AI-based intelligent matching method for litigation materials

Abstract

The invention relates to the technical field of intelligent matching, in particular to an AI-based intelligent matching method for litigation materials, which comprises the following steps of: performing case text analysis and part-of-speech tagging on the basis of a case database, determining legal attributes of keywords, filtering non-critical stop words, and according to the frequencies of the keywords and legal concepts, determining the legal attributes of the keywords and the legal concepts; and obtaining a case feature vector set. According to the method, similar cases can be effectively and automatically clustered through multi-dimensional space mapping and case group recognition, an obvious advantage is brought to the accuracy of information retrieval, the accuracy of case similarity evaluation is further improved through introduction of cosine similarity measurement, it is ensured that a reference case is highly matched with a current case, and a community detection algorithm is applied, so that the accuracy of information retrieval is improved. Through the establishment of the legal quotation feature set and the quotation relation graph, the understanding of the relation between the legal documents is deepened so as to accurately judge the importance and the correlation of the materials, so that the efficiency and the accuracy of the matching of the litigation materials are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent matching technology, and in particular to an AI-based intelligent matching method for litigation materials. Background Art

[0002] The field of intelligent matching technology is to use artificial intelligence technology to automatically match, classify and retrieve large amounts of text data. Its core purpose is to improve the efficiency and accuracy of data processing through intelligent algorithms, especially in situations where complex and large-scale data sets need to be processed. Intelligent matching technology has a wide range of applications in multiple industries such as law, medicine, and finance, and can assist professionals in quickly locating the required information and improving work efficiency.

[0003] Among them, the AI-based intelligent litigation material matching method, at its core, uses artificial intelligence technology to automatically identify and match relevant litigation materials, thereby helping legal professionals quickly find the evidence or reference materials needed for the case. By reducing the time required for manual material screening and improving the efficiency and effectiveness of the litigation preparation process, it can not only help lawyers and judges focus more on the legal analysis and judgment of the case, but also reduce litigation costs to a certain extent.

[0004] Traditional methods rely on manual screening and classification when processing complex and large-scale material data, which not only consumes a lot of time and labor, but is also easily affected by personal experience and subjective judgment, resulting in low efficiency and accuracy of information retrieval and matching, delays in case analysis or misjudgments, and thus affecting litigation results and work efficiency. For example, during the trial of a legal case, due to the failure to accurately match relevant legal documents or cases, lawyers spend too much time in the case preparation stage, which in turn affects the efficiency of the case and the quality of argumentation. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an AI-based intelligent matching method for litigation materials.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: an AI-based litigation material intelligent matching method, comprising the following steps:

[0007] S1: Based on the case database, case text analysis and part-of-speech tagging are performed to determine the legal attributes of keywords, filter non-critical stop words, and obtain a set of case feature vectors based on the frequency of keywords and legal concepts;

[0008] S2: Based on the case feature vector set, perform multidimensional space mapping on the case data, identify the case groups clustered in the feature space, group the cases, and generate a case clustering data set;

[0009] S3: Based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical case, and the case with the best similarity is selected as a reference to generate a case similarity evaluation record;

[0010] S4: Based on the case similarity assessment records and in combination with the legal material database, semantic analysis of the document is performed, including identification of legal terms and extraction of legal citations, and a legal citation feature set is constructed through legal terminology and citation information;

[0011] S5: Based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the correlation and influence between legal documents, and mine the legal connections between materials to generate citation network analysis information;

[0012] S6: Based on the citation network analysis information and in combination with the case similarity assessment record, by analyzing the similarity of the cases and the citation criticality of the legal materials, the legal materials associated with the current case are matched to generate a litigation material matching list.

[0013] The present invention has the following improvements: the case feature vector set includes the frequency of key legal terms, case type labels, and the number of citations of relevant regulations; the case similarity assessment record includes the reference case identification, similarity score, and analysis of key legal points of the case; the legal citation feature set includes the cited case number, the relevant legal article index, and the legal concept analysis in the citation; the citation network analysis information includes the node strength score, the core case and regulation link, and the citation path with the best correlation; the litigation material matching list includes the legal document matching list, the material correlation score, and the case application scenario proposal.

[0014] The present invention has the following improvements: based on the case feature vector set, multi-dimensional space mapping is performed on the case data, case groups clustered in the feature space are identified, and the cases are grouped. The specific steps of generating a case clustering data set are:

[0015] S201: Based on the case feature vector set, each case is represented as a point in a multidimensional space, and spatial mapping of the case data is performed to generate case mapping data;

[0016] S202: Based on the case mapping data, identify case points clustered in space, and regard the cases as groups with similar characteristics, and generate case group identification results;

[0017] S203: Based on the case group identification results, the cases are grouped, a unique identifier is assigned to each group, and key features are described to generate a case clustering data set.

[0018] The present invention is improved in that, based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical case, and the case with the best similarity is selected as a reference. The steps of generating the case similarity evaluation record are specifically as follows:

[0019] S301: Based on the case clustering data set, using the cosine similarity measurement, the feature vector of the current case is compared with the feature vector of each case in the data set, and the angle difference is calculated to generate case similarity calculation data;

[0020] S302: Based on the case similarity calculation data, a case with the best similarity is selected, and the case is determined as a reference case associated with the current case, and a screening result of similar cases is generated;

[0021] S303: Based on the screening results of similar cases, the information of the selected cases is integrated, including the case number, relevant legal clauses and judgment results, to provide a legal judgment basis for the current case and generate a case similarity assessment record.

[0022] The present invention is improved in that the cosine similarity metric is based on the formula:

[0023]

[0024] Calculate case similarity scores;

[0025] Among them, A is the feature vector of the current case, B is the feature vector of the case in the data set, W is the feature weight vector, and Θ is the adjustment coefficient.

[0026] The present invention is improved in that, based on the case similarity evaluation record, combined with the legal material database, the semantic analysis of the document is performed, including the recognition of legal terms and the extraction of legal citations, and the steps of constructing a legal citation feature set through legal proper nouns and citation information are specifically as follows:

[0027] S401: Based on the case similarity evaluation record, access the legal material database, extract the legal terms and citation information mentioned in the record, and generate a legal data extraction record;

[0028] S402: Based on the legal data extraction record, analyzing the contextual relationship between legal proper nouns and citation information, identifying the role and significance in legal argumentation, and generating legal citation analysis results;

[0029] S403: Based on the legal citation analysis results, a collection is established, including the definition of relevant legal terms, cited laws and cases, and the relationship between multiple elements, to construct a legal citation feature collection.

[0030] The present invention is improved in that, based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the correlation and influence between legal documents, and mine the legal connections between materials. The specific steps of generating citation network analysis information are:

[0031] S501: Based on the legal citation feature set, collect and organize direct citation relationships between legal documents, including citations between legal articles or cases, and summarize the information to generate a legal document citation relationship set;

[0032] S502: Based on the legal document citation relationship set, using a community detection algorithm, identifying mutual citations and one-way citations between documents, constructing a preliminary citation relationship graph between legal materials, and generating a preliminary citation network graph;

[0033] S503: Based on the preliminary citation network diagram, analyze the correlation and influence between documents, identify the core connections and key nodes between legal materials, and generate citation network analysis information.

[0034] The present invention improves the community detection algorithm according to the formula:

[0035]

[0036] Calculate modularity Q′ and generate a preliminary graph of the citation network;

[0037] Among them, A ij is the edge weight between nodes i and j, k i and k j are the degrees of nodes i and j respectively, m is the sum of the weights of the edges in the graph, δ(c i ,c j ) is an indicator function, which is 1 when nodes i and j belong to the same community, otherwise it is 0. i and c j represents the community of nodes i and j, S ij is the similarity score between documents i and j, L i and L j represents the legal criticality index of documents i and j, T is the sum of the legal criticality indexes of the documents, and α, β and γ are weight coefficients.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are:

[0039] In the present invention, similar cases can be automatically clustered effectively through multidimensional space mapping and case group identification, which brings obvious advantages in the accuracy of information retrieval. The introduction of cosine similarity measurement further improves the accuracy of case similarity assessment, ensuring that the reference case is highly matched with the current case. Using the community detection algorithm, through the creation of legal citation feature sets and citation relationship graphs, the understanding of the relationship between legal documents is deepened to accurately judge the importance and relevance of materials, thereby improving the efficiency and accuracy of litigation material matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of an AI-based intelligent matching method for litigation materials is proposed for the present invention;

[0041] Figure 2 The present invention proposes a detailed flow chart of step S1 in a litigation material intelligent matching method based on AI;

[0042] Figure 3 The present invention proposes a detailed flow chart of step S2 in a litigation material intelligent matching method based on AI;

[0043] Figure 4 The present invention proposes a detailed flow chart of step S3 in a litigation material intelligent matching method based on AI;

[0044] Figure 5 The present invention proposes a detailed flow chart of step S4 in a litigation material intelligent matching method based on AI;

[0045] Figure 6 The present invention proposes a detailed flow chart of step S5 in a litigation material intelligent matching method based on AI;

[0046] Figure 7 The present invention proposes a detailed flow chart of step S6 in a litigation material intelligent matching method based on AI. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0049] Example

[0050] See also Figure 1 The present invention provides a technical solution: an AI-based litigation material intelligent matching method, comprising the following steps:

[0051] S1: Based on the case database, case text analysis and part-of-speech tagging are performed to determine the legal attributes of keywords, filter non-critical stop words, and obtain a set of case feature vectors based on the frequency of keywords and legal concepts;

[0052] S2: Based on the case feature vector set, the case data is mapped into a multidimensional space, the case groups clustered in the feature space are identified, and the cases are grouped to generate a case clustering data set;

[0053] S3: Based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical cases, and the case with the best similarity is selected as a reference to provide a legal judgment basis for the current case and generate a case similarity evaluation record;

[0054] S4: Based on the case similarity assessment records and in combination with the legal material database, the document semantic analysis is performed, including the identification of legal terms and the extraction of legal citations. A legal citation feature set is constructed through legal terminology and citation information.

[0055] S5: Based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the correlation and influence between legal documents, and mine the legal connections between materials to generate citation network analysis information;

[0056] S6: Based on the citation network analysis information and combined with the case similarity assessment records, by analyzing the similarity of the cases and the citation criticality of the legal materials, the legal materials most relevant to the current case are matched and a litigation material matching list is generated.

[0057] The case feature vector set includes the frequency of key legal terms, case type labels, and the number of citations of relevant regulations. The case similarity assessment records include reference case identification, similarity scores, and analysis of key legal points in the case. The legal citation feature set includes the cited case number, relevant legal article index, and analysis of legal concepts in the citation. The citation network analysis information includes node strength scores, core case and regulatory links, and highly correlated citation paths. The litigation material matching list includes the legal document matching list, material correlation scores, and case application scenario proposals.

[0058] See also Figure 2 ,Based on the case database, the steps of case text analysis and part-of-speech tagging are carried out to determine the legal attributes of keywords, filter non-critical stop words, and obtain the case feature vector set according to the frequency of keywords and legal concepts are as follows:

[0059] S101: Based on the case database, perform word segmentation processing on the case text, convert the case description into a set of words or phrases, and generate a case text word segmentation result;

[0060] S102: Based on the word segmentation results of the case text, perform part-of-speech tagging, mark each word segmentation result with its attribute in the legal context, including nouns, verbs or proprietary legal terms, and generate a part-of-speech tag list;

[0061] S103: Based on the part-of-speech tag list, non-critical stop words are filtered, and a case feature vector set is obtained according to the frequencies of the remaining keywords and legal concepts.

[0062] In sub-step S101, based on the case database, the natural language processing (NLP) algorithm is used to perform word segmentation on the case text using Python's NLTK library. The Tokenize function is executed to convert the case description into a collection of words or phrases. The word segmentation parameters are set to identify English and legal terms, and the case text segmentation results are generated.

[0063] In sub-step S102, based on the case text segmentation results, a part-of-speech tagging algorithm is used, and Python's spaCy library is used to perform part-of-speech tagging. The part-of-speech tagging model is set to a legal context optimization model to identify nouns, verbs, or proprietary legal terms. The assign_tags function is executed to assign appropriate legal attributes and generate a part-of-speech tagging list.

[0064] In sub-step S103, based on the part-of-speech tag list, a frequency analysis algorithm is used to filter non-critical stop words using Python's Pandas library, and the remove_stopwords function is executed to delete non-keywords. The calculate_frequency function is used to set threshold parameters based on the frequency of occurrence of remaining keywords and legal concepts to filter high-frequency words and generate a set of case feature vectors.

[0065] See also Figure 3 Based on the case feature vector set, the case data is mapped into a multidimensional space, the case groups clustered in the feature space are identified, and the cases are grouped. The specific steps for generating the case clustering data set are as follows:

[0066] S201: Based on the case feature vector set, each case is represented as a point in a multidimensional space, and spatial mapping of the case data is performed to generate case mapping data;

[0067] S202: Based on the case mapping data, identify case points clustered in space, and regard the cases as groups including similar features, and generate case group identification results;

[0068] S203: Based on the case group identification results, the cases are grouped, a unique identifier is assigned to each group, and key features are described to generate a case clustering data set.

[0069] In sub-step S201, based on the case feature vector set, the principal component analysis (PCA) method is used to perform multidimensional spatial mapping on each case representation using Python's Scikit-learn library, and the pac.fit_transform function is executed. The n_components parameter is set to select the number of principal components to be retained, so as to reduce the data dimension while retaining the most original data variation, perform spatial mapping of the case data, and generate case mapping data.

[0070] In sub-step S202, based on the case mapping data, the K-means clustering algorithm is used to identify case points clustered in space using Python's Scikit-learn library, and the KMeans.fit_predict function is executed. The n_clusters parameter is set to determine the number of clusters. The case points are assigned to the nearest cluster center according to the distance in the feature space to generate case group identification results.

[0071] In sub-step S203, based on the case group identification results, the label assignment method is adopted, Python is used to perform case grouping operations, the unique_labels function is executed to assign unique identifiers, and the groupby function is used to aggregate cases according to cluster identifiers, describing the key features of each group, including the cluster center and the number of cases, to generate a case clustering data set.

[0072] See also Figure 4 ,Based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical case, and the case with the best similarity is selected as a reference to provide a legal basis for the current case. The specific steps for generating case similarity evaluation records are as follows:

[0073] S301: Based on the case clustering data set, using the cosine similarity measurement, the feature vector of the current case is compared with the feature vector of each case in the data set, the angle difference is calculated, and the case similarity calculation data is generated;

[0074] S302: Based on the case similarity calculation data, a case with the best similarity is selected, and the case is determined as a reference case most relevant to the current case, and a screening result of similar cases is generated;

[0075] S303: Based on the screening results of similar cases, the information of the selected cases is integrated, including the case number, relevant legal clauses and judgment results, to provide a legal judgment basis for the current case and generate a case similarity assessment record.

[0076] In sub-step S301, based on the case clustering data set, the cosine similarity measurement method is adopted, and the Python Scikit-learn library is used to compare the feature vector of the current case with the feature vector of each case in the data set. The cosine_similarity function is executed, and the parameters are set to calculate the cosine similarity between the vectors, thereby performing angle difference calculation and generating case similarity calculation data.

[0077] The cosine similarity metric is based on the formula:

[0078]

[0079] Calculate case similarity scores;

[0080] Among them, A is the feature vector of the current case, B is the feature vector of the cases in the dataset, W is the feature weight vector, and Θ is the adjustment coefficient to consider additional differences or similarities between cases.

[0081] Execution process:

[0082] First, define the feature vectors A and B of the current case and the cases in the dataset. Each feature vector contains several dimensions, representing different attributes of the case. Then, introduce a new feature weight vector W, in which each element corresponds to the importance of each dimension in A and B, which is determined by data analysis methods. Then, weight each element of A and B, and use A·W and B·W to calculate the weighted vector dot product and modulus. Finally, add the adjustment coefficient Θ, which is determined by case analysis and is used to fine-tune the final similarity score to ensure that the calculation result can more accurately reflect the actual similarity or difference between cases.

[0083] In sub-step S302, based on the case similarity calculation data, the optimal selection algorithm is adopted, and Python is used to select the case with the best similarity. The argmax function is executed to determine the case index with the highest cosine similarity value, thereby determining the reference case most relevant to the current case and generating a screening result for similar cases.

[0084] In sub-step S303, based on the screening results of similar cases, a data integration method is adopted to use Python's Pandas library to integrate the information of the selected cases, including case numbers, relevant legal provisions and judgment results. The merge function is executed to merge the current case data with the detailed information of the selected case to generate a case similarity assessment record.

[0085] See also Figure 5 Based on the case similarity assessment records and combined with the legal material database, the semantic analysis of the document is carried out, including the identification of legal terms and the extraction of legal citations. The specific steps of constructing the legal citation feature set are as follows:

[0086] S401: Based on the case similarity assessment record, access the legal material database, extract the legal terms and citation information mentioned in the record, and generate a legal data extraction record;

[0087] S402: Based on the legal data extraction records, analyze the contextual relationship between legal proper nouns and citation information, identify the role and significance in legal argumentation, and generate legal citation analysis results;

[0088] S403: Based on the results of the legal citation analysis, a collection is established, including the definition of relevant legal terms, the cited laws and cases, and the relationship between multiple elements, to construct a legal citation feature collection.

[0089] In sub-step S401, based on the case similarity assessment records, the legal material database is accessed, a text data extraction algorithm is adopted, the HTML or XML data is parsed using Python's BeautifulSoup library, the find_all function is executed to extract the legal terms and citation information mentioned in the records, and the filtering parameters are set to locate the tags containing legal terms and citations, and the legal data extraction records are generated.

[0090] In sub-step S402, based on the legal data extraction records, the natural language processing (NLP) algorithm is adopted, and the contextual relationship between legal proper nouns and citation information is analyzed using Python's NLTK library. The pos_tag function is executed for part-of-speech tagging, and the identify_context function is used to identify the role and significance in legal argumentation, and the legal citation analysis results are generated.

[0091] In sub-step S403, based on the results of legal citation analysis, a feature set is constructed. An information aggregation algorithm is used to integrate relevant legal term definitions, cited laws and cases, and the relationship between multiple elements using Python's Pandas library. The concat function is executed to merge data from different sources to construct a comprehensive legal citation feature set and generate a legal citation feature set.

[0092] See also Figure 6 Based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the correlation and influence between legal documents, and mine the legal connections between materials. The specific steps for generating citation network analysis information are as follows:

[0093] S501: Based on the legal citation feature set, collect and organize direct citation relationships between legal documents, including citations between legal articles or cases, and summarize the information to generate a legal document citation relationship set;

[0094] S502: Based on the legal document citation relationship set, a community detection algorithm is used to identify mutual citations and one-way citations between documents, a preliminary citation relationship graph between legal materials is constructed, and a preliminary citation network graph is generated;

[0095] S503: Based on the preliminary citation network diagram, analyze the correlation and influence between documents, identify the core connections and key nodes between legal materials, and generate citation network analysis information.

[0096] In sub-step S501, based on the legal citation feature set, a data integration method is adopted to use Python's Pandas library to collect and organize direct citation relationships between legal documents, perform DataFrame operations to summarize information on citations between legal articles or cases, set data filtering conditions to include only valid legal document citations, and generate a legal document citation relationship set.

[0097] In sub-step S502, based on the citation relationship set of legal documents, a community detection algorithm, specifically the Louvain method, is used to identify mutual citations and one-way citations between documents using Python's community library. The best_partition function is executed to construct a preliminary citation relationship graph between legal materials. The weight parameters are set to take into account the frequency and importance of citations, and a preliminary graph of the citation network is generated.

[0098] Community detection algorithm, according to the formula:

[0099]

[0100] Calculate modularity Q′ and generate a preliminary graph of the citation network;

[0101] Among them, A ij is the edge weight between nodes i and j, k i and k j are the degrees of nodes i and j respectively, m is the sum of the weights of all edges in the graph, δ(c i ,c j ) is an indicator function, which is 1 when nodes i and j belong to the same community, otherwise it is 0. i and c j represents the community of nodes i and j, S ij is the similarity score between documents i and j, L i and L j represents the legal criticality index of documents i and j, T is the sum of the legal importance indexes of all documents, and α, β and γ are weight coefficients.

[0102] Execution process:

[0103] First, calculate the mutual reference weight A between each pair of nodes ij , and through the additional indicator S ij Consider the similarity between documents to enhance the accuracy of community detection, and then introduce the legal importance index L of the document i and L j , and adjust the degree of each node by the weight coefficient β to reflect its importance in the legal system. Then, the weight coefficient α is used to enhance or weaken the similarity score S ijThe influence of isotope is taken into account to ensure that similarity is properly considered in community construction. At the same time, the denominator of the modularity formula is adjusted by introducing the adjustment coefficient γ of the total legal importance index T to balance the weight distribution of edges in the network. Finally, the optimal parameter setting is determined by comparing different α, β and γ values. The specific value is iteratively searched within a given range through optimization algorithms, such as grid search or genetic algorithm, to maximize the module Q′, thereby generating a network diagram that accurately reflects the mutual citation relationship of legal documents.

[0104] In sub-step S503, based on the preliminary graph of the citation network, a network analysis method is used, specifically the NetworkX library, to analyze the correlation and influence between documents, and the degree_centrality function is executed to identify the core connections and key nodes between legal materials, considering the connectivity of the nodes and their position in the network, to generate citation network analysis information.

[0105] See also Figure 7 Based on the citation network analysis information, combined with the case similarity assessment records, by analyzing the similarity of the cases and the citation criticality of the legal materials, the legal materials most relevant to the current case are matched, and the steps for generating a litigation material matching list are as follows:

[0106] S601: Based on the citation network analysis information, filter the legal document nodes that are highly relevant to the current case, including core legal provisions and frequently cited cases, and generate relevant legal document screening information;

[0107] S602: Based on the related legal document screening information, combined with the case similarity evaluation record, the similarity and citation criticality between the selected document and the current case are compared, a case related document list is determined, and a case document similarity analysis result is generated;

[0108] S603: Based on the case document similarity analysis results, screen the materials that can reflect the current case issues and provide legal support, match the legal materials required for the current case, and generate a litigation material matching list.

[0109] In sub-step S601, based on the citation network analysis information, a key node identification method is adopted, specifically the degree_centrality function of the NetworkX library. Python is used to analyze the centrality of legal document nodes, and legal document nodes with high relevance to the current case are screened, including core legal provisions and frequently cited cases. Filter conditions are executed to locate nodes with centrality higher than the set threshold, and related legal document screening information is generated.

[0110] In sub-step S602, based on the information of related legal documents screening and combined with the case similarity assessment records, a similarity comparison algorithm is adopted, specifically the cosine similarity measurement method, and Python's Scikit-learn library is used to compare the similarity and citation criticality between the selected document and the current case. The cosine_similarity function is executed to calculate the similarity, determine the case-related document list, and generate the case document similarity analysis results.

[0111] In sub-step S603, based on the case document similarity analysis results, a material matching algorithm is adopted, specifically a screening method based on similarity scoring. Python is used to screen materials that can reflect the current case issues and provide legal support. Screening criteria are set to match legal materials with high similarity scores. The legal materials required for the current case are matched to generate a litigation material matching list.

[0112] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An AI-based intelligent matching method for litigation materials, characterized in that: The following steps are involved: Based on the case database, case text analysis and part-of-speech tagging are performed to determine the legal attributes of keywords, filter out non-critical stop words, and obtain a set of case feature vectors based on the frequency of keywords and legal concepts; Based on the case feature vector set, the case data is mapped into a multidimensional space, the case groups clustered in the feature space are identified, and the cases are grouped to generate a case clustering data set; Based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical case, and the case with the best similarity is selected as a reference to generate a case similarity evaluation record; Based on the case similarity assessment records and in combination with the legal material database, the document semantic analysis is performed, including the identification of legal terms and the extraction of legal citations, and a legal citation feature set is constructed through legal terminology and citation information; Based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the relevance and influence between legal documents, and mine the legal connections between materials to generate citation network analysis information; Based on the citation network analysis information, combined with the case similarity assessment record, by analyzing the similarity of the cases and the citation criticality of the legal materials, the legal materials associated with the current case are matched to generate a litigation material matching list.

2. The AI-based litigation material intelligent matching method according to claim 1 is characterized in that: The case feature vector set includes the frequency of key legal terms, case type labels, and the number of citations of relevant regulations. The case similarity assessment record includes the reference case identification, similarity score, and analysis of key legal points of the case. The legal citation feature set includes the cited case number, the relevant legal article index, and the analysis of legal concepts in the citation. The citation network analysis information includes node strength scores, core case and regulatory links, and the citation path with the best correlation. The litigation material matching list includes a legal document matching list, a material correlation score, and a case application scenario proposal.

3. The AI-based litigation material intelligent matching method according to claim 1 is characterized in that: Based on the case feature vector set, the case data is mapped into a multidimensional space, the case groups clustered in the feature space are identified, and the cases are grouped. The specific steps of generating a case clustering data set are as follows: Based on the case feature vector set, each case is represented as a point in a multidimensional space, and spatial mapping of the case data is performed to generate case mapping data; Based on the case mapping data, identifying case points clustered in space, and treating the cases as groups with similar characteristics, generating case group identification results; Based on the case group identification results, the cases are grouped, a unique identifier is assigned to each group, and key features are described to generate a case clustering data set.

4. The AI-based litigation material intelligent matching method according to claim 1 is characterized in that: Based on the case clustering data set, the cosine similarity measurement is used to evaluate the similarity between the current case and the historical case, and the case with the best similarity is selected as a reference. The specific steps of generating the case similarity evaluation record are as follows: Based on the case clustering data set, using the cosine similarity measurement, the feature vector of the current case is compared with the feature vector of each case in the data set, the angle difference is calculated, and the case similarity calculation data is generated; Based on the case similarity calculation data, a case with the best similarity is selected, and the case is determined as a reference case associated with the current case to generate a screening result of similar cases; Based on the screening results of similar cases, the information of the selected cases is integrated, including the case number, relevant legal clauses and judgment results, to provide a legal judgment basis for the current case and generate a case similarity assessment record.

5. The AI-based litigation material intelligent matching method according to claim 4 is characterized in that: The cosine similarity metric is based on the formula: Calculate case similarity scores; Among them, A is the feature vector of the current case, B is the feature vector of the case in the data set, W is the feature weight vector, and Θ is the adjustment coefficient.

6. The AI-based litigation material intelligent matching method according to claim 1 is characterized in that: Based on the case similarity assessment records, combined with the legal material database, the semantic analysis of the document is performed, including the identification of legal terms and the extraction of legal citations, and the steps of constructing a legal citation feature set through legal terminology and citation information are as follows: Based on the case similarity assessment records, access the legal material database, extract the legal terms and citation information mentioned in the records, and generate legal data extraction records; Based on the legal data extraction records, analyzing the contextual relationship between legal proper nouns and citation information, identifying the role and significance in legal argumentation, and generating legal citation analysis results; Based on the legal citation analysis results, a collection is established, including the definition of relevant legal terms, cited laws and cases, and the relationship between multiple elements to construct a legal citation feature collection.

7. The AI-based litigation material intelligent matching method according to claim 1 is characterized in that: Based on the legal citation feature set, a community detection algorithm is used to create a citation relationship graph between legal materials, analyze the correlation and influence between legal documents, and mine the legal connections between materials. The specific steps for generating citation network analysis information are as follows: Based on the legal citation feature set, direct citation relationships between legal documents are collected and sorted, including citations between legal articles or cases, and information is summarized to generate a legal document citation relationship set; Based on the legal document citation relationship set, using a community detection algorithm, identifying mutual citations and one-way citations between documents, constructing a preliminary citation relationship graph between legal materials, and generating a preliminary citation network graph; Based on the preliminary citation network diagram, the correlation and influence between documents are analyzed, the core connections and key nodes between legal materials are identified, and citation network analysis information is generated.

8. The AI-based litigation material intelligent matching method according to claim 7 is characterized in that: The community detection algorithm is based on the formula: Calculate the modularity Q′ and generate a preliminary graph of the citation network; Among them, A ij is the edge weight between nodes i and j, k i and k j are the degrees of nodes i and j respectively, m is the sum of the weights of the edges in the graph, δ(c i ,c j ) is an indicator function, which is 1 when nodes i and j belong to the same community, otherwise it is 0. i and c j represents the community of nodes i and j, S ij is the similarity score between documents i and j, L i and L j represents the legal criticality index of documents i and j, T is the sum of the legal criticality indexes of the documents, and α, β and γ are weight coefficients.

Images