to be updated, the first correct data stored in the data storage unit (304) during the cycle is retrieved to update the model, and the updated model is transmitted to the comprehensive semantic prediction model unit (302) to replace the original model; The first correct data is the fifth target data generated by the data processing and input unit (301) whose prediction result within the period is accurate, and the prediction result generated by the comprehensive semantic prediction model unit (302) for the fifth target data.

4. The method for generating a semantic association analysis model according to claim 3, characterized in that: The comprehensive semantic prediction model unit (302) comprises: a first text prediction model part (302a), a second text prediction model part (302b) and a third text prediction model part (302c); The first text prediction model unit (302a) is used to predict the first text data in the fifth target data, the second text prediction model unit (302b) is used to predict the second text data in the fifth target data, and the third text prediction model unit (302c) is used to predict the third text data in the fifth target data.

5. The method for generating a semantic association analysis model according to claim 4, characterized in that: The comprehensive semantic analysis module (400) comprises: a first analysis unit (401), a second analysis unit (402), a third analysis unit (403), and a fourth analysis unit (404); The first analysis unit (401) is used to receive the prediction result transmitted by the first text prediction model unit (302a), and perform a first text analysis on the prediction result; The second analysis unit (402) is used to receive the prediction result transmitted by the second text prediction model part (302b), and perform a second text analysis on the prediction result; The third analysis unit (403) is used to receive the prediction result transmitted by the third text prediction model unit (302c), and perform a third text analysis on the prediction result; The fourth analysis unit (404) is used to perform a first correlation analysis on the first text analysis result, the second text analysis result and the third text analysis result, and transmit the first correlation analysis to the visualization module (500) for visualization processing.

6. The method for generating a semantic association analysis model according to claim 5, characterized in that: The data preprocessing module (200) comprises: a first classification unit (201), a first preprocessing unit (202), a first data output unit (203) and a first preprocessing adjustment unit (204); The first classification unit (201) is used to perform a first classification on the first target data according to a preset classification rule; The first preprocessing unit (202) is used to perform a first preprocessing on the second target data after the first classification, and generate third target data; The first data output unit (203) transmits the pre-processed third target data to the comprehensive semantic prediction model module (300); The first preprocessing adjustment unit (204) is used to obtain first error data stored in the data storage unit (304) in the comprehensive semantic prediction model module (300), and adjust the preprocessing parameters of the first preprocessing unit (202) according to the first error data; The first error data is the fifth target data generated by the data processing and input unit (301) whose prediction result within the period is erroneous, and the prediction result generated by the comprehensive semantic prediction model unit (302) for the fifth target data.

7. A semantic association analysis method applied to the semantic association analysis model of claim 1, characterized in that: The method comprises: Acquiring historical target data, and preprocessing the historical target data; Establish a comprehensive text semantic analysis model based on deep neural network; Training the comprehensive text semantic analysis model based on the preprocessed historical target data; Input real-time target data, pre-process the real-time target data, and input the pre-processed real-time target data into the trained comprehensive text semantic analysis model to obtain output results; Perform semantic association analysis on the output results.

8. The semantic association analysis method of the semantic association analysis model according to claim 7, characterized in that: The comprehensive text semantic analysis model includes: a first text prediction model, a second text prediction model and a third text prediction model; The first text prediction model is used to predict text-type target data; The second text prediction model is used to predict the target data which is a picture type; The third text prediction model is used to predict the target data which is audio or video type data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 7 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 7 to 8 are implemented.

Images