Judicial data file generation method and system and computer program product

Abstract

The invention discloses a judicial data file generation method and system and a computer program product. The method comprises the following steps: acquiring image data and audio data of an electronic file; preprocessing the electronic file image data; identifying the preprocessed electronic file image data by adopting an OCR (Optical Character Recognition) technology to obtain first text data; recognizing the audio data by adopting a language recognition technology to obtain second text data; fusing the first text data and the second text data to obtain fused data; analyzing the fusion data by using a pre-trained large language model, and extracting key information of the fusion data; and classifying the fusion data according to the key information, and allocating a corresponding label to each piece of fusion data to obtain a judicial data file in structured storage. According to the judicial data file generation method and system and the computer program product provided by the invention, the limitation of a conventional regular expression and NLP technology is overcome, and the accuracy and efficiency of judicial file structuring are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and specifically to a judicial data file generation method, system and computer program product. Background Art

[0002] Since the Supreme Court proposed the concept of "smart courts", Internet justice has gone through three development stages: "justice + Internet" - "Internet + justice" - "Internet justice". Smart courts are an important achievement of the intelligent construction of politics and law, and an important part of the construction of the People's Courts' informatization version 4.0. At present, smart courts have gone through multiple stages of updates and iterations, and have basically established an information system with "digital rule of law and smart justice" as the core.

[0003] Since 2013, the total number of cases in courts across the country has increased rapidly at an average annual rate of 13%, increasing 2.4 times in 10 years; the average annual number of cases handled by judges has increased from 187 cases in 2017 to 357 cases in 2023, and the contradiction between people and cases has become increasingly prominent.

[0004] Digital reform is an important opportunity to comprehensively enhance the modernization capabilities of judges. At present, the large number of cases and small number of judges has become a nationwide phenomenon, which has risen from a minor and local contradiction in the past to a basic contradiction in the judiciary. Practice has proved that it is difficult to completely solve the problems of large number of cases and small number of judges and trial supervision by relying solely on traditional means, tools, methods, mechanisms and systems. Give full play to the traction of digitalization, the biggest variable in today's world, and promote institutional reshaping through digital empowerment, so as to optimize the allocation of trial resources to the maximum extent, realize the cross-regional, cross-level and cross-court allocation of judicial resources, and effectively solve the problems of uneven workload of judges and blind spots in supervision.

[0005] To achieve digital reform and digital empowerment, we must go through the three stages of "platformization + paperless + intelligentization". Intelligent empowerment is to achieve the transformation from "judicial informatization" to "judicial intelligence". It is necessary to make full use of the advantages of "algorithms + data", make full efforts in intelligent trial, intelligent execution, intelligent service and intelligent management, and actively explore the deep integration of artificial intelligence and judicial activities. File data is the starting point and foundation of the above-mentioned intelligent empowerment, and is a "data element".

[0006] Data files rely on technologies such as OCR text recognition, ASR voice recognition, NLP natural semantic processing, and large models to complete the electronicization, digitization, and intelligentization of file data. They are the foundation of the construction of smart courts and the source of intelligent trials.

[0007] At present, most methods for constructing judicial data files use methods based on regular expressions and natural language processing (NLP). Due to the diversity of file materials, the use of traditional regular expressions and natural language processing technologies faces some technical limitations. These limitations can be summarized as follows:

[0008] 1. Limitations of Regular Expressions

[0009] 1. Lack of flexibility: Regular expressions are suitable for processing strings with fixed patterns. However, in judicial files, text formats are diverse, including handwriting, documents in different formats (such as PDF, text in pictures, etc.). These non-standard formats are often difficult to match with simple regular expression patterns.

[0010] 2. Difficulty in processing complex layouts: The text in the dossier materials often contains complex layouts, such as tables, mixed text and graphics, etc. Traditional regular expressions are unable to handle such complex layouts and it is difficult to accurately separate text and image content.

[0011] 3. Ineffective handwriting recognition: Regular expressions are completely unable to recognize and process handwritten evidence materials because handwritten text does not have a fixed character pattern.

[0012] 2. Limitations of NLP Technology

[0013] 1. Limitations of contextual understanding: NLP technology is inadequate in understanding complex contexts and implicit meanings. The texts in judicial files often contain rich contextual information, which is crucial for accurately understanding the content of the text, but it is difficult for NLP technology to fully capture it.

[0014] 2. Challenges of diversity and difference: The texts in judicial files involve multiple languages ​​and expressions. NLP technology needs to adapt to this diversity and difference, but current NLP models still face challenges in processing cross-language and cross-domain texts.

[0015] 3. Challenges of data quality and scale: The performance of NLP technology depends on high-quality training data. However, in the judicial field, high-quality annotated data is often scarce, and large-scale data sets may contain bias and inaccuracy, which limits the application effect of NLP technology in the structuring of judicial files.

[0016] 4. Explainability issue: NLP models based on deep learning are often viewed as “black boxes”, and their internal decision-making processes lack explainability. In the judicial field, this unexplainability may lead to distrust of the model results, thus affecting its application.

[0017] 5. Security and privacy issues: NLP technology may face security and privacy issues when processing sensitive and personal data. Judicial files may contain sensitive information, and it is necessary to ensure the security and privacy of this information during the processing process.

[0018] In short, with regard to the problem of judicial file structuring, traditional regular expressions and NLP technologies have certain limitations, resulting in low efficiency and poor accuracy in the construction of judicial data files. Summary of the invention

[0019] To this end, the present application provides a judicial data file generation method, system and computer program product to solve the problems of low efficiency and poor accuracy of judicial data file construction methods in the prior art.

[0020] In order to achieve the above objectives, this application provides the following technical solutions:

[0021] In a first aspect, a method for generating a judicial data file comprises:

[0022] Step 1: Obtain electronic file image data and audio data;

[0023] Step 2: preprocessing the electronic file image data; the preprocessing includes data enhancement processing and image content classification;

[0024] Step 3: Using OCR technology to identify the pre-processed electronic file image data to obtain first text data;

[0025] Step 4: Using language recognition technology to identify the audio data to obtain second text data;

[0026] Step 5: Fusing the first text data and the second text data to obtain fused data;

[0027] Step 6: Analyze the fused data using a pre-trained large language model and extract key information of the fused data;

[0028] Step 7: Classify the fused data according to the key information, and assign corresponding labels to each piece of fused data to obtain a structured storage of judicial data files.

[0029] Preferably, in step 2, the enhancement process is performed by using image processing technology.

[0030] Preferably, in step 2, a deep learning algorithm is used for classifying the image content.

[0031] Preferably, in step 5, a cross-modal learning algorithm is used when fusing the first text data and the second text data.

[0032] Preferably, in step 6, the key information includes case type, cause of action, party information, legal terms and focus of the case dispute.

[0033] Preferably, it also includes: constructing a dynamically updated legal knowledge graph, integrating the extracted entities and relationships into the legal knowledge graph, and forming a comprehensive legal knowledge base.

[0034] Preferably, the method further includes: utilizing a pre-trained large language model to perform reasoning analysis on implicit information and logical relationships in the fused data, and predicting a decision result.

[0035] Preferably, the method further includes: displaying the structured stored judicial data files through a front-end page.

[0036] In a second aspect, a judicial data file generation system includes:

[0037] A data acquisition module, used to acquire electronic file image data and audio data;

[0038] A data preprocessing module, used for preprocessing the electronic file image data; the preprocessing includes data enhancement processing and image content classification;

[0039] An image data recognition module, used to recognize the pre-processed electronic file image data using OCR technology to obtain first text data;

[0040] An audio data recognition module, used to recognize the audio data using a language recognition technology to obtain second text data;

[0041] A data fusion module, used for fusing the first text data and the second text data to obtain fused data;

[0042] An information extraction module, used to analyze the fused data using a pre-trained large language model and extract key information of the fused data;

[0043] The data classification module is used to classify the fused data according to the key information and assign a corresponding label to each fused data to obtain a structured storage of judicial data files.

[0044] In a third aspect, a computer program product includes a computer program or instructions, which, when executed by a processor, implement the steps of a method for generating a judicial data file.

[0045] Compared with the prior art, this application has at least the following beneficial effects:

[0046] The present application provides a method, system and computer program product for generating judicial data files, which obtain electronic file image data and audio data; pre-process the electronic file image data; use OCR technology to identify the pre-processed electronic file image data to obtain first text data; use language recognition technology to identify audio data to obtain second text data; fuse the first text data and the second text data to obtain fused data; use a pre-trained large language model to analyze the fused data and extract key information of the fused data; classify the fused data according to the key information and assign corresponding labels to each fused data to obtain a structured stored judicial data file. The method, system and computer program product for generating judicial data files provided by the present application overcome the limitations of traditional regular expressions and NLP technologies and improve the accuracy and efficiency of judicial file structuring. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more intuitively illustrate the prior art and the present application, exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing the present application; for example, those skilled in the art are capable of easily making conventional adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components) based on the technical concepts and exemplary drawings disclosed in the present application.

[0048] Figure 1 A flow chart of a method for generating judicial data files provided in Example 1 of the present application. DETAILED DESCRIPTION

[0049] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0050] In the description of this application: unless otherwise specified, the meaning of "plurality" is two or more. The terms "first", "second", "third", etc. in this application are intended to distinguish the objects referred to, and do not have special meanings in terms of technical connotations (for example, they should not be understood as emphasizing the importance or order, etc.). Expressions such as "including", "comprising", "having", etc. also mean "not limited to" (certain units, components, materials, steps, etc.).

[0051] The terms such as "upper", "lower", "left", "right", "middle", etc. cited in this application are usually used to indicate the general relative position relationship for the purpose of intuitive understanding by referring to the drawings, and are not absolute limitations on the position relationship in the actual product.

[0052] The data file is a comprehensive information set generated after in-depth analysis and organization of court file data. It integrates technologies such as image recognition, natural language processing (NLP) and large language models, and makes full use of the understanding and generation capabilities of large language models. By training a large-scale corpus, the file data is deconstructed and analyzed, and the trial elements are extracted. The case is expressed in a graphical way, which sorts out the case elements, case facts and basic information that the judge is concerned about during the trial, forms the case context information, and generates a case portrait.

[0053] Embodiment 1

[0054] See also Figure 1 This embodiment provides a method for generating a judicial data file, including:

[0055] S1: Obtaining electronic file image data and audio data;

[0056] Specifically, audio data includes court recordings, interrogation records and other audio materials.

[0057] S2: preprocessing the electronic file image data; the preprocessing includes data enhancement processing and image content classification;

[0058] In this step, image processing technology is used for data enhancement, specifically including: using advanced image processing technology, including high-resolution scanning, image denoising, contrast enhancement, etc., to ensure that the image data in the file (such as handwritten notes, evidence photos, charts, etc.) are clear and legible.

[0059] Deep learning algorithms are used to classify image content, including: using deep learning algorithms for image recognition, automatically identifying and classifying image content, such as person recognition, object detection, etc., laying the foundation for subsequent intelligent analysis.

[0060] S3: using OCR technology to recognize the pre-processed electronic file image data to obtain first text data;

[0061] Specifically, OCR technology is used to convert electronic file image data into text data to achieve preliminary digitization of information.

[0062] S4: using language recognition technology to recognize the audio data to obtain second text data;

[0063] Specifically, this step performs high-precision digital conversion of audio data such as trial recordings and interrogation records, and combines advanced speech recognition technology to accurately transcribe the speech content into text. Multi-language and multi-dialect recognition models are used to improve transcription accuracy, and keyword retrieval is supported to facilitate subsequent rapid location of key information in the audio. Among them, language recognition technology is used to identify audio materials such as recording materials submitted by the parties and trial recordings, and convert them into text records to supplement text data.

[0064] S5: Fusing the first text data and the second text data to obtain fused data;

[0065] Specifically, this step effectively integrates data from multiple modalities, such as text converted from images, text converted from audio, and text, and establishes associations between data from different modalities through cross-modal learning algorithms to achieve information complementarity and verification. For example, combining the text of a court trial recording with synchronized video footage can help understand the emotions, body language, and other non-verbal information of the parties involved, providing a more comprehensive perspective for case analysis.

[0066] This step combines information from multiple modalities, such as text, images, and audio, to provide a more comprehensive description of the case, thereby more accurately extracting the cause of the case.

[0067] S6: Use the pre-trained large language model to analyze the fused data and extract the key information of the fused data;

[0068] This step uses a pre-trained large language model to conduct in-depth analysis of the text data in the case file, including word segmentation, part-of-speech tagging, syntactic analysis, entity recognition and linking, etc., to achieve accurate extraction of key information such as case type, cause of action, case facts, party information, legal provisions, and focus of case disputes. This not only facilitates data retrieval and utilization, but also provides a basis for subsequent intelligent analysis.

[0069] The large language model has learned a large amount of text data through pre-training, and has mastered rich language knowledge and context understanding capabilities. In data file generation, the large language model can conduct in-depth analysis of text data obtained by OCR and language recognition, and extract key information of the case, such as the relationship between the parties, the focus of the case dispute, and the applicable legal provisions. This in-depth analysis capability helps judges and lawyers grasp the case more quickly and accurately, and provides strong support for case hearing and decision-making.

[0070] Based on its powerful text generation capabilities, the large language model can automatically generate a case summary based on the input text data. The case summary is a concise summary of the overall situation of the case, which helps users quickly understand the case background and main facts. The case summary generated by the large language model is not only fluent in language and clear in logic, but also can accurately reflect the core content of the case, greatly improving the readability and usability of the case file data.

[0071] The large language model has learned a lot of legal knowledge and case data during the pre-training process, so it has a certain legal reasoning ability. In the data file generation, the large language model can make legal application reasoning based on the facts of the case and the legal provisions, and provide legal application suggestions and references for judges and lawyers. This legal application reasoning ability helps to ensure the fairness and legality of case trials and improve the accuracy and authority of judicial judgments. The training process of the large language model is as follows:

[0072] S601: Data preparation and preprocessing;

[0073] Data Collection:

[0074] (1) Legal texts and regulations: Collect the latest legal texts, judicial interpretations, administrative regulations and other text data from national legal databases, official websites of courts at all levels and other channels.

[0075] (2) Trial data: File data from publicly available historical trial cases, including verdicts, rulings, mediation documents, etc.

[0076] (3) File data: including case facts, evidence materials, statements of the parties, defense opinions, and other detailed file information.

[0077] Data preprocessing:

[0078] (1) Cleaning and deduplication: Delete useless characters, duplicate data, and noisy data to ensure the accuracy and consistency of the data set.

[0079] (2) Word segmentation and tokenization: Split text data into words or sub-word units to facilitate model processing.

[0080] (3) Build a vocabulary: Based on the cleaned text data, build a vocabulary containing common legal terms.

[0081] (4) Data labeling: For some tasks (such as legal question answering, case verdict prediction, etc.), data labeling is required to provide the labels required for supervised learning.

[0082] S602: Model design and establishment;

[0083] (1) Base model selection: The Hanwang Tiandi model is selected as the base. Based on the characteristics of the legal field, the large language model is adapted to the field and further pre-trained or fine-tuned to make it better at processing legal texts.

[0084] (2) Parameter setting, number of layers, hidden state dimensions: Set appropriate model parameters according to computing resources and model requirements. Optimizer and loss function: Select appropriate optimization algorithm (such as Adam) and loss function (such as cross entropy loss) to optimize the model training process.

[0085] S603: Model training;

[0086] (1) Batch training: The preprocessed data is input into the model in batches for training.

[0087] (2) Mixed precision training: Use data and parameters of different precisions for training to speed up the training process and reduce memory usage.

[0088] (3) Forward propagation: The model receives an input text sequence and gradually generates a prediction for the next word or character.

[0089] (4) Backpropagation: By comparing the output predicted by the model with the actual next word or character, the loss function is calculated and the error is backpropagated to update the model parameters.

[0090] (5) Optimization algorithm: Use optimization algorithms such as gradient descent to minimize the loss function.

[0091] (6) Performance evaluation: Regularly use the test set to evaluate model performance, such as perplexity, BLEU score and other indicators.

[0092] (7) Adjust hyperparameters such as learning rate and batch size based on the evaluation results to improve the model performance.

[0093] S604: Model evaluation and tuning.

[0094] (1) Multi-dimensional evaluation: Model performance is evaluated from multiple dimensions, including legal language understanding, legal knowledge question answering, legal prediction, and legal text generation.

[0095] (2) Combination of manual and automatic evaluation: A comprehensive evaluation is performed by combining automatic evaluation indicators (such as BLEU, ROUGE, etc.) with manual evaluation.

[0096] (3) Instruction fine-tuning: Construct instruction datasets for fine-tuning training for specific tasks in the legal field (such as contract information extraction, case judgment generation, etc.).

[0097] (4) Knowledge distillation: The “soft” output of the large model is used as the input of the teacher model and then passed to the student model through the teacher model to reduce the model size and computational complexity.

[0098] This step uses large language model technology to improve the understanding of text semantics and context, thereby improving the accuracy of case details and case causes extraction.

[0099] S7: Classify the fused data according to key information and assign corresponding labels to each piece of fused data to obtain a structured storage of judicial data files.

[0100] This step intelligently classifies text data based on case type, legal field, and other dimensions, and automatically assigns corresponding labels to each document or paragraph, such as "contract dispute", "criminal liability", etc., to facilitate users to quickly locate the required information. At the same time, a dynamically updated legal knowledge graph is constructed, and the extracted entities and relationships are integrated into it to form a comprehensive legal knowledge base.

[0101] A judicial data file generation method provided in this embodiment is also capable of context perception and reasoning analysis, specifically including: combining the context understanding ability of a large language model to perform reasoning analysis on implicit information and logical relationships in the text, such as inferring the motivations behind the case, predicting possible judgment results, etc., to provide judges and lawyers with deeper insights.

[0102] The judicial data file generation method provided in this embodiment can also intelligently present the data file in a digital manner. Specifically, the text data that has been deeply analyzed and structured is presented in an intuitive graphical and visual form, such as a timeline showing the progress of the case, a relationship map showing the relationship between the parties, etc., making the data file an intelligent tool that is easy to understand and efficiently used. At the same time, it supports interactive query and exploration, and users can customize views as needed to further enhance the user experience.

[0103] It should be noted that this implementation strictly abides by relevant laws and regulations throughout the data digitization process, and adopts technical means such as encrypted storage, access control, and anonymization processing to ensure that personal privacy and sensitive information are not leaked. In particular, strict desensitization processing is implemented for image and audio data involving personal privacy, thereby protecting the legitimate rights and interests of the parties and ensuring the privacy and security of user data.

[0104] The judicial data file generation method provided in this embodiment overcomes the limitations of traditional regular expressions and NLP technologies, and combines a variety of technical means, such as OCR technology, machine learning, deep learning, large language models, etc., to improve the accuracy and efficiency of judicial file structuring. Through advanced technical solutions, data files realize the comprehensive digitization and intelligent processing of file data, provide strong support for smart courts in terms of smart trial, smart execution, smart service, and smart management, and improve the utilization efficiency and accessibility of file data.

[0105] Embodiment 2

[0106] This embodiment provides a judicial data file generation system, which includes three layers: data layer, processing layer, and service layer:

[0107] Data Layer:

[0108] (1) Establish a raw data area to store unprocessed file images, audio files, etc.

[0109] (2) Establish a processing data area to store text data after OCR and language recognition, as well as comprehensive information after multimodal fusion.

[0110] (3) Implement data encryption and access control mechanisms to ensure data security and privacy.

[0111] Processing layer:

[0112] (1) Use language recognition technology to process court audio and video and convert them into text data;

[0113] (2) Use OCR text recognition technology to perform text recognition on the image data in the file and convert it into text data;

[0114] (3) Construct an NLP processing module to be responsible for basic processing such as word segmentation, part-of-speech tagging, and syntactic analysis of the text;

[0115] (4) Deploy a large language model analysis module to perform deep semantic understanding, key information extraction, multimodal fusion analysis, etc. Perform NLP processing on the case file data to extract key elements such as basic case information, party information, litigation requests, factual reasons, cause of action, and judgment results; perform NLP processing on the text of the court trial speech transcription to extract key information such as questions and answers, statements, and debates during the trial.

[0116] Service layer:

[0117] (1) Provide RESTful API interface to support data request and interaction of front-end applications.

[0118] (2) Realize intelligent management of data files, including the creation, editing, retrieval, and analysis of electronic files.

[0119] (3) Establish a compliance review module to ensure that all data processing activities comply with relevant laws and regulations.

[0120] (4) Develop a user rights management system to assign data access rights based on user roles to ensure data security.

[0121] Therefore, this embodiment provides a judicial data file generation system which specifically includes:

[0122] A data acquisition module, used to acquire electronic file image data and audio data;

[0123] A data preprocessing module, used for preprocessing the electronic file image data; the preprocessing includes data enhancement processing and image content classification;

[0124] An image data recognition module, used to recognize the pre-processed electronic file image data using OCR technology to obtain first text data;

[0125] An audio data recognition module, used to recognize the audio data using a language recognition technology to obtain second text data;

[0126] A data fusion module, used for fusing the first text data and the second text data to obtain fused data;

[0127] An information extraction module, used to analyze the fused data using a pre-trained large language model and extract key information of the fused data;

[0128] The data classification module is used to classify the fused data according to the key information and assign a corresponding label to each fused data to obtain a structured storage of judicial data files.

[0129] For the specific implementation content of each module in a judicial data file generation system, please refer to the above definition of a judicial data file generation method, which will not be repeated here.

[0130] Embodiment 3

[0131] This embodiment provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of a method for generating a judicial data file.

[0132] The technical features of the above embodiments may be arbitrarily combined (as long as there is no contradiction in the combination of these technical features). To make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

Claims

1. A method for generating judicial data files, characterized in that: include: Step 1: Obtain electronic file image data and audio data; Step 2: pre-processing the electronic file image data; The preprocessing includes data enhancement processing and image content classification; Step 3: Using OCR technology to identify the pre-processed electronic file image data to obtain first text data; Step 4: Using language recognition technology to identify the audio data to obtain second text data; Step 5: Fusing the first text data and the second text data to obtain fused data; Step 6: Analyze the fused data using a pre-trained large language model and extract key information of the fused data; Step 7: Classify the fused data according to the key information, and assign corresponding labels to each piece of fused data to obtain a structured storage of judicial data files.

2. The method for generating judicial data files according to claim 1, characterized in that: In the step 2, the enhancement process is performed by using image processing technology.

3. The method for generating judicial data files according to claim 1, characterized in that: In step 2, a deep learning algorithm is used for classifying the image content.

4. The method for generating judicial data files according to claim 1, characterized in that: In the step 5, a cross-modal learning algorithm is used when fusing the first text data and the second text data.

5. The method for generating judicial data files according to claim 1, characterized in that: In step 6, the key information includes case type, cause of action, party information, legal terms and focus of the case dispute.

6. The method for generating judicial data files according to claim 1, characterized in that: Also includes: Construct a dynamically updated legal knowledge graph, integrate the extracted entities and relationships into the legal knowledge graph, and form a comprehensive legal knowledge base.

7. The method for generating judicial data files according to claim 1, characterized in that: Also includes: A pre-trained large language model is used to perform reasoning and analysis on the implicit information and logical relationships in the fused data, and predict the judgment result.

8. The method for generating judicial data files according to claim 1, characterized in that: Also includes: The structured stored judicial data files are displayed through the front-end page.

9. A judicial data file generation system, characterized in that: include: A data acquisition module, used to acquire electronic file image data and audio data; A data preprocessing module, used for preprocessing the electronic file image data; The preprocessing includes data enhancement processing and image content classification; An image data recognition module, used to recognize the pre-processed electronic file image data using OCR technology to obtain first text data; An audio data recognition module, used to recognize the audio data using a language recognition technology to obtain second text data; A data fusion module, used for fusing the first text data and the second text data to obtain fused data; An information extraction module, used to analyze the fused data using a pre-trained large language model and extract key information of the fused data; The data classification module is used to classify the fused data according to the key information and assign a corresponding label to each fused data to obtain a structured storage of judicial data files.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images