Legal element analysis method and system based on large language model and knowledge graph

Through the combination of large language model and knowledge graph, the knowledge graph of legal requirements is dynamically constructed and updated, and the problem of poor capture of the complexity of specific cases in the existing technology is solved, and efficient and accurate analysis of legal requirements and Quick response.

CN119046476APending Publication Date: 2024-11-29NANJING WEIQUE INFORMATION TECH CO LTD +1

Patent Information

Application Number
CN202411525294.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-30
Publication Date
2024-11-29

Smart Images

  • Figure CN119046476A_ABST
    Figure CN119046476A_ABST
Patent Text Reader

Abstract

The invention discloses a legal element analysis method and system based on a large language model and a knowledge graph, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining a legal element of a target case, carrying out the information extraction through the large language model, and generating knowledge graph construction information containing entity information and relation information; and constructing a knowledge graph of the legal elements of the target case based on the construction information. And obtaining analysis information of the user for the legal elements of the target case, and performing second information extraction in the constructed knowledge graph by using the large language model so as to generate an analysis result. In addition, through a cache mechanism, processing of legal terms and judgment paths of high-frequency access in the analysis process is optimized, and the response efficiency of the system is improved through an elimination strategy. According to the method, the deep semantic understanding ability of the large language model and the legal logic analysis ability of the knowledge graph are effectively combined, and the accuracy and processing efficiency of legal analysis can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to a legal requirements analysis method and system based on a large language model and knowledge graph. Background Art

[0002] With the advancement of science and technology and the development of artificial intelligence, legal assistance tools have been widely used. For example, the Chinese patent with the authorization announcement number CN117743590B discloses a legal assistance method and system based on a large language model. The method constructs legal large language model prompt words based on user input questions to ensure that the pre-trained large language model can better understand the user input questions; constructs a legal knowledge graph based on the acquired legal corpus, so that the pre-trained large language model can better understand the relationship between legal entities and legal concepts; and queries the legal knowledge graph based on the pre-trained large language model according to the legal large language model prompt words, and uses the powerful knowledge acquisition and context analysis capabilities of the pre-trained large language model to quickly and accurately obtain professional answers to user input questions.

[0003] The above methods have some shortcomings. For example, the existing solutions rely on users to input questions to construct prompt words and query the knowledge graph based on the prompt words. This process has high requirements for the data content of the knowledge graph. The general knowledge graph is not effective for specific cases and may not fully capture the complexity of specific cases, especially in the context of polysemous words, legal terms or special cases. In addition, the construction and query of knowledge graphs in the existing technology are relatively static, and fail to dynamically adjust the structure or content of the knowledge graph according to the user's analysis questions and case background. Summary of the invention

[0004] In response to the deficiencies in the prior art, this application discloses a legal requirements analysis method and system based on a large language model and knowledge graph.

[0005] In the first aspect, the present application provides a legal requirements analysis method based on a large language model and a knowledge graph, including: Obtain the legal requirements of the target case, use the large language model to extract the first information, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; Based on the knowledge graph construction information, a target knowledge graph corresponding to the legal requirements of the target case is constructed; Obtain analysis information of the target user on the legal requirements of the target case; based on the analysis information, use the large language model to perform second information extraction in the target knowledge graph to generate analysis result information.

[0006] As an optional implementation manner, the generating analysis result information further includes: Counting the legal clauses and precedent paths whose access times are greater than a first threshold within a preset time window, and determining the legal clauses and precedent paths with high frequency of access; caching the frequently accessed legal clauses and case paths with a preset cache priority; When generating analysis result information, firstly, frequently accessed legal clauses and precedent paths are retrieved from the cache; Counting the legal clauses and precedent paths whose access times within a preset time window are less than a first threshold, and determining the legal clauses and precedent paths with low frequency of access; The legal clauses and case paths that are accessed less frequently are cleared using a first elimination strategy; wherein the first elimination strategy includes: a least recently used strategy and a least frequently used strategy.

[0007] As an optional implementation, the step of obtaining the legal requirements of the target case, extracting the first information using a large language model, and generating knowledge graph construction information includes: Determine the type of elements of the legal elements of the target case; Based on the element type of the legal element of the target case and the historical data of elements of the same element type in a preset legal element database, determine the legal clauses and precedent paths corresponding to the legal element of the target case; Based on the knowledge graph construction information and the legal clauses and precedent paths corresponding to the legal requirements of the target case, construct a target knowledge graph corresponding to the legal requirements of the target case; Among them, the types of requirements include: criminal cases, intellectual property cases, and civil cases.

[0008] As an optional implementation, before using the large language model to extract the first information, the method further includes: Select the information extraction mechanism based on the data size and task complexity of the legal requirements of the target case; The task complexity is determined based on the sentence length, number of terms and density of polysemous words in the legal requirements of the target case.

[0009] As an optional implementation, the information extraction mechanism selected based on the data size and task complexity of the legal requirements of the target case includes: In response to the data size of the legal requirements of the target case being greater than a first preset threshold, or the task complexity being greater than a second preset threshold, a preprocessing step is initiated; wherein the preprocessing step includes: preliminary screening of entity information and relationship information; In response to the data size of the legal requirements of the target case being less than or equal to the first preset threshold, and the task complexity being less than or equal to the second preset threshold, the large language model is directly used to extract entity information and relationship information.

[0010] As an optional implementation, the preprocessing step includes: Generate a global word frequency and inverse document frequency weight for each entity by calculating the word frequency and inverse document frequency of each entity in the legal requirements of the target case; Based on the global word frequency and inverse document frequency weights, entity information with weights higher than a preset threshold is screened out as preliminary screening entity data; Extract the relationship information between entities through context dependency parsing and entity co-occurrence frequency; Based on the co-occurrence frequency of the entities and the strength of the dependency relationship, the relationship information that meets the preset threshold is screened out as the initial screening relationship data; The preliminary screening entity data and preliminary screening relationship data are input into a large language model for deep entity extraction and relationship construction.

[0011] As an optional implementation, it also includes: The entity information and relationship information in the legal requirements of the target case are segmented into logical paragraphs; and the word frequencies of the entity information and relationship information in each logical paragraph are calculated respectively; Generate local word frequency and inverse document frequency weights for each logical paragraph; Combine the local word frequency and inverse document frequency weights with the global word frequency and inverse document frequency weights to generate comprehensive word frequency and inverse document frequency weights; The comprehensive word frequency and inverse document frequency weights are used to optimize the deep extraction of entity information and relationship information by a large language model.

[0012] Secondly, this application also provides a legal requirements analysis system based on a large language model and knowledge graph, including: An extraction module is used to obtain the legal requirements of the target case, use the large language model to perform first information extraction, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; A construction module, used to construct information based on the knowledge graph to construct a target knowledge graph corresponding to the legal requirements of the target case; The interactive module is used to obtain the target user's analysis information on the legal requirements of the target case; based on the analysis information, the large language model is used to perform second information extraction in the target knowledge graph to generate analysis result information.

[0013] In a third aspect, an optional implementation of the present application further provides a computer device, comprising: a processor and a memory, the memory storing machine-readable instructions executable by the processor, the processor being used to execute the machine-readable instructions stored in the memory, and when the machine-readable instructions are executed by the processor, the machine-readable instructions perform the steps in the above-mentioned first aspect, or any possible implementation of the first aspect.

[0014] In a fourth aspect, an optional implementation of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are executed.

[0015] Compared with the prior art, the present invention has the following beneficial effects: it can automatically extract entity and relationship information through a large language model, quickly build a knowledge graph corresponding to the legal requirements of the target case, and dynamically update it as the case changes. By using the structured information of the knowledge graph, the accuracy and response speed of problem analysis are improved.

[0016] By presetting the cache priority mechanism, the present invention can prioritize the retrieval of frequently accessed legal clauses and case paths, reducing the time overhead of repeated queries. In particular, when dealing with multiple related cases, the cache mechanism accelerates the access to frequently accessed clauses, significantly improving the response speed of the system.

[0017] The present invention also intelligently selects information extraction mechanisms based on the data size and task complexity of the target case's legal requirements. For cases with large data size or high task complexity, the system will enable preprocessing steps to reduce the computational burden of the large language model; for simpler cases, the large language model is directly used for analysis, thereby improving the overall processing efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a legal requirements analysis method based on a large language model and a knowledge graph provided in an embodiment of the present application; Figure 2 A flowchart of a cache content adjustment method provided in an embodiment of the present application; Figure 3 A schematic diagram of a legal requirements analysis system based on a large language model and knowledge graph provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0020] To facilitate understanding of this embodiment, the legal requirements analysis method based on a large language model and a knowledge graph disclosed in the embodiment of this application is first introduced in detail. The execution subject of the legal requirements analysis method based on a large language model and a knowledge graph provided in the embodiment of this application is generally a computer device with a certain computing capability, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the legal requirements analysis method based on a large language model and a knowledge graph can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0021] The present invention reduces the reliance on large-scale annotated data by combining a large language model with a knowledge graph, significantly improving the efficiency of building and updating the knowledge graph. By utilizing the deep semantic understanding capability and knowledge reasoning function of the large language model, semantic analysis of text information is achieved, and verification and supplementation are performed through the knowledge graph, thereby improving the accuracy and completeness of knowledge extraction. By integrating the powerful semantic analysis capability of the large language model with the dynamic reasoning capability of the knowledge graph, complex legal documents can be flexibly processed and intelligently analyzed.

[0022] See also Figure 1 As shown, it is a flowchart of a legal requirement analysis method based on a large language model and a knowledge graph provided in an embodiment of the present application, and the method includes steps S101 to S103, wherein: S101: Obtain the legal requirements of the target case, use the large language model to extract the first information, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; S102: constructing a target knowledge graph corresponding to the legal requirements of the target case based on the knowledge graph construction information; S103: Obtain analysis information of the target user on the legal requirements of the target case; based on the analysis information, use the large language model to perform second information extraction in the target knowledge graph to generate analysis result information.

[0023] In specific implementation, the present disclosure first obtains the legal requirements of the target case. The legal requirements of the target case can be obtained in one of the following two ways: User input: Users directly input the legal requirements involved through the system interface, including the core issues of the case, relevant legal provisions, precedents, and basic background information of the case.

[0024] Automatic document parsing: Through natural language processing technology, relevant legal information can be automatically parsed from legal documents or files uploaded by users. This process includes technical steps such as keyword extraction and entity recognition to obtain valuable legal information.

[0025] Input the legal requirements of the target case into a pre-trained large language model, such as GPT. The large language model will perform semantic analysis on the input text, identify entity information such as legal clauses, important people in the case, events, and extract the relationship between these entities. Entity information includes: legal clauses, roles of parties in the case, etc., while relationship information includes the association between entities, such as the reference relationship of legal clauses, the applicable relationship of precedents, etc.

[0026] Based on the extracted entity information and relationship information, the target knowledge graph is constructed. Each node in the knowledge graph represents a legal entity (such as a legal clause, a case, a party), and each edge represents the relationship between entities (such as a reference relationship, a case support relationship). The knowledge graph can be stored and managed using a graph database (such as Neo4j or ArangoDB), and efficient management and retrieval of entities and relationships can be achieved through a graph database query language (such as Cypher).

[0027] After obtaining the user's analysis information, the user's question is directly analyzed and semantically parsed through the large language model. The analysis question entered by the user (for example: "In this case, how should the liability for breach of contract in the contract be determined?") will be input into the large language model, and the model will conduct an in-depth semantic analysis of the question to identify the key legal entities and concepts in the question. The large language model first parses the semantic structure of the user's question and extracts important keywords and legal-related entity information (such as "liability for breach of contract", "contract", etc.). Based on these extracted entities, the semantic parsing results of the user's question are then associated with the constructed knowledge graph. Through the model's powerful semantic understanding capabilities, the large language model can identify the potential association between user questions and corresponding entities and relationships in the knowledge graph without relying on explicit keyword matching.

[0028] Regarding S101 above: In the specific implementation, the large language model is used to perform the first stage of information extraction (i.e., first information extraction) on the input legal text. In this process, the large language model performs semantic analysis through preset prompt words or specially tuned prompt word templates to identify entity information (such as legal terms, case parties, etc.) and relationship information (such as the relationship between contract terms and precedents).

[0029] For example, the prompt words are "Please extract the legal provisions and case studies related to breach of contract liability." or "Identify the legal provisions and case studies related to Party A in the contract." It is understandable that all the set prompt words can be automatically interacted with the large language model through scripts.

[0030] Regarding S102 above: After the big language model extracts the entity information and relationship information in the legal file, a dedicated target knowledge graph is generated for the file. For example, in a specific contract dispute case, the knowledge graph will be built specifically for the case, including the core legal entities and corresponding relationships in the case. The files and corresponding knowledge graphs for each case can be completely independent. Different cases will have their own independent nodes and edges to ensure that the information between files is not confused. For a new case, the system will build a new knowledge graph from scratch without affecting the existing case graph.

[0031] In addition, the knowledge graph structure of each case file is dynamically generated according to the case content. For example, a node can represent "legal clause", "precedent", "party", "event", etc. For example, "Criminal Law Article 233" is a legal clause node, and "Precedent A" is a precedent node. Edges can represent relationships such as "reference", "application", and "participation". For example, there may be an applicable relationship between a party and a legal clause, which will be reflected in the knowledge graph through edges.

[0032] In a specific implementation, the query and insertion interface of the graph database can be used to insert the extracted entities and relationships into the graph in the form of nodes and edges.

[0033] Regarding S103 above: In specific implementations, an interactive interface (such as a web interface or an App interface) can be provided to users, where they can ask legal questions about the case through an input box or voice input. These questions can be open-ended or specific questions about legal provisions, precedents, or case details.

[0034] The analysis question entered by the user will be directly passed to the large language model (such as pre-trained models such as GPT). The large language model can automatically parse the natural language text entered by the user and identify the core legal terms and the focus of the question. The large language model performs semantic analysis on the text entered by the user based on its built-in pre-trained semantic analysis capabilities. The model can identify key entities in the question (such as "plaintiff", "right to terminate the contract", "Article 52 of the Contract Law") and infer the user's intention. For example, the model may recognize that the user is asking about legal terms and precedents related to the right to terminate the contract.

[0035] After analyzing the user's analysis question, the large language model will perform secondary information extraction (i.e., second information extraction) in combination with the knowledge graph of the target case. The large language model will generate vector representations (embeddings) for the above keywords. These vector representations contain not only the surface meaning of the words, but also the deep semantic relationships. Through these vector representations, the model can better understand the user's legal issues and match them with the nodes in the knowledge graph.

[0036] Each legal entity in the knowledge graph (such as "Article 52 of the Contract Law", "Case A", "Plaintiff", etc.) has a corresponding semantic vector representation. Semantic matching of keywords can be performed through the cosine similarity between the word vectors generated by the large language model and the node vectors of the knowledge graph, or through the query interface of the graph database such as Neo4j. For example, the nodes in Neo4j can be pre-associated with the word vectors of the large language model, and then the legal entity nodes most relevant to the keywords can be retrieved through the query language of the graph database (such as Cypher).

[0037] After identifying the legal entity nodes related to the problem, it is necessary not only to return these nodes, but also to retrieve the edges related to these nodes in the knowledge graph, that is, the relationships between entities. Relationship edges usually represent the relevance of entities to other legal clauses, precedents, or interpretations. For example, using the relationship retrieval function of the graph database, obtain the relationship related to the target node (such as "Article 52 of the Contract Law") through an API or query language (such as Cypher).

[0038] The large language model uses the expansion of the context window and the multi-head attention mechanism to deeply analyze the semantics of the question and generate corresponding legal advice. The logic behind this is that the model understands the core of the user's question, combines all relevant nodes and relationship information in the knowledge graph, and infers the legal advice that the user is most likely to need.

[0039] After combining the nodes and relationships retrieved from the knowledge graph, the large language model will generate the final legal advice based on the user's original analysis question. For example, for "Article 52 of the Contract Law", the user may be advised: "According to Article 52 of the Contract Law, the plaintiff has the right to terminate the contract, and according to Case A, the court supports the plaintiff's request in similar circumstances." See also Figure 2 , Figure 2 This is a flow chart of a cache content adjustment method provided in an embodiment of the present application. As an optional implementation, when generating the analysis result information, the cache content can also be adjusted, including steps S201 to S204: S201: Counting the legal clauses and precedent paths whose access times are greater than a first threshold within a preset time window, and determining the legal clauses and precedent paths with high frequency of access; S202: caching the frequently accessed legal clauses and case paths with a preset cache priority; S203: when generating analysis result information, firstly retrieve frequently accessed legal clauses and precedent paths in the cache; S204: Count the legal clauses and case paths whose access times within a preset time window are less than a first threshold, and determine the legal clauses and case paths that are accessed less frequently; and remove the legal clauses and case paths that are accessed less frequently using a first elimination strategy; wherein the first elimination strategy includes: a least recently used strategy and a least frequently used strategy.

[0040] Among them, the precedent path refers to the court judgment cases related to the current analysis case identified from previous cases and the legal clauses and basis cited by them. These precedents can be obtained through legal retrieval systems or other legal databases, and will contain similar case judgments related to the current legal requirements. The precedent path includes not only the specific judgment content, but also the citation order of relevant legal clauses and legal logic, so it can provide users with a complete reference framework. For example, the precedent path in a case may contain information such as relevant court judgments, cited legal clauses, and relevant factual determinations. The present invention records the number of times users visit these precedents in an automated manner, and performs cache management based on the access frequency.

[0041] In a specific implementation, a preset time window is first set to count the number of times users visit certain legal clauses and precedent paths. The preset time window can be set to one day, one week, one month or other suitable time range according to the needs of the specific system. Within this time window, the number of times users visit specific legal clauses and precedent paths is counted. In addition, a first threshold is set to distinguish between high-frequency and low-frequency access clauses and paths. The threshold can be adjusted based on historical access data or system resource load conditions. For example, if the number of visits to a legal clause or precedent path exceeds a set threshold, such as 10 times, then the clause or path is considered to be a high-frequency access clause, otherwise it is a low-frequency access.

[0042] In addition, the threshold can be adjusted dynamically based on user behavior. For example, if the analysis of a certain type of case usually requires accessing the same legal clause multiple times, the threshold can be increased accordingly; for simple case queries, the threshold can be set lower. Different types of cases can have different thresholds. Complex cases may have a higher threshold, such as 10 visits, while simple cases may have a lower threshold, such as 2 visits.

[0043] After the statistics are completed, the legal clauses and case paths with access times greater than the first threshold are marked and set as high-priority caches. The cache priority can be set based on the importance of the clause or path, the frequency of user queries, and the performance requirements of the system. For example, for frequently accessed legal clauses, they are stored in the cache for quick retrieval during subsequent analysis. The cache priority can be divided into multiple levels, which can usually be dynamically adjusted based on the number of accesses. For example, the legal clauses with the most access times will be assigned the highest cache priority.

[0044] When a user initiates a new legal analysis, the frequently accessed legal terms and case paths are first retrieved from the cache. This means that the system does not have to re-query the knowledge graph every time, which greatly improves the response speed. The retrieval method of cached data is based on the previously set cache priority. The system will check the high-priority terms in the cache in turn to ensure a quick response to the user's analysis needs.

[0045] For legal clauses and case paths whose statistical results show that the number of accesses within the preset time window is less than the first threshold, they are marked as low-frequency access clauses. To ensure the reasonable allocation of cache resources, they can be cleared based on one of the following two elimination strategies: Least Recently Used (LRU) strategy: removes legal terms or case paths that have not been accessed again for a long time. This strategy ensures that only terms frequently used by recent users are retained in the cache.

[0046] Least Frequently Used (LFU) policy: prioritizes deleting legal terms and paths with the lowest access frequency, ensuring that frequently accessed data can stay in the cache longer.

[0047] In the specific implementation, cache cleaning is performed regularly, and low-frequency access terms and paths are cleared according to the LRU or LFU strategy to ensure that the cache space can be effectively utilized.

[0048] In this way, by retrieving frequently accessed legal clauses from the cache first, the response time of user analysis is significantly shortened, improving the user experience of the system. Through the LRU and LFU strategies, limited cache resources can be reasonably allocated to prevent invalid data from occupying cache space for a long time. The cache data is dynamically adjusted according to the actual access frequency of users to ensure that the system can better respond to changing analysis needs.

[0049] It should be noted that the cache mechanism can be designed as a global cache system that runs independently from the knowledge graph, mainly to optimize the frequently accessed clauses and precedent paths in the entire legal analysis system. At this time, the cache is not limited to the knowledge graph of a certain case, but a globally shared cache. For example, when multiple cases involve similar legal clauses, the system will cache these frequently accessed clauses, and there is no need to repeatedly extract them from the knowledge graphs of different cases each time, thereby improving the response speed of the overall system.

[0050] Another way is that the cache mechanism is a local cache space created independently for the knowledge graph of each case. The knowledge graph corresponding to the legal requirements of each target case has a separate cache space. In this way, when the user queries the same case clause or precedent path multiple times, the system will respond quickly from the local cache of the case. This method can better improve the analysis efficiency of specific cases and avoid competition and conflicts in the global cache system.

[0051] In terms of specific implementation, more efficient query response can be achieved through the joint scheduling of cache strategy and knowledge graph. For example, when a user initiates a query, it first checks whether there is a corresponding knowledge graph for the case queried by the user. If there is a knowledge graph, the knowledge graph will work together with the cache mechanism to prioritize the retrieval of high-frequency legal clauses or precedent paths in the cache. If not in the cache, continue to extract information from the knowledge graph.

[0052] If it is a globally shared cache space, it will first check whether the global cache has stored the relevant terms or precedents. If so, the result will be returned directly. If it does not exist in the cache, it will need to be retrieved in the target knowledge graph, and then these frequently accessed data will be updated to the global cache to enable faster response to subsequent queries.

[0053] By combining the cache mechanism with the knowledge graph, the present invention can effectively improve the response speed of the large language model and the knowledge graph analysis method, especially when processing multiple repeated queries. The cache mechanism can reduce repeated data extraction work and improve the overall processing efficiency of the system, while the knowledge graph ensures the professionalism and comprehensiveness of the analysis content. The combination of the two enables the system to provide efficient legal analysis services while ensuring that the analysis results of complex cases are highly accurate.

[0054] As an optional implementation, the method of obtaining the legal requirements of the target case, using a large language model to perform first information extraction, and generating knowledge graph construction information includes: determining the requirement type of the legal requirements of the target case; determining the legal clauses and precedent paths corresponding to the legal requirements of the target case based on the requirement type of the legal requirements of the target case and the historical data of requirements of the same requirement type in the legal requirement database; constructing a target knowledge graph corresponding to the legal requirements of the target case based on the knowledge graph construction information and the legal clauses and precedent paths corresponding to the legal requirements of the target case; wherein the requirement types include: criminal cases, intellectual property cases, and civil cases.

[0055] The knowledge graph corresponding to the legal requirements of the target case in the present disclosure can also be combined with additional legal requirements databases when constructing, so as to form a more complete and richer target knowledge graph.

[0056] In specific implementation, the type of legal requirements of the target case is one of the core of case analysis. First, it is necessary to clarify the type of legal requirements of the target case, including criminal cases, intellectual property cases, civil cases, etc. In this step, the large language model will analyze the input text of the legal requirements of the target case and use the preset prompt words (PromptEngineering) to identify which type the legal requirements belong to.

[0057] For example, similar to "determine whether this case is a criminal case, an intellectual property case, or a civil case." When the large language model parses the legal requirements text of the target case and identifies relevant legal clauses, case terms (such as "patent", "infringement", "criminal law"), etc., it will automatically classify the case type.

[0058] For example, for input text containing the keywords "criminal law" and "robbery", the system will automatically classify the case as a criminal case; while for input text containing the keywords "patent" and "infringement", it will be classified as an intellectual property case.

[0059] Next, based on the historical data of the legal requirements database and combined with the case requirements type, the legal provisions and precedent paths corresponding to the target case type are screened out.

[0060] The legal requirements database pre-stores legal clauses and precedent information related to different types of cases. For example, criminal cases store commonly used criminal law clauses and corresponding precedents, while intellectual property cases store clauses and precedents such as patent law and trademark law. Using the large language model, legal clauses and precedents that match the case type can be retrieved from the legal requirements database. Through the contextual analysis capabilities of the large language model, the model can determine the relevant clauses and precedent paths for similar cases based on the background of historical cases.

[0061] In the specific implementation, based on the legal clauses and precedent paths extracted from historical data, we start to build a knowledge graph of the legal elements of the target case. The construction of the knowledge graph requires converting the above legal clauses and precedent paths into nodes and edges.

[0062] For example, in the construction of nodes and edges, each legal clause and precedent path will be displayed as a node in the knowledge graph, and the relationship between different nodes is represented by edges. The specific relationship can be based on the judge's reference to the legal clause in the precedent, or the reference relationship between legal clauses.

[0063] It should be noted that the construction here is a supplement to the knowledge graph construction in the above text that is not combined with the legal requirements database.

[0064] In this way, the generated knowledge graph not only contains the legal logic of the case, but also integrates judicial interpretations and precedent resources that are closely related to the case type.

[0065] As an optional implementation method, the legal requirements database itself can also be a knowledge graph. For example, each field such as intellectual property, criminal cases, and civil cases can have a large domain knowledge graph. This domain knowledge graph stores public information such as common legal clauses, relevant precedents, and legal interpretations.

[0066] The case knowledge graph is a small subgraph, namely the target knowledge graph mentioned above, which represents the personalized information such as legal clauses and precedents involved in a specific case. These small case knowledge graphs can be nested in the large domain knowledge graph, pointing to related domain knowledge through links, or directly linking nodes (such as legal clauses and cases).

[0067] In the big knowledge graph, the domain knowledge graph stores common legal knowledge in the industry, such as relevant legal clauses, relevant classic cases, interpretations, legal principles, etc. The specific case graph stores the entities and relationships of each case and establishes a relationship with the big knowledge graph.

[0068] When the large language model parses the case information, it will first create a sub-knowledge graph for the case, and then search for the legal clauses and precedents involved in the case in the domain knowledge graph, and then associate them with the nodes in the large knowledge graph. For example, if a case involves "patent infringement", the case knowledge graph can contain specific infringement facts and details, and find further legal basis in the intellectual property domain knowledge graph by pointing to the relevant clauses and precedents of the "Patent Law".

[0069] When users conduct analysis, they first analyze the nodes of the case knowledge graph (the target knowledge graph corresponding to a single case), such as which legal clauses, evidence, and participants are involved. Then, they further use the legal clauses, interpretations, and related precedents in the domain knowledge graph to provide more authoritative legal advice.

[0070] For example, a graph database such as Neo4j or ArangoDB can be used to store domain knowledge graphs and case knowledge graphs, and through relationship queries in the graph database, legal clauses and historical precedents related to the case can be quickly retrieved. In the graph construction process, the large language model is responsible for extracting entities and relationships from the files and legal texts input by the user, and then using the graph database API to build the graph structure.

[0071] If there are multiple cases involving the same legal terms or similar scenarios, their case knowledge graphs can be connected to the domain knowledge graph through a common node (such as a legal term).

[0072] As an optional implementation, before using a large language model to perform the first information extraction, it also includes: selecting an information extraction mechanism based on the data scale and task complexity of the legal requirements of the target case; wherein the task complexity is determined based on the sentence length, number of terms, and density of polysemous words in the legal requirements of the target case.

[0073] As an optional implementation, the information extraction mechanism selected based on the data scale and task complexity of the legal requirements of the target case includes: in response to the data scale of the legal requirements of the target case being greater than a first preset threshold, or the task complexity being greater than a second preset threshold, enabling a preprocessing step; wherein the preprocessing step includes: performing preliminary screening of entity information and relationship information; in response to the data scale of the legal requirements of the target case being less than or equal to the first preset threshold, and the task complexity being less than or equal to the second preset threshold, directly using a large language model to extract entity information and relationship information.

[0074] In the specific implementation, before information extraction, the legal requirements of the target case are first analyzed to determine the data size and task complexity of the requirements. The purpose of this step is to decide whether to directly use the large language model for information extraction or to use some preprocessing mechanism to optimize the extraction process.

[0075] By parsing the text in the legal requirements of the target case, the data size is calculated. Common metrics include: the total number of characters or words in the text, the number of pages or paragraphs in the document, and the number of legal clauses involved.

[0076] For example, longer case files or documents containing a large number of legal clauses are marked as "large data scale", while shorter case summaries are classified as "small data scale".

[0077] In addition, the complexity of the task can be evaluated based on the following parameters: Sentence length: the average length of sentences in the text. Longer sentences usually indicate more complex legal issues. The system can detect the length of each sentence and calculate the average through natural language processing tools; Number of terms: Identify legal terms, proper nouns, and technical terms in the text. Through the term base or dictionary that comes with the large language model, the system can count the number of terms that appear; Polysemy density: By analyzing the frequency of polysemous words in the text, the system can determine whether the language of a case is ambiguous. Polysemous words (such as "contract" may refer to different legal concepts) usually increase the complexity of the analysis.

[0078] For example, when a case involves a large number of terms and polysemous words, the task complexity is marked as “high”, otherwise it is marked as “low”.

[0079] Based on the above assessment of data size and task complexity, select a suitable information extraction mechanism.

[0080] For example, when the data scale is small and the task complexity is low, the large language model can be directly called to extract information. In this case, no complicated preprocessing steps are required, and the legal requirements of the entire target case can be automatically input, and the large language model directly returns entity information and relationship information. For example, the data scale is less than 5,000 words, the number of terms is less than 100, and the sentences are relatively simple (the average sentence length is less than 20 words). In this case, the system will directly pass the text to the large language model and process it using the set prompt words: For example, when the data scale is large and the task complexity is high, a preprocessing mechanism can be selected. During the preprocessing process, the system will first perform a preliminary screening of the terms, entities, and sentences in the legal requirements to reduce the processing burden of the large language model. For example, the data scale exceeds 10,000 words and the number of terms exceeds 200. You can first filter out unimportant terms through a term extraction algorithm, or reduce the amount of data through segmented processing.

[0081] In this way, by judging the data scale and task complexity, we can intelligently select the appropriate information extraction mechanism to avoid large language models directly processing large-scale and highly complex data, reducing unnecessary computational burden.

[0082] As an optional implementation, the preprocessing step includes: generating global word frequency and inverse document frequency weights for each entity by calculating the word frequency and inverse document frequency of each entity in the legal requirements of the target case; screening out entity information with weights higher than a preset threshold as preliminary screening entity data based on the global word frequency and inverse document frequency weights; extracting relationship information between entities through context dependency analysis and entity co-occurrence frequency; screening out relationship information that meets a preset threshold as preliminary screening relationship data based on the entity co-occurrence frequency and the strength of the dependency; and inputting the preliminary screening entity data and preliminary screening relationship data into a large language model for deep entity extraction and relationship construction.

[0083] In the specific implementation, the preprocessing step mainly involves filtering out entity information with higher weights through the calculation of term frequency and inverse document frequency (TF-IDF), and extracting relationship information between entities through context dependency analysis and entity co-occurrence frequency. This implementation method is mainly to optimize the input of the large language model and improve its efficiency and accuracy in extracting entities and relationships.

[0084] First, we need to calculate the term frequency and inverse document frequency (TF-IDF) value of each entity to measure its importance in the legal requirements text of the target case.

[0085] Among them, term frequency (TF) is used to measure the frequency of an entity (such as legal terms, legal terms, etc.) in the legal requirements document of the target case. For each entity, the system will count its occurrence times in the entire document and divide it by the total number of words in the document to obtain the term frequency of the entity.

[0086] For example, in a document of a contract dispute case, "Article 52 of the Contract Law" appears 10 times, and the total number of words in the document is 1000, then the word frequency of "Article 52 of the Contract Law" is 0.01.

[0087] Inverse Document Frequency (IDF) is used to evaluate the rarity of an entity in the entire document collection. If an entity appears in a large number of documents, its IDF value will be low, indicating that it has poor discriminability; conversely, entities with high IDF values ​​have high uniqueness. The present disclosure obtains its IDF value by calculating the proportion of entities appearing in all case files.

[0088] For example, if "Article 52 of the Contract Law" appears only 50 times in 1,000 documents in the system, then its IDF is log(1000 / 50) = 1.3.

[0089] The TF-IDF weight of each entity is generated by multiplying the term frequency by the inverse document frequency. The higher the weight, the higher the distinction and importance of the entity in the document. For example, the TF-IDF value of "Contract Law Article 52" is 0.013.

[0090] After calculating the TF-IDF value of each entity, based on the preset threshold, the entity information with a weight higher than the threshold is screened out as the initial screening entity data. This process ensures that the large language model only needs to process entity information that is highly relevant to the case during processing, avoiding processing irrelevant or repeated terms. The preset threshold can be adjusted according to the specific case type and system performance. For example, a TF-IDF weight threshold can be set to 0.01, and all entities below this value will not participate in subsequent analysis.

[0091] In addition to filtering entity information, it is also necessary to extract the relationship between entities in the legal requirements text of the target case. This process is achieved through context dependency parsing and entity co-occurrence frequency.

[0092] Among them, dependency parsing is a type of NLP technology that is used to analyze the grammatical relationship between words in a sentence to understand their subordinate relationships. Through this technology, the relationship between entities (usually nouns, pronouns, etc.) and other words can be identified, and then the "relationship information" in the sentence can be extracted.

[0093] In specific implementation, commonly used tools include Stanford NLP, spaCy, and AllenNLP, which can perform dependency syntactic analysis. The tool inputs a sentence, parses its grammatical structure, generates a dependency tree, and then identifies the grammatical relationship between the entities in the sentence. For example, in the sentence "The contract is signed by both parties", there is a subject-predicate relationship between "contract" and "sign", and there is an owner relationship between "both parties" and "contract". Through these tools, contextual dependencies related to entities can be extracted from the text, and these relationships can help identify the semantic associations between entities.

[0094] Entity co-occurrence frequency refers to the frequency with which multiple entities co-appear in the same piece of text (such as a sentence, paragraph, or document). By counting the number of times entities appear in the same context, the strength of the relationship between them can be measured.

[0095] In specific implementation, some open source text processing libraries or tools can be used, such as Gensim, the text processing module in scikit-learn, and the co-occurrence frequency can be counted and calculated through preprocessing steps (such as word segmentation, stop word removal, etc.). The co-occurrence frequency is used to measure the association strength between two entities in actual documents. High-frequency co-occurrence of entity pairs may mean that there is some implicit relationship between them. For example, "signed" and "effective" in a contract usually appear together.

[0096] Based on the dependency tree, the grammatical relationship between entities can be quantified, and different weights are given to different relationships (such as subject, predicate, object) to calculate the relationship strength. Based on the statistical co-occurrence frequency, a preset threshold is set to filter out entity pairs with high co-occurrence frequency. By combining the grammatical weight of dependency parsing and the co-occurrence frequency strength, entity relationships with practical significance can be more accurately filtered out. For example, only when there is both a high co-occurrence frequency and a strong grammatical dependency relationship between two entities will they be considered a valid relationship.

[0097] When faced with a large number of legal documents or complex cases, the use of preprocessing methods can effectively reduce the input data size of large language models and improve the processing efficiency of the overall system.

[0098] As an optional implementation method, it is also possible to: segment the entity information and relationship information in the legal requirements of the target case into logical paragraphs; calculate the word frequency of the entity information and relationship information in each logical paragraph respectively; generate local word frequency and inverse document frequency weights for each logical paragraph; combine the local word frequency and inverse document frequency weights with the global word frequency and inverse document frequency weights to generate comprehensive word frequency and inverse document frequency weights; and use the comprehensive word frequency and inverse document frequency weights to optimize the deep extraction of entity information and relationship information by a large language model.

[0099] In the specific implementation, the entity information and relationship information in the legal requirements of the target case are first segmented into logical paragraphs. The segmentation can be based on the semantic structure of the document or the natural paragraph division. For example, a legal document usually contains different chapters or paragraphs, and each paragraph may revolve around a specific legal issue or concept. First, these paragraphs are marked and divided, and each paragraph contains several legal entities (such as clauses, terms) and their relationship information.

[0100] In each logical paragraph, extract each entity information (such as "Article 52 of the Contract Law", "Party A", etc.) and its associated relationship information (such as "Perform contractual obligations") one by one. Calculate the word frequency of entity information and relationship information in each logical paragraph. The word frequency refers to the number of times the entity or relationship appears in the current paragraph. Then, based on these word frequencies, the system generates local word frequency and inverse document frequency weights (local TF-IDF) to measure the importance of each entity and relationship information in the current paragraph.

[0101] In addition, a global analysis is performed on the legal requirements text of the entire target case, and the global word frequency and inverse document frequency weights of each entity information and relationship information are calculated. Global TF-IDF refers to the importance assessment in the entire case file. Finally, the local and global weights are combined to facilitate the large language model to better extract and understand these entities and relationships.

[0102] Local term frequency and inverse document frequency weights (local TF-IDF) and global term frequency and inverse document frequency weights (global TF-IDF) can be combined in a weighted average or maximum selection manner to generate a composite weight. The composite weight is used to balance local and global importance. For example, a legal clause may appear frequently in a paragraph, but is relatively common throughout the case file, so the local TF-IDF is high and the global TF-IDF is low. The composite weight combines these two types of information to ensure that the large language model can fully utilize local and global importance for more accurate information extraction.

[0103] Through the above-mentioned comprehensive word frequency and inverse document frequency weights, these weights are passed as input features to the large language model. Based on these comprehensive weights, the large language model can focus on entity information and relationship information with higher comprehensive weights, and perform in-depth semantic analysis and extraction. For example, if a legal clause has a higher weight in a paragraph, the large language model will give priority to processing this entity, thereby improving the accuracy and efficiency of the large language model in the case analysis or legal analysis process.

[0104] For example, a user submits a case file involving intellectual property rights. The present disclosure first segments each paragraph in the file and extracts the legal entities (such as "Patent Law Article 42") and their relationship information that appear in each paragraph. The local word frequency of "Patent Law Article 42" in the current paragraph is calculated, and combined with the global word frequency in the entire case file to generate a comprehensive weight. Finally, the large language model can more accurately extract the core legal entities and relationship information in the case based on the comprehensive weight.

[0105] In this way, by combining local and global weights, it is possible to better distinguish important entity information in different scenarios and avoid missing key content. By filtering information with higher weights and passing it to the large language model, the processing of irrelevant information is effectively reduced, improving the analysis efficiency and performance of the model.

[0106] Based on the same inventive concept, a legal requirements analysis system based on a large language model and a knowledge graph is also provided in the embodiments of the present application. Since the principle of the system in the embodiments of the present application to solve the problem is similar to the legal requirements analysis method based on a large language model and a knowledge graph in the embodiments of the present application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0107] Reference Figure 3 As shown, Figure 3 A schematic diagram of a legal requirements analysis system based on a large language model and a knowledge graph provided in an embodiment of the present application, the system comprising: The extraction module 41 is used to obtain the legal requirements of the target case, use the large language model to perform first information extraction, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; A construction module 42, configured to construct information based on the knowledge graph to construct a target knowledge graph corresponding to the legal requirements of the target case; The interactive module 43 is used to obtain the target user's analysis information on the legal requirements of the target case; based on the analysis information, the large language model is used to perform second information extraction in the target knowledge graph to generate analysis result information.

[0108] For descriptions of the processing flow of each module in the system and the interaction flow between each module, reference can be made to the relevant descriptions in the above method embodiments, which will not be described in detail here.

[0109] The embodiment of the present application further provides a computer device, comprising: a processor and a memory; the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor performs the following steps: Obtain the legal requirements of the target case, use the large language model to extract the first information, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; Based on the knowledge graph construction information, a target knowledge graph corresponding to the legal requirements of the target case is constructed; Obtain analysis information of the target user on the legal requirements of the target case; based on the analysis information, use the large language model to perform second information extraction in the target knowledge graph to generate analysis result information.

[0110] The above-mentioned memory includes internal memory and external memory; the internal memory here is also called internal memory, which is used to temporarily store the calculation data in the processor and the data exchanged with external memory such as hard disk. The processor exchanges data with the external memory through the internal memory.

[0111] The specific execution process of the above instructions can refer to the steps of the legal requirements analysis method based on a large language model and a knowledge graph described in the embodiment of the present application, which will not be repeated here.

[0112] The present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for analyzing legal requirements based on a large language model and a knowledge graph described in the above method embodiment are executed. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0113] An embodiment of the present application also provides a computer program product, which carries a program code. The instructions included in the program code can be used to execute the steps of the legal requirements analysis method based on a large language model and a knowledge graph described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0114] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0115] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system and method can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some communication interfaces, devices or units, which can be electrical, mechanical or other forms.

[0116] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0117] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A legal requirements analysis method based on a large language model and knowledge graph, characterized in that: include: Obtain the legal requirements of the target case, use the large language model to extract the first information, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; Based on the knowledge graph construction information, a target knowledge graph corresponding to the legal requirements of the target case is constructed; Obtain analysis information of the target user on the legal requirements of the target case; based on the analysis information, use the large language model to perform second information extraction in the target knowledge graph to generate analysis result information.

2. The method according to claim 1, characterized in that The generating analysis result information also includes: Counting the legal clauses and precedent paths whose access times are greater than a first threshold within a preset time window, and determining the legal clauses and precedent paths with high frequency of access; caching the frequently accessed legal clauses and case paths with a preset cache priority; When generating analysis result information, firstly, frequently accessed legal clauses and precedent paths are retrieved from the cache; Counting the legal clauses and precedent paths whose access times within a preset time window are less than a first threshold, and determining the legal clauses and precedent paths with low frequency of access; The legal clauses and case paths that are accessed less frequently are cleared using a first elimination strategy; wherein the first elimination strategy includes: a least recently used strategy and a least frequently used strategy.

3. The method according to claim 2, characterized in that The step of obtaining the legal requirements of the target case, extracting the first information using a large language model, and generating knowledge graph construction information includes: Determine the type of elements of the legal elements of the target case; Based on the element type of the legal element of the target case and the historical data of elements of the same element type in a preset legal element database, determine the legal clauses and precedent paths corresponding to the legal element of the target case; Based on the knowledge graph construction information and the legal clauses and precedent paths corresponding to the legal requirements of the target case, construct a target knowledge graph corresponding to the legal requirements of the target case; Among them, the types of requirements include: criminal cases, intellectual property cases, and civil cases.

4. The method according to claim 3, characterized in that Before using the large language model to extract the first information, the method further includes: Select the information extraction mechanism based on the data size and task complexity of the legal requirements of the target case; The task complexity is determined based on the sentence length, number of terms, and density of polysemous words in the legal requirements of the target case.

5. The method according to claim 4, characterized in that Based on the data size and task complexity of the legal requirements of the target case, the information extraction mechanism is selected to include: In response to the data size of the legal requirements of the target case being greater than a first preset threshold, or the task complexity being greater than a second preset threshold, a preprocessing step is initiated; wherein the preprocessing step includes: preliminary screening of entity information and relationship information; In response to the data size of the legal requirements of the target case being less than or equal to the first preset threshold, and the task complexity being less than or equal to the second preset threshold, the large language model is directly used to extract entity information and relationship information.

6. The method according to claim 5, characterized in that The pre-processing step comprises: Generate a global word frequency and inverse document frequency weight for each entity by calculating the word frequency and inverse document frequency of each entity in the legal requirements of the target case; Based on the global word frequency and inverse document frequency weights, entity information with weights higher than a preset threshold is screened out as preliminary screening entity data; Extract the relationship information between entities through context dependency parsing and entity co-occurrence frequency; Based on the co-occurrence frequency of the entities and the strength of the dependency relationship, the relationship information that meets the preset threshold is screened out as the initial screening relationship data; The preliminary screening entity data and preliminary screening relationship data are input into a large language model for deep entity extraction and relationship construction.

7. The method according to claim 6, characterized in that Also includes: Processing the entity information and relationship information in the legal requirements of the target case by logical paragraphs; Calculate the word frequency of entity information and relationship information in each logical paragraph respectively; Generate local word frequency and inverse document frequency weights for each logical paragraph; Combine the local word frequency and inverse document frequency weights with the global word frequency and inverse document frequency weights to generate comprehensive word frequency and inverse document frequency weights; The comprehensive word frequency and inverse document frequency weights are used to optimize the deep extraction of entity information and relationship information by a large language model.

8. The legal requirements analysis system based on large language model and knowledge graph is characterized by: include: An extraction module is used to obtain the legal requirements of the target case, use the large language model to perform first information extraction, and generate knowledge graph construction information; wherein the knowledge graph construction information includes: entity information and relationship information; A construction module, used to construct information based on the knowledge graph to construct a target knowledge graph corresponding to the legal requirements of the target case; The interactive module is used to obtain the target user's analysis information on the legal requirements of the target case; based on the analysis information, the large language model is used to perform second information extraction in the target knowledge graph to generate analysis result information.

9. A computer device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor executes the steps of the legal requirements analysis method based on a large language model and a knowledge graph as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program. When the computer program is executed by a computer device, the computer device executes the steps of the legal requirements analysis method based on a large language model and a knowledge graph as described in any one of claims 1 to 7.

Citation Information

Patent Citations

  • Legal consultation report generation system and method based on legal knowledge graph

    CN117033721A

  • Tobacco enterprise intelligent information question and answer method based on knowledge graph and large language model

    CN117216227A

  • Law and regulation question-answering system based on generative language model and construction method

    CN117271716A

  • Legal knowledge graph construction method and device, equipment and storage medium

    CN118036736A

  • Visual development system for knowledge graph

    CN118093895A

Cited By

  • Generative multi-hop law rule chain mining and evaluating method

    CN119990135A

  • Application program authorization protocol analysis method, computer device, medium and product

    CN120234798A

  • Application authorization protocol parsing method, computer device, medium, and product

    CN120234798B

  • Intelligent information extraction system based on multi-mode adaptive analysis

    CN120449886A

  • An intelligent information extraction system based on multimodal adaptive analysis

    CN120449886B