Neural network-based legal case relation graph dynamic updating method

Abstract

The invention relates to the technical field of data processing, in particular to a legal case relation graph dynamic updating method based on a neural network. Comprising the steps that original legal case data are collected and preprocessed to obtain target legal data; entities in the target legal data are extracted to serve as nodes, node sub-graphs are obtained through a machine learning model, weight parameters corresponding to the nodes are selected from weight parameter sub-graphs, weights of node edges are obtained based on the weight parameters and a weight prediction model, and a first graph is formed; acquiring newly added legal data, extracting new nodes, acquiring newly added sub-graphs through a machine learning model, and fusing the newly added sub-graphs with the first graph to obtain a second graph; and generating a search vector according to key weight parameters of adjacent nodes and edges in the second map, and embedding the search vector into the second map. According to the method, the problems that the legal case relationship is complex and cannot be dynamically updated are solved, the legal case relationship graph can be dynamically updated, and the accuracy and timeliness of the graph are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method for dynamically updating a legal case relationship graph based on a neural network. Background Art

[0002] With the advancement of the legal informatization process, a large amount of legal case data presents the characteristics of being unstructured, having complex associations, and being dynamically updated. Traditional legal case analysis methods mostly rely on manual experience to construct static knowledge graphs, and the legal entity relationships rely on manual rules extraction, making it difficult to adapt to the complex and changeable case associations. Similar prior arts include a Chinese patent with the publication number CN112527924A, which proposes a method and device for dynamically updating a knowledge graph, including: constructing a knowledge graph based on a first knowledge base to generate a knowledge graph of a target domain, where the first knowledge base is composed of original data of the target domain; generating a second knowledge base based on new knowledge of the target domain, and expanding the data in the second knowledge base into the knowledge graph to obtain an updated knowledge graph; performing a knowledge empowerment operation based on the knowledge graph to obtain an application solution of the knowledge graph, and performing a knowledge reasoning operation on the application solution to obtain a third knowledge base, and expanding the data in the third knowledge base into the knowledge graph to obtain an updated knowledge graph. This invention can achieve the dynamic expansion of the knowledge graph, ensuring the dynamic update and real-time nature of the knowledge graph. In addition, a similar prior art is a US patent with the publication number US11386064B2, which proposes a method for updating a server knowledge graph, executed by a server, including obtaining the server knowledge graph of the server and obtaining a plurality of device knowledge graphs by receiving device knowledge graphs from each of a plurality of devices. The method further includes generating a knowledge graph for expanding the server knowledge graph based on the obtained plurality of device knowledge graphs, and updating the obtained server knowledge graph using the generated knowledge graph for expanding the server knowledge graph. The technical solutions in the above two patent documents both solve the problem of updating the knowledge graph, and are all about the expansion and update of the existing graph, with the weight parameters being fixed, unable to adaptively change according to data updates, accurately quantify the association relationships, and support efficient retrieval. Summary of the Invention

[0003] The present invention provides a method for dynamically updating a legal case relationship graph based on a neural network, which realizes the efficient construction and dynamic maintenance of legal case relationships through multi-stage data processing and model collaboration. The method includes the following steps: Step S1: Collect original legal case data, preprocess the collected original legal case data to obtain a plurality of target legal data; Step S2: Extract multiple entities in the target legal data as nodes, obtain a node subgraph based on all the nodes and a machine learning model, select the weight parameters corresponding to the nodes from the weight parameter subgraph, obtain the weights of the node edges based on the weight parameters and a weight prediction model, and use the weight parameter subgraph and the node subgraph as the first graph; Step S3: Obtain new legal data, extract new nodes from the new legal data, obtain a new subgraph based on the new nodes and the machine learning model, and fuse the new subgraph with the first graph to obtain a second graph; Step S4: Generate a search vector based on adjacent nodes in the second graph and key weight parameters of the edges between the adjacent nodes, and embed the search vector into the second graph.

[0004] As a preferred technical solution of the present invention, after step S4, there is also step S5: Statistically analyze the search information of users at preset time intervals, obtain search nodes from the search information, and display different colors according to the search times of the search nodes, where the more search times, the brighter the color, and vice versa the darker the color. Also obtain the core weight parameters with the largest weight contribution values of the edges corresponding to the N search nodes with the largest search times, increase the corresponding weights of the core weight parameters in the weight prediction model, and correspondingly decrease the weights of the other weight parameters corresponding to the edges, and calculate the weights of the edges corresponding to the weight prediction model and the adjusted weight parameter model, and update the second graph.

[0005] As a preferred technical solution of the present invention, the obtaining of the node subgraph includes: Extract multiple entities as nodes from the target legal data through semantic analysis, determine multiple core nodes in the nodes according to the elements of legal cases, obtain a sub-graph including related nodes related to the core nodes based on each core node through a machine learning model, and fuse different sub-graphs based on the related relationships between the nodes of different sub-graphs through an analysis unit, and obtain the node subgraph.

[0006] As a preferred technical solution of the present invention, the obtaining of the first graph includes: Select the weight parameters corresponding to each node in the node subgraph from the weight parameter subgraph and connect them. According to the text information corresponding to the node in the target legal data and the scoring criteria corresponding to the weight parameters, obtain the scores of each weight parameter corresponding to the node, and use the node, the scores of each corresponding weight parameter, and the weight prediction model to obtain the weight value corresponding to the node. Add the weight parameter scores and the weight value to the edge corresponding to the node. Take the weight parameter subgraph, the node subgraph, and the connection relationship between the two as the first graph.

[0007] As a preferred technical solution of the present invention, obtaining the second graph includes: Extract the new nodes from the new legal data, obtain the connected nodes of the new nodes through semantic analysis, connect the new nodes with the corresponding connected nodes, and select and connect the corresponding weight parameters from the weight parameter subgraph. When the new nodes cannot obtain the corresponding weight parameters from the weight parameter subgraph, calculate the similarity between the new nodes and each node in the first graph, and use the weight parameter corresponding to the node with the highest similarity as the target weight parameter of the new nodes. Calculate the weights of the edges corresponding to the new nodes by the method of step S2, add the weights of the new nodes to the corresponding edges, and obtain the second graph.

[0008] As a preferred technical solution of the present invention, the generation of the search vector includes: Convert the adjacent nodes connected by each node and the corresponding edges in the second graph into search vectors through a semantic conversion model, and based on the weight prediction model, obtain the weights of each weight parameter corresponding to the edges between the nodes in the search vector and the scores of the weight parameters, and obtain the contribution value of each weight parameter to the weight of the edge. Add the weight parameter with the highest corresponding contribution value to the corresponding nodes in the search vector, and also add each search vector to the second graph.

[0009] As a preferred technical solution of the present invention, the machine learning model is a neural network model trained by the correlation relationships between each core node and other nodes in different types of legal cases.

[0010] As a preferred technical solution of the present invention, the value range of N is a positive integer greater than or equal to 1 and less than M, and M is the total number of the search nodes.

[0011] The present invention also provides a legal case relationship graph dynamic update system based on a neural network for implementing the above method. The system includes: The acquisition unit is used to acquire the original legal case data, preprocess the acquired original legal case data, and obtain multiple target legal data; The generation unit is used to extract multiple entities in the target legal data as nodes, obtain a node subgraph based on all the nodes and a machine learning model, select the weight parameters corresponding to the nodes from the weight parameter subgraph, obtain the weights of the node edges based on the weight parameters and a weight prediction model, and use the weight parameter subgraph and the node subgraph as the first graph; The fusion unit is used to obtain new legal data, extract new nodes from the new legal data, obtain an added subgraph based on the new nodes and the machine learning model, and fuse the added subgraph with the first graph to obtain the second graph; The embedding unit is used to generate search vectors based on adjacent nodes in the second graph and key weight parameters of the edges between the adjacent nodes, and embed the search vectors into the second graph.

[0012] The present invention also provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed by a processor, the above method is implemented.

[0013] The beneficial effects of the present invention are as follows: The present invention proposes a method for dynamically updating a legal case relationship graph based on a neural network. Through multi-stage data processing and model collaboration, the efficient construction and dynamic maintenance of legal case relationships are realized. First, after the original legal case data is preprocessed and entity extraction is performed, a node subgraph is generated by combining semantic analysis and a machine learning model, and the strength of the relationships between nodes is quantified through a weight parameter subgraph to form an initial graph, that is, the first graph. This process converts unstructured legal texts into structured relationship networks, solves the problems of fuzzy entity relationships and difficult-to-quantify weights in traditional methods, and significantly improves the interpretability of the graph. Second, for new legal data, new nodes are dynamically fused with the existing graph through similarity matching and a weight prediction model, avoiding the lag of a static graph. At the same time, the integrity of the graph structure is ensured through an adaptive weight allocation mechanism, enabling the system to flexibly handle the diversity and complexity of legal cases. In addition, adjacent nodes are generated through a semantic conversion model and search vectors of key weight parameters are embedded. By combining the user's search behavior, the weight prediction model is dynamically optimized, achieving a double improvement in retrieval efficiency and result accuracy. The user's high-frequency search nodes and associated weight parameters are fed back to the model in real time to drive the dynamic adjustment of the weights of the second graph, enabling the system to adaptively optimize according to actual needs and meet the legal practitioners' requirements for timeliness, accuracy, and personalized retrieval. Through the mutual cooperation of the above technical solutions, the problem of complex legal case relationships and difficult updates is effectively solved, providing efficient and reliable knowledge support for legal research, case analysis, and judicial decision-making. Brief Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0015] Figure 1 It is a flowchart of the method for dynamically updating the legal case relationship graph based on neural network of the present invention; Figure 2 It is a flowchart of the first graph acquisition method of the present invention; Figure 3 It is a flowchart of the method for generating search vectors of the present invention; Figure 4 It is a structure diagram of the system for dynamically updating the legal case relationship graph based on neural network of the present invention. Detailed Embodiments

[0016] The embodiments of the present invention provide a method for dynamically updating the legal case relationship graph based on neural network. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above accompanying drawings of the present invention are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0017] For ease of understanding, the specific processes of the embodiments of the present invention are described below. As Figure 1 shown, the method for dynamically updating the legal case relationship graph based on neural network in the embodiments of the present invention includes: Step S1: Collect the original legal case data, preprocess the collected original legal case data, and obtain a plurality of target legal data; Specifically, the above-mentioned original legal case data is collected by the collection unit and preprocessed to remove duplicate, incomplete or incorrectly formatted cases, such as documents missing key fields, and convert unstructured text, such as judgment documents, into structured data, such as case types, parties, court levels, evidence lists, etc., and unify the data formats from different sources. Through the above technical solutions, the original data is converted into normalized target legal data, providing high-quality input for the subsequent construction of the first knowledge graph and laying a foundation for entity recognition as nodes and relevant relationships.

[0018] Step S2: Extract multiple entities from the target legal data as nodes, obtain a node subgraph based on all the nodes and a machine learning model, select the weight parameters corresponding to the nodes from the weight parameter subgraph, obtain the weights of the node edges based on the weight parameters and a weight prediction model, and use the weight parameter subgraph and the node subgraph as the first knowledge graph; Specifically, based on the above-mentioned target legal data, through semantic analysis, such as dependency syntactic analysis, the nodes in the target legal data are obtained, and the core nodes, such as cases, parties, and evidence, and relevant nodes, such as legal provision citations, are further identified according to legal case elements. Using a pre-trained machine learning model, such as a graph neural network or BERT, the associations between nodes are analyzed to generate local sub-knowledge graphs. By analyzing the overlapping nodes of different sub-knowledge graphs, if multiple sub-knowledge graphs involve the same node, the sub-knowledge graphs are combined into a node subgraph. Weight parameters are assigned to each node from a predefined weight parameter subgraph, such as case quality and sufficiency of evidence. Combining text information, such as the detailed description of evidence in the judgment document, and scoring criteria, such as the integrity of the evidence chain, the scores of each weight parameter are obtained, and the weights of the edges are calculated through a weight prediction model. Through the above technical solutions, a static relationship network of legal cases is constructed through the node subgraph, and the weight parameter subgraph gives quantitative meanings to the nodes and edges, such as the sufficiency score of evidence. The first knowledge graph formed by the combination of the two not only presents the associations between legal entities but also distinguishes the strength of the relationships through weight values, enhancing the interpretability and practicality of the knowledge graph.

[0019] Step S3: Obtain new legal data, extract new nodes from the new legal data, obtain a new subgraph based on the new nodes and the machine learning model, and fuse the new subgraph with the first knowledge graph to obtain the second knowledge graph; Specifically, perform the same preprocessing on the new legal case as in step S1, extract new nodes, obtain the new subgraph corresponding to the above-mentioned newly added legal data through a machine learning model, analyze the correlation between the above-mentioned new nodes and the nodes in the first graph, and select corresponding weight parameters for the above-mentioned new nodes. When no corresponding weight parameters can be found for the above-mentioned new nodes, calculate the similarity between its semantic vector and the existing nodes, borrow the weight parameters of the similar nodes, and recalculate the weights of the new edges through a weight prediction model to ensure the integrity of the fused graph. Through the above technical solution, the second graph can incorporate newly added case information in real time through a dynamic fusion mechanism, avoiding the lag of traditional static graphs. The addition of new types of evidence will not damage the graph structure, but will adaptively expand through similarity matching, ensuring the flexibility and timeliness of the system.

[0020] Step S4: Generate a search vector based on the adjacent nodes in the second graph and the key weight parameters of the edges between the adjacent nodes, and embed the search vector into the second graph.

[0021] Specifically, use a semantic transformation model, such as Sentence-BERT, to encode the adjacent nodes and their adjacency relationships in the second graph into vectors, combine with a weight prediction model, analyze the key weight parameters on the edges and the contribution values of the weights, and screen out the key weight parameters with the highest contribution values, such as "integrity of the evidence chain" with a contribution value ratio of 80%. Embed the selected key weight parameters and their scores into the search vector, and embed the above search vector into the above second graph. Through the above technical solution, the second graph embedded with the search vector supports efficient semantic retrieval. When the user searches, the system preferentially returns highly relevant high-weight evidence rather than secondary information. At the same time, the explicit embedding of the key weight parameters improves the interpretability of the results, meeting the dual needs of users for accuracy and transparency.

[0022] Furthermore, the acquisition of the node subgraph includes: Extract multiple entities as nodes from the target legal data through semantic analysis, determine multiple core nodes among the nodes according to the elements of the legal case, obtain a subgraph including related nodes related to each core node through a machine learning model based on each core node, and fuse different subgraphs based on the correlation between the nodes of different subgraphs through an analysis unit to obtain the node subgraph.

[0023] Specifically, by semantically analyzing the above-mentioned target legal data and extracting multiple entities as mentioned above, such as cases, parties, Evidence 1, Evidence 2, first instance, second instance, etc., and determining the core nodes among the above-mentioned nodes, i.e., entities, according to the elements of legal cases, and obtaining a sub-graph corresponding to the above-mentioned core nodes based on the above-mentioned core nodes and other nodes through a machine learning model. The above-mentioned machine learning model is a neural network model trained with each core word in each legal case and other node data carrying relevant relationship labels. The above-mentioned machine learning model can fully learn the relevant relationships between each core word and other nodes, so as to obtain a relatively accurate above-mentioned sub-graph, and by analyzing the relevant relationships between the nodes of the sub-graph and the nodes of other sub-graphs and fusing based on the above-mentioned relevant relationships, the above-mentioned node sub-graph is obtained. Through the above technical solution, the node sub-graph corresponding to the above-mentioned target legal data can be obtained, laying a foundation for obtaining the first graph corresponding to the above-mentioned target legal data.

[0024] Further, the obtaining of the first graph is as Figure 2 shown and includes: Select the weight parameters corresponding to each node from the weight parameter sub-graph and connect them. According to the text information corresponding to the node in the target legal data and the scoring criteria corresponding to the weight parameters, obtain the scores of the node corresponding to each weight parameter, and use the node, the scores of the node corresponding to each weight parameter, and the weight prediction model to obtain the weight value of the edge corresponding to the node, and add the weight parameter scores and the weight value to the edge corresponding to the node. Use the weight parameter sub-graph and the node sub-graph as the first graph.

[0025] Specifically, the sub-graphs corresponding to different core nodes are also connected according to the relevant relationships between the above-mentioned core nodes and the relevant relationships between the relevant nodes corresponding to different core nodes, and the above-mentioned first graph is obtained. Since the above-mentioned first graph only reflects the relevant relationships between nodes, but cannot reflect the specific relevant content of each node, therefore, the weight parameters corresponding to the above-mentioned nodes are selected from the above-mentioned weight parameter sub-graph, and the above-mentioned nodes and the above-mentioned weight parameters are connected. For example: Case (Node A): such as "labor dispute case", the weight parameters are case quality and sufficiency of evidence, and the weight parameters of the parties (Nodes B, C), for example: compliance of the parties' behavior, consistency of statements, and relevance of the statement content to the case content. According to the relevant content of the above-mentioned nodes, that is, the text information and scoring criteria corresponding to the above-mentioned nodes in the above-mentioned target legal data, the score of each weight parameter is obtained, and the score of each weight parameter corresponding to the above-mentioned node is added to the corresponding edge. By inputting the above-mentioned nodes and the scores of each weight parameter into the above-mentioned weight prediction model, the weight value of the edge corresponding to the above-mentioned node is obtained, where the above-mentioned weight prediction model is a pre-trained neural network model. The above-mentioned weight value is also added to the corresponding edge, and the above-mentioned weight parameter sub-graph and the above-mentioned node sub-graph are used as the above-mentioned first graph. Through the above technical solution, not only can the above-mentioned node sub-graph be obtained, but also the specific content of each node in the above-mentioned node sub-graph can be obtained, increasing the interpretability of the above-mentioned first graph.

[0026] Further, obtaining the second graph includes: From the above-mentioned newly added legal data, the new nodes are extracted, the connection nodes of the new nodes are obtained through semantic analysis, the new nodes are connected to the corresponding connection nodes, and the corresponding weight parameters are selected and connected from the above-mentioned weight parameter sub-graph. When the new nodes cannot obtain the corresponding weight parameters from the above-mentioned weight parameter sub-graph, the similarity between the new nodes and each node in the above-mentioned first graph is calculated, and the weight parameter corresponding to the node with the highest similarity is used as the target weight parameter of the new nodes, and the weight of the edge corresponding to the new nodes is calculated by the method of step S2. The weight of the new nodes is added to the corresponding edge to obtain the second graph.

[0027] Specifically, after obtaining the above-mentioned newly added legal data, preprocessing is first performed, and new nodes are extracted. Based on the above-mentioned new nodes, the above-mentioned local graph is obtained through the above-mentioned machine learning model. By semantic analysis of the first graph and the above-mentioned new nodes, connection nodes related to the above-mentioned new nodes are obtained, and the above-mentioned new nodes and the above-mentioned connection nodes are connected to achieve the fusion of the new nodes and the above-mentioned first graph. The method of step S2 is also repeated to obtain the score of the weight parameter corresponding to the above-mentioned new node, which does not reflect the content of the above-mentioned new node. Since the nodes in the above-mentioned new nodes and the above-mentioned first graph are selected according to the set rules when selecting the above-mentioned weight parameters, when a new type of node appears, for example: WeChat chat evidence, when the weight parameter corresponding to the "WeChat chat evidence" node is not included in the above-mentioned set rules, the cosine value of the semantic vector corresponding to the above-mentioned new node and the semantic vector of each node in the above-mentioned first graph is calculated, and the target node with the highest similarity is obtained according to the cosine value, and the weight parameter corresponding to the above-mentioned target node is used as the weight parameter of the above-mentioned new node. According to the method in step S2, the weight of the edge corresponding to the above-mentioned new node is obtained, and the weight and the score of the weight parameter corresponding to the new node are added to the corresponding edge, so as to obtain the above-mentioned second relationship graph after fusion. Through the above technical solution, the latest case information can be updated to the corresponding case relationship graph in real time according to the progress of the legal case, which is convenient for users to view in real time.

[0028] Further, the generation of the search vector is as Figure 3 shown, and includes: Converting the adjacent nodes connected by each node and the corresponding edge in the second graph into search vectors through a semantic conversion model, obtaining the weight of each weight parameter corresponding to the edge between the nodes in the search vector and the score of the weight parameter based on the weight prediction model, and obtaining the contribution value of each weight parameter to the weight of the edge, and adding the weight parameter with the highest corresponding contribution value and the corresponding score ratio to the corresponding nodes in the search vector, and adding each search vector to the second graph.

[0029] Specifically, as the new legal data increases, the above-mentioned second knowledge graph will also become larger and its content will increase. It will become difficult for users to quickly obtain the core content or target content of the corresponding legal cases. To solve this problem, each node in the above-mentioned second knowledge graph and the adjacent nodes connected by the corresponding edges are transformed into the above-mentioned search vectors through the above-mentioned semantic transformation model. Also, based on the above-mentioned weight prediction model, the weight of each weight parameter corresponding to the edges between the nodes in the above-mentioned search vectors and the score of each above-mentioned weight parameter are obtained. The product of the weight of each above-mentioned weight parameter and the corresponding score is used as the contribution value of the above-mentioned weight parameter to the edge weight. The above-mentioned weight parameter with the largest contribution value is used as the key parameter and embedded between the corresponding nodes. Also, the score corresponding to the above-mentioned weight parameter is added to the above-mentioned search vectors. Through the above technical solution, it not only provides convenience for users to search for target content from the above-mentioned second knowledge graph, but also enables the acquisition of the content of relevant information between nodes during the search process, thereby improving the accuracy of obtaining the above-mentioned target content.

[0030] Further, after the step S4, there is also a step S5: Statistically analyze the search information of users at a preset time interval, obtain the search nodes from the search information, and display different colors according to the search times of the search nodes. The more the search times, the brighter the color, and vice versa, the darker the color. Also, obtain the core weight parameters with the largest weight contribution values of the edges corresponding to the N search nodes with the largest search times, increase the corresponding weights of the core weight parameters in the weight prediction model, and correspondingly decrease the weights of the other weight parameters corresponding to the edges. Then, calculate the weights of the edges through the weight prediction model and the adjusted weight parameter model, and update the second knowledge graph.

[0031] Specifically, by statistically analyzing the above-mentioned search information of users at the above-mentioned preset time interval, the information of the nodes that users are concerned about can be obtained in a timely manner. Different colors are displayed to highlight the degree of attention of users to different nodes. The nodes with higher attention are set with brighter colors, and vice versa, the nodes with lower attention are set with relatively darker colors, so as to facilitate users to quickly understand the case relationships and content. Also, according to the core weight parameters with the largest weight contribution values of the edges corresponding to the above-mentioned search nodes, the weights of the above-mentioned core parameters in the above-mentioned weight prediction model are increased, and the weights of the other weight parameters corresponding to the above-mentioned edges in the above-mentioned weight prediction model are correspondingly adjusted. Then, the weights of the above-mentioned edges are recalculated through the above-mentioned weight prediction model, and the above-mentioned second knowledge graph is updated. Through the above technical solution, it not only facilitates users to quickly understand the core content of legal cases, but also can dynamically adjust the above-mentioned second knowledge graph according to the user's focus points to meet the attention requirements of users at different stages.

[0032] Further, the machine learning model is a neural network model trained with the correlation relationships between each core node and other nodes in different types of legal cases.

[0033] Further, N ranges from a positive integer greater than or equal to 1 and less than M, where M is the total number of the search nodes.

[0034] The present invention also provides a dynamic update system for a legal case relationship graph based on a neural network, which is used to implement the above method. As Figure 4 shown, the system includes: An acquisition unit, configured to acquire original legal case data, preprocess the acquired original legal case data, and obtain a plurality of target legal data; A generation unit, configured to extract a plurality of entities in the target legal data as nodes, obtain a node subgraph based on all the nodes and the machine learning model, select the weight parameters corresponding to the nodes from the weight parameter subgraph, obtain the weights of the node edges based on the weight parameters and the weight prediction model, and use the weight parameter subgraph and the node subgraph as the first graph; A fusion unit, configured to obtain new legal data, extract new nodes from the new legal data, obtain an added subgraph based on the new nodes and the machine learning model, and fuse the added subgraph with the first graph to obtain a second graph; An embedding unit, configured to generate search vectors based on adjacent nodes in the second graph and key weight parameters of the edges between the adjacent nodes, and embed the search vectors into the second graph.

[0035] The present invention also provides a computer-readable storage medium, on which instructions are stored. The instructions are characterized in that when being executed by a processor, the above method is implemented.

[0036] In summary, after preprocessing and entity extraction of the above-mentioned original legal case data, the present invention generates a node subgraph by combining semantic analysis and a machine learning model, and quantifies the strength of the relationship between nodes through a weight parameter subgraph to form an initial graph, i.e., the first graph. This process transforms unstructured legal texts into a structured relationship network, solves the problems of fuzzy entity relationships and difficult-to-quantify weights in traditional methods, and significantly improves the interpretability of the graph. Secondly, for newly added legal data, new nodes are dynamically fused with the existing graph through similarity matching and a weight prediction model, avoiding the lag of static graphs. At the same time, the integrity of the graph structure is ensured through an adaptive weight allocation mechanism, enabling the system to flexibly handle the diversity and complexity of legal cases. In addition, a search vector embedding key weight parameters is generated through a semantic conversion model, and the weight prediction model is dynamically optimized in combination with the user's search behavior, achieving a double improvement in retrieval efficiency and result accuracy. The user's high-frequency search nodes and associated weight parameters are fed back to the model in real time to drive the dynamic adjustment of the weights of the second graph, enabling the system to adaptively optimize according to actual needs and meet the needs of legal practitioners for timeliness, accuracy, and personalized retrieval. Through the mutual cooperation of the above technical solutions, the problems of complex legal case relationships and difficult updates are effectively solved, providing efficient and reliable knowledge support for legal research, case analysis, and judicial decision-making.

[0037] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0038] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0039] As described above, 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 foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamically updating a legal case relationship graph based on a neural network, characterized in that: The following steps are involved: Step S1: collecting original legal case data, preprocessing the collected original legal case data, and obtaining a plurality of target legal data; Step S2: extract multiple entities in the target legal data as nodes, obtain a node subgraph based on all the nodes and the machine learning model, select weight parameters corresponding to the nodes from the weight parameter subgraph, obtain the weights of the node edges based on the weight parameters and the weight prediction model, and use the weight parameter subgraph and the node subgraph as the first graph; Step S3: acquiring newly added legal data, extracting new nodes from the newly added legal data, acquiring a newly added subgraph based on the new nodes and the machine learning model, and fusing the newly added subgraph with the first graph to acquire a second graph; Step S4: Generate a search vector based on key weight parameters of adjacent nodes and edges between adjacent nodes in the second graph, and embed the search vector into the second graph.

2. The method according to claim 1, characterized in that The step S4 further includes a step S5: The user's search information is counted at preset time intervals, and the search nodes are obtained from the search information. Different colors are displayed according to the number of searches of the search nodes, wherein the more searches, the brighter the color, and vice versa. The core weight parameter with the largest weight contribution value of the edge corresponding to the N search nodes with the largest number of searches is also obtained, and the corresponding weight of the core weight parameter in the weight prediction model is increased, and the weights of other weight parameters corresponding to the edge are correspondingly reduced, and the weight corresponding to the edge is calculated through the weight prediction model and the adjusted weight parameter model, and the second graph is updated.

3. The method for dynamically updating the legal case relationship graph based on a neural network according to claim 1 is characterized in that: The acquisition of the node subgraph includes: Through semantic analysis, multiple entities are extracted from the target legal data as nodes, and according to the elements of the legal case, multiple core nodes in the nodes are determined, and based on each core node, a sub-graph including related nodes that have a correlation with the core node is obtained through a machine learning model, and the analysis unit is used according to the correlation between the nodes of different sub-graphs, and based on the correlation, different sub-graphs are fused to obtain the node sub-graph.

4. The method for dynamically updating the legal case relationship graph based on a neural network according to claim 1 is characterized in that: The acquisition of the first atlas includes: The weight parameters corresponding to each of the nodes in the node subgraph are selected from the weight parameter subgraph and connected. According to the text information corresponding to the node in the target legal data and the scoring criteria corresponding to the weight parameters, the score of each weight parameter corresponding to the node is obtained, and the node, the score corresponding to each weight parameter and the weight prediction model are used to obtain the weight value of the edge corresponding to the node, and the weight parameter score and the weight value are added to the edge of the corresponding node, and the weight parameter subgraph, the node subgraph and the connection relationship between the two are used as the first graph.

5. The method for dynamically updating the legal case relationship graph based on a neural network according to claim 1 is characterized in that: Acquiring the second atlas includes: Extract the new node from the newly added legal data, obtain the connection node with the new node through semantic analysis, connect the new node with the corresponding connection node, select the corresponding weight parameter from the weight parameter subgraph and connect, when the new node cannot obtain the corresponding weight parameter from the weight parameter subgraph, calculate the similarity between the new node and each node in the first graph, and use the weight parameter corresponding to the node with the highest similarity as the target weight parameter of the new node, calculate the weight of the corresponding edge of the new node through the method of step S2, add the weight of the new node to the corresponding edge, and obtain the second graph.

6. The method according to claim 1, characterized in that The generation of the search vector includes: Each node in the second graph and the adjacent nodes connected by the corresponding edge are converted into a search vector through a semantic conversion model, and the weight of each weight parameter corresponding to the edge between the nodes in the search vector and the score of the weight parameter are obtained based on the weight prediction model to obtain the contribution value of each weight parameter to the weight of the edge, and the weight parameter with the highest corresponding contribution value is added between the corresponding nodes in the search vector, and each search vector is added to the second graph.

7. The method according to claim 1, characterized in that The machine learning model is a neural network model that has been trained on the correlation between each core node and other nodes in different types of legal cases.

8. The method according to claim 6, characterized in that The value range of N is a positive integer greater than or equal to 1 and less than M, where M is the total number of the search nodes.

9. A system for dynamically updating a legal case relationship graph based on a neural network, used to implement the method according to any one of claims 1 to 8, characterized in that: The system comprises: A collection unit, used for collecting original legal case data, preprocessing the collected original legal case data, and obtaining a plurality of target legal data; A generating unit, configured to extract multiple entities in the target legal data as nodes, obtain a node subgraph based on all the nodes and a machine learning model, select a weight parameter corresponding to the node from a weight parameter subgraph, obtain a weight of a node edge based on the weight parameter and a weight prediction model, and use the weight parameter subgraph and the node subgraph as a first graph; A fusion unit, used to obtain newly added legal data, extract new nodes from the newly added legal data, obtain a newly added subgraph based on the new nodes and the machine learning model, and fuse the newly added subgraph with the first graph to obtain a second graph; An embedding unit is used to generate a search vector based on key weight parameters of adjacent nodes and edges between adjacent nodes in the second graph, and embed the search vector into the second graph.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images