Litigation strategy generation and complaint rate prediction method based on multi-modal data fusion and intelligent reasoning

Abstract

The invention discloses a litigation strategy generation and complaint rate prediction method based on multi-modal data fusion and intelligent reasoning. The method aims at improving the efficiency and pertinence of legal consultation. Case description and evidence files are analyzed by means of natural language processing, an OCR (optical character recognition) technology, image recognition and the like, and a comprehensive solution of complaint risk early warning, litigation request recommendation and complaint rate prediction can be realized by the method in combination with a dynamic knowledge graph and a probability graph model constructed by a historical judgment database. Specifically, the system firstly analyzes case information and related evidences input by a user through NLP, OCR and image recognition technologies, then constructs a dynamic knowledge graph based on the information, and analyzes a case in a manner of aligning with the dynamic knowledge graph. Then, through comparison of historical discriminant data, the system can generate a complaint risk early warning, recommend a corresponding litigation request, and predict a complaint rate and influence factors thereof based on a probabilistic graph model and a path reasoning method. In addition, the method also emphasizes the interpretability of a prediction result, and helps a user to understand a prediction basis through knowledge graph path visualization and an LIME algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer networks, and particularly relates to a method for generating litigation strategies and predicting the winning rate based on multi-modal data fusion and intelligent reasoning. Background Art

[0002] Traditional case analysis work often relies on manual experience and manual retrieval methods, which are not only inefficient but also prone to missing key information. Ordinary parties usually lack legal professional knowledge and do not know how to correctly file litigation requests. However, with the continuous development of cutting-edge technologies such as natural language processing, large models, knowledge graph technology, and probabilistic graphical models, automated, intelligent, and interpretable litigation request recommendations and case winning rate predictions are gradually becoming a reality. The application of these technologies can effectively improve the efficiency and accuracy of case analysis, provide more accurate and professional legal services for parties, and contribute to the intelligent development of judicial work. Summary of the Invention

[0003] Object of the Invention: To provide a method for generating litigation strategies and warning against losing risks based on multi-modal data and intelligent reasoning, and to solve the problems of low efficiency and poor pertinence in traditional legal consultations; through a dynamic knowledge graph and a probabilistic graphical model, an integrated solution for warning against losing risks, recommending litigation requests, and predicting the winning rate is realized. Technical Solution

[0004] Input and Parsing of Multi-modal Case Data: Parse the case description text input by the user through natural language processing (NLP) technology and a large language model (LLM) to extract the case situation; Parse the unstructured evidence files (including images, videos, and audios) uploaded by the user through OCR technology, image recognition technology, and multi-modal feature fusion methods, and perform object detection (YOLO algorithm) on the image evidence to extract key physical evidence regions (such as contract signatures, infringing product logos); Perform temporal action recognition (3D-CNN model) on video evidence to generate behavioral semantic descriptions (such as the process of infringement behavior) to generate structured evidence data; Construction and Alignment of Dynamic Knowledge Graph: Construct a knowledge graph in the legal field based on entities (case types, involved amounts, evidence, disputes, legal provisions, case acceptance courts, attorneys, judgment results, etc.) and relationships in the historical judgment database; Use a graph embedding algorithm (such as GraphSAGE) to vectorize the user's case elements and perform semantic alignment with the nodes in the knowledge graph to generate a multi-dimensional association network; dynamically update the node attributes of the graph based on the legal provision revision records and the timestamp of judicial policy changes; Multi-modal Similar Judgment Screening and Interactive Form Generation: Based on text semantic matching (Sentence-BERT), knowledge graph path weights (Dijkstra algorithm), and evidence contrast learning (SimCLR), calculate the comprehensive similarity score between the user's case and historical judgments; Extract the defendant's arguments and disputed focuses in historical cases to generate structured risk labels; through multi-modal similarity calculation, screen the risk points of losing the lawsuit that are closest to the user's case description, and generate a risk form of cases that may lead to losing the lawsuit according to the score ranking, supporting the user to select according to the actual situation through the interactive interface; Dynamic Recommendation and Real-time Editing of Litigation Requests: According to the user's selection, count the adoption rate of litigation requests in historical similar judgments to generate a litigation request recommendation list; Allow the user to perform real-time editing on the recommended litigation requests (modify the request content, adjust the amount); Winning Rate Prediction and Interpretability Analysis Based on Path: Use a probabilistic graph model (Bayesian network or Markov network) to model the conditional dependence relationship between case elements, and perform joint probability reasoning in combination with the causal path in the knowledge graph (litigation request → legal provision → risk → judgment result); adopt Monte Carlo sampling (MCMC) to simulate the judicial decision-making scenario, and output the winning rate distribution and confidence interval; Through an interpretability algorithm (LIME) and knowledge graph path visualization, highlight key influencing factors (such as the weight of evidence integrity); Dynamically adjust the weight of historical cases based on a time decay factor to reduce the interference of judgments with expired or modified legal provisions on the prediction results. Description of the Drawings

[0005] Figure 1 It is a schematic diagram of the process of a technical patent for a litigation strategy generation and winning rate prediction method based on multi-modal data fusion and intelligent reasoning.

[0006] Technical Effects High efficiency: Automatically process a large amount of historical judgment data to improve the efficiency of litigation strategy generation; Accuracy: Improve the pertinence of litigation strategies through losing lawsuit risk warnings and dynamic recommendations; Interpretability: Enhance the credibility of prediction results through knowledge graph path visualization and the LIME algorithm.

[0007] Embodiment The user inputs the case description and evidence files; The system generates a warning form for the risk of losing a lawsuit. Risk 1: The defendant may argue that the contract terms are invalid. Risk 2: The statute of limitations has expired. After the user confirms the risk points, the system recommends litigation requests: "Request the court to order the defendant to fulfill the contract obligations and pay liquidated damages of 500,000 yuan", and "Request the court to order the defendant to bear attorney fees of 5,000 yuan". The user confirms the two litigation requests and edits the request content: "The attorney fees are adjusted to 10,000 yuan". The system predicts that the probability of this litigation request being supported by the court is 50%.

Claims

1. A litigation strategy generation and winning rate prediction method based on multimodal data fusion and intelligent reasoning, characterized in that: The following steps are involved: Multimodal case data input and analysis: Use natural language processing (NLP) technology and large language model (LLM) to analyze the case description text entered by the user and extract the case situation; use OCR technology, image recognition technology and multimodal feature fusion methods to analyze the unstructured evidence files (including images, videos, and audio) uploaded by the user to generate structured evidence data; Dynamic knowledge graph construction and alignment: Based on the entities (case type, amount involved, evidence, disputes, legal provisions, courts accepting cases, attorneys, judgment results, etc.) and relationships in the historical judgment database, a legal knowledge graph is constructed; a graph embedding algorithm (such as GraphSAGE) is used to vectorize the user's case elements and semantically align them with the nodes in the knowledge graph to generate a multi-dimensional association network; Multimodal similar judgment screening and interactive form generation: Based on text semantic matching (Sentence-BERT), knowledge graph path weight (Dijkstra algorithm) and evidence comparison (SimCLR), the comprehensive similarity score between the user's case and historical judgments is calculated; a risk form that may lead to a losing case is generated by sorting the scores, and users are supported to make choices based on actual conditions through an interactive interface; Dynamic recommendation and real-time editing of litigation requests: Based on user selection, statistics on the adoption rate of litigation requests in similar historical judgments are used to generate a list of recommended litigation requests; allowing users to edit recommended litigation requests in real time (modify request content, adjust amount); Path-based win rate prediction and explainability analysis: Use probabilistic graph models (Bayesian networks or Markov networks) to model the conditional dependencies between case elements, and combine the causal paths in the knowledge graph (litigation requests and evidence → legal provisions → risks → judgment results) to perform joint probabilistic reasoning; use Monte Carlo sampling (MCMC) to simulate judicial decision-making scenarios and output the win rate distribution and confidence interval; highlight key influencing factors (such as evidence integrity weight) through the interpretability algorithm (LIME) and knowledge graph path visualization.

2. The method according to claim 1, characterized in that The unstructured evidence analysis further includes: performing target detection (YOLO algorithm) on image evidence to extract key physical evidence areas (such as contract signatures, infringing product logos); performing temporal action recognition (3D-CNN model) on video evidence to generate behavioral semantic descriptions (such as the infringement process).

3. The method according to claim 1, characterized in that: The construction of the knowledge graph further includes: dynamically updating graph node attributes based on legal text revision records and judicial policy change timestamps.

4. The method according to claim 1, characterized in that: The generation of the lawsuit loss risk warning form further includes: extracting the defendant's arguments and controversial points in historical cases to generate structured risk labels; through multimodal similarity calculation, screening the lawsuit loss risk points closest to the user's case description, and generating an interactive form for the user to confirm.

5. The method according to claim 1, characterized in that The winning rate prediction further includes: dynamically adjusting the weights of historical cases based on a time decay factor to reduce the interference of judgments containing expired or amended laws on the prediction results.

6. A litigation strategy generation and litigation risk warning system for implementing the method described in claims 1-5, characterized in that: include: Multimodal parsing engine: integrates NLP, OCR, image / video recognition and multimodal fusion modules; dynamic knowledge graph platform: supports real-time update, semantic alignment and path reasoning; The interactive decision-making terminal provides a loss risk warning form, an editable litigation request panel, and a win rate visualization dashboard; distributed computing cluster: used for large-scale MCMC sampling, graph traversal, and real-time adversarial simulation.

7. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images