Intelligent case generation method based on judicial knowledge domain graph

Abstract

The invention relates to the technical field of legal text generation, in particular to an intelligent case generation method based on a judicial knowledge domain atlas, which comprises the following steps: S1, constructing the judicial knowledge domain atlas; s2, extracting case facts and judgment reason parts of the document; s3, classifying and splitting case facts of the document; s4, carrying out deep learning to generate an abstract; s5, correlation comparison of the content is carried out, and integrity and accuracy are verified; s6, optimizing a judgment key through a convolutional neural network model; s7, automatically generating correlation indexes with other historical cases according to correlation analysis in the judicial knowledge domain graph; s8, summarizing the results, and splicing the results into a complete case document; according to the method, by combining the judicial knowledge domain graph and the deep learning technology, the legal document is automatically generated, the integrity of document content and the accuracy of legal logic are ensured, historical cases are efficiently associated, and the efficiency and quality of judgment document generation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of legal text generation, and in particular to a case intelligent generation method based on a judicial knowledge domain graph. Background Art

[0002] In modern judicial practice, the generation of judicial documents is a complex and time-intensive process, especially in the handling of multi-trial cases and the writing of judicial reasons. The traditional document writing process relies on judges and legal workers to conduct in-depth analysis of case facts and the application of legal provisions, which is not only time-consuming but also prone to errors due to human factors. At the same time, with the increase in the number of cases, the efficiency problem of manual document processing has gradually become prominent, especially when it is necessary to correlate and compare the current case with a large number of historical cases and legal provisions. The task becomes particularly arduous. In addition, manually generated documents also face challenges in maintaining content consistency, completeness and legal accuracy, and often require multiple revisions to meet the court submission standards.

[0003] Therefore, it is necessary to develop an intelligent case generation method based on judicial knowledge domain graph to solve the above problems. Summary of the invention

[0004] Based on the above objectives, the present invention provides an intelligent case generation method based on judicial knowledge domain graph.

[0005] The case intelligent generation method based on the judicial knowledge domain map includes the following steps:

[0006] S1: constructing a judicial knowledge domain map, which includes information on case types, trial procedures, adjudication basis, and judgment results;

[0007] S2: Parse the input multi-trial legal documents, extract the case facts and judgment reasons of the documents, and mark the newly added facts in the multi-trial cases;

[0008] S3: Based on the judicial knowledge domain map constructed in S1 and the document information extracted in S2, the segmented rule extraction algorithm is used to classify and split the case facts in the documents to generate a preliminary summary of the basic case facts;

[0009] S4: For the judgment reasons extracted from S2, the BERT+LSTM model is used to perform deep learning generative summarization to obtain the judgment reasons and judgment essence.

[0010] S5: Combine the judicial knowledge domain map of S1 with the judgment reasons and judgment essence generated by S4 to compare the contents and verify their completeness and accuracy;

[0011] S6: Using the results of the correlation comparison in S5, the convolutional neural network model is used to optimize the judgment essence, and enhance the legal logic of the judgment essence and the applicability of the case type;

[0012] S7: Based on the judgment essence optimized by S6, the correlation index with other historical cases is automatically generated according to the correlation analysis in the judicial knowledge domain map;

[0013] S8: Summarize the results of S3, S4, S6 and S7 into a complete case document and format it to meet the standard output requirements of judicial documents.

[0014] Optionally, the S1 specifically includes:

[0015] S11: Obtain historical judgment documents from an existing judicial database, wherein the historical judgment documents include first-instance judgments, second-instance judgments, and retrial judgments, wherein the data source is a public judicial database;

[0016] S12: preprocessing the obtained judgment documents, including standardization of data format, cleaning of redundant information and elimination of abnormal data;

[0017] S13: Based on the preprocessed data, a graph structure algorithm is applied to construct a judicial knowledge domain graph, with case types as nodes and case trial procedures, adjudication basis and judgment results as edges, forming a directed graph connecting different case types;

[0018] S14: Label the nodes and edges in the graph, wherein the labels include the case type corresponding to each node, the trial procedure corresponding to each edge, the basis for adjudication, and relevant information of the judgment result.

[0019] Optionally, the S2 specifically includes:

[0020] S21: Perform text segmentation on the input multi-review legal documents, and preliminarily divide the case facts part and the judgment reasons part in the documents according to the fixed legal document format based on natural language processing technology. The division criteria include paragraph markers, legal terms and common document structures;

[0021] S22: Use named entity recognition technology to extract predetermined elements from documents, including the persons involved, time, location, case type and case facts of the dispute;

[0022] S23: Based on word frequency analysis and semantic parsing, extract the content of the judgment reasons, including the basis for the judgment, legal analysis and key points of the judgment;

[0023] S24: Compare and analyze documents of multi-trial cases, and mark newly added case facts and judgment reasons in the second-instance or retrial documents based on timestamp information and semantic similarity calculations to ensure that no information in multi-trial documents is missed.

[0024] Optionally, the S3 specifically includes:

[0025] S31: Obtain the document information extracted in S2, including the case facts and judgment reasons, and preliminarily classify the documents according to the case type, trial procedure and judgment basis. The specific classification is based on the nodes and edges in the judicial knowledge domain graph constructed by S1;

[0026] S32: applying a segmented rule extraction algorithm to classify case facts, wherein the segmented rule extraction algorithm performs classification based on a fixed paragraph structure of the document and a set of legal terms;

[0027] S33: Classify and store the paragraphs that meet the rules in S32 according to categories. The classified case fact information includes the time of the case, the identity of the parties, and the elements of the litigation request.

[0028] S34: Split the classified case fact information into specific events, dispute focus and key points of evidence according to the case type and trial procedure;

[0029] S35: Based on the results of classification and splitting, a preliminary basic case summary is generated, which includes the main facts of the case, the focus of the dispute and the description of relevant evidence, and is formatted and output according to a standardized template.

[0030] Optionally, the S32 specifically includes:

[0031] S321: Define a set of rules, where each rule corresponds to a class of case facts;

[0032] S322: Convert the input document paragraph into a word vector representation by using word embedding technology. Specifically, convert the words in each paragraph into corresponding numerical representations, and average the vectors of all words in the paragraph to obtain the vector representation of the entire paragraph.

[0033] S323: for each paragraph, compare its vector representation with the vector representation of the predefined rule, and calculate the similarity between the paragraph and each rule. The higher the similarity, the closer the paragraph is to the rule.

[0034] S324: According to the calculated similarity, classify the paragraph into the category corresponding to the rule with the highest similarity.

[0035] Optionally, the S4 specifically includes:

[0036] S41: pre-processing the text of the judgment reasons extracted in S2, and dividing the text into sentences and paragraphs;

[0037] S42: Input each sentence into the BERT model, encode the words in each sentence, and generate a word vector representation with contextual relevance;

[0038] S43: Input the vector of each sentence generated by BERT into the LSTM model to capture the contextual dependencies between sentences;

[0039] S44: After LSTM processes all the sentences, an attention mechanism is used to generate a summary. The attention mechanism weights the hidden state of each sentence to identify the sentences that contribute more to the summary, thereby generating summary content.

[0040] S45: Generate a summary of the reasons for the judgment and the essence of the judgment based on the hidden state of the LSTM through the decoder.

[0041] Optionally, the S5 specifically includes:

[0042] S51: Preliminarily match the judgment reasons and judgment essence generated in S4 with the historical cases in the judicial knowledge domain map constructed in S1, and extract the historical case information related to the current case based on the nodes of case type, judgment basis and judgment result in the map;

[0043] S52: semantically match the judgment reasons generated by S4 with the judgment reasons of the same type of cases in the graph, and calculate the similarity between the two to ensure that the generated judgment reasons are consistent with the existing legal provisions and historical judgment logic;

[0044] S53: Compare the judgment summary. Based on the historical case data extracted from the graph, use the semantic similarity algorithm to compare the generated judgment summary with the historical judgment summary to determine whether the judgment summary covers all the legal points.

[0045] S54: Verify the completeness and accuracy of the judgment reasons and judgment essence through the results of the correlation comparison. If any omissions or inconsistencies are found in the comparison, they will be marked and corresponding correction suggestions will be generated based on the legal provisions and historical case information in the map;

[0046] S55: Finally, a verification report is generated, the comparison results and correction suggestions are output, and the generated reasons for judgment and judgment summary are adjusted according to the verification results.

[0047] Optionally, the S6 specifically includes:

[0048] S61: taking the judgment summary generated in S5 and the correction suggestions after the associated comparison as input, and converting the judgment summary text into a sequence of word vectors through word embedding technology, wherein each word is converted into a corresponding numerical representation to form a series of word vectors;

[0049] S62: Input the word vector sequence into the convolution layer of the convolutional neural network. The convolution layer extracts local features of the word vector sequence through a sliding convolution kernel to identify important vocabulary combinations and expression patterns in the judgment essence.

[0050] S63: The output of the convolution layer is processed by the pooling layer. The role of the pooling operation is to extract the most important features from the convolution result;

[0051] S64: The pooled features are passed to the fully connected layer, where all local features are integrated into global features, and all extracted features are combined to generate the final feature representation of the judgment essence;

[0052] S65: Based on the final feature representation, determine whether the judgment summary complies with the legal logic and language norms. If it does not meet the standards, feedback modification suggestions will be provided; if it meets the standards, the final optimized judgment summary text will be generated.

[0053] Optionally, the S7 specifically includes:

[0054] S71: Match the judgment essence optimized in S6 with the historical cases in the judicial knowledge domain map;

[0055] S72: Calculate the similarity between the optimized judgment summary and the historical case summary in the graph, convert the words in the judgment summary into vector representations based on the word vector model, and then calculate the semantic similarity between the summary text and the historical case text;

[0056] S73: Screen the matched relevant historical cases, and evaluate the correlation strength between different historical cases according to the correlation weights in the graph;

[0057] S74: Generate a correlation index, specifically associate the current adjudication essence with the screened historical cases, and display the case type, adjudication result and related information of each historical case in the graph through the correlation index;

[0058] S75: Output an index report based on the generated association index, which includes association details between the current case and the historical case, the correlation score, and summary information of the historical case.

[0059] Optionally, the S8 specifically includes:

[0060] S81: Obtain the basic case summary from S3, extract the core facts of the case, including the case background, persons involved, focus of dispute and the course of the case, and perform standardization to form a basic facts paragraph as the opening part of the document;

[0061] S82: Integrate the judgment reasons and judgment summary generated in S4 into the document according to the predetermined format. The judgment summary serves as the summary paragraph of the judgment reasons, which briefly summarizes the court's legal opinions and judgment basis;

[0062] S83: Based on the optimized judgment summary in S6, confirm whether its content complies with legal logic and language specifications, and insert the optimized judgment summary into the corresponding position of the document;

[0063] S84: append the associated index generated in S7 to the end of the case document;

[0064] S85: Format the summarized documents to ensure that the layout of the documents conforms to the standard format of judicial documents, including adding headers, paragraph numbers, titles and footers, and finally forming a complete case document.

[0065] Beneficial effects of the present invention:

[0066] The present invention, by introducing an intelligent generation method based on the judicial knowledge domain map, can effectively improve the efficiency and accuracy of legal document generation. By automatically extracting case facts, judgment reasons and judgment essence, and optimizing them in combination with deep learning technology, the generated documents not only ensure the integrity and consistency of the content, but also can provide more accurate legal judgment basis based on historical cases and legal provisions. This method solves the problem of difficult extraction of complex information in multi-trial cases, making the document generation process more efficient and without omissions.

[0067] The present invention, through the combination with the judicial knowledge domain map, realizes the intelligent association between judicial documents and historical cases, which not only provides legal workers with a convenient reference to historical precedents, but also ensures the legal logic and language norms of the generated documents. By deeply optimizing the judicial essence and automatically generating related indexes, users can quickly query relevant legal information, which significantly improves the accuracy and efficiency of case handling and provides strong technical support for the judicial system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0069] Figure 1 A schematic diagram of a case intelligent generation method flow chart of an embodiment of the present invention;

[0070] Figure 2 Schematic diagram of the steps of classifying case facts using the segmented rule extraction algorithm of an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0072] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0073] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0074] like Figure 1-Figure 2 As shown in the figure, the case intelligent generation method based on the judicial knowledge domain map includes the following steps:

[0075] S1: Construct a judicial knowledge domain map, which includes information on case types, trial procedures, adjudication basis, and judgment results, providing basic data support for case generation;

[0076] S2: Parse the input multi-trial legal documents, extract the case facts and judgment reasons of the documents, and mark the newly added facts in the multi-trial cases to ensure the integrity and continuity of the information;

[0077] S3: Based on the judicial knowledge domain map constructed in S1 and the document information extracted in S2, the segmented rule extraction algorithm is used to classify and split the case facts in the documents to generate a preliminary summary of the basic case facts;

[0078] S4: For the judgment reasons extracted from S2, the BERT+LSTM model is used to perform deep learning generative summarization to obtain the judgment reasons and judgment essence.

[0079] S5: Combine the judicial knowledge domain map of S1 with the judgment reasons and judgment essence generated by S4 to compare the correlation of the contents, verify their completeness and accuracy, and ensure the comprehensive presentation of the facts of the multi-trial case;

[0080] S6: Using the results of the correlation comparison in S5, the convolutional neural network (DGCNN) model is used to optimize the judgment essence, and enhance the legal logic of the judgment essence and the applicability of the case type;

[0081] S7: Based on the judgment essence optimized by S6, according to the correlation analysis in the judicial knowledge domain map, it automatically generates the correlation index with other historical cases, providing reference basis and support for legal judgment;

[0082] S8: Summarize the results of S3, S4, S6 and S7 into a complete case document and format it to meet the standard output requirements of judicial documents.

[0083] The construction of judicial knowledge domain map in S1 specifically includes:

[0084] S11: Obtain historical judgment documents from existing judicial databases. Historical judgment documents include first-instance judgments, second-instance judgments, and retrial judgments. The data source is a public judicial database;

[0085] S12: Preprocess the data of the obtained judgment documents, including standardization of data format, cleaning of redundant information and elimination of abnormal data, so as to ensure the integrity and consistency of the input data;

[0086] S13: Based on the preprocessed data, a graph structure algorithm is applied to construct a judicial knowledge domain graph, with case types as nodes and case trial procedures, adjudication basis and judgment results as edges, forming a directed graph connecting different case types;

[0087] S14: Label the nodes and edges in the graph. The labels include the case type corresponding to each node, the trial procedure corresponding to each edge, the basis for adjudication, and the relevant information of the judgment result, so that each element in the graph can accurately represent the actual judicial information. Through the above steps, it is ensured that the data in the construction process of the judicial knowledge field graph is complete and the structure is clear, providing an accurate reference for the intelligent generation of subsequent cases.

[0088] S2 specifically includes:

[0089] S21: Perform text segmentation on the input multi-review legal documents. Based on the natural language processing (NLP) technology, the case facts part and the judgment reasons part in the documents are preliminarily divided according to the fixed legal document format. The division criteria include paragraph markers, legal terms and common document structures.

[0090] The common document structure is as follows:

[0091] The cover page and header. The cover page usually contains basic information about the case, such as the case number, names of the parties, and the hearing court. The header may contain repeated information such as the case number and document category.

[0092] Information about the parties involved in the litigation, including the names and roles of the plaintiff, defendant, attorneys, judges, etc.;

[0093] Basic information of the case, briefly describing the background of the case, the legal relationships involved, the focus of the dispute, etc.;

[0094] Case facts, which describe the facts, including evidence, witness statements, court investigations, etc. This part usually describes the specific circumstances of the case in detail and is the key part of information extraction;

[0095] The court's opinion, which includes the court's application of law, evaluation of evidence, and determination of controversial issues in the case, including paragraphs with introductory words such as "This court holds" that focus on the court's legal views;

[0096] The judgment result is usually introduced as "the judgment is as follows" or "the ruling is as follows", which clearly lists the court's decision, including the judgment result, the effective date of the judgment, the payment of fees, etc.;

[0097] The concluding part includes appeal guidelines, legal effect explanations, etc., to guide the parties on how to exercise their rights in accordance with the law;

[0098] Attachments. Some documents may include attachments, such as relevant legal documents, copies of evidence materials, etc.

[0099] S22: Use named entity recognition (NER) technology to extract predetermined elements from the document, including the persons involved, time, place, case type and case facts of the dispute. The extraction process is based on the fixed expression patterns and legal terms in the document. Specifically, the input document is D, which contains multiple paragraphs P. i and sentence S j , and define a fixed expression pattern set M, which includes common legal terms such as "plaintiff", "defendant", "public request", etc.; use the pattern matching algorithm f match (S j , M), in each sentence S j Matching fixed expression patterns in the formula is: match (Sj , M)={t k ∈S j |t k ∈M}, where t k is a term in a sentence, and if the match is successful, the elements in the corresponding paragraph are extracted as predetermined elements;

[0100] S23: Based on word frequency analysis and semantic parsing, extract the content of the judgment reason part, including the judgment basis, legal analysis and judgment points. Specifically, the context-related semantic model is used for extraction. Suppose the judgment reason part of the document is C, the judgment basis part is B, and the semantic model f is used. semantic (C) represents the context-related semantic vector representation, the formula is: Among them, v(S j ) is the semantic vector representation of the jth sentence, w j The weights are related to the context. The core content of the judgment basis is extracted by calculating the similarity sim(v(C), v(B)) with the judgment basis in the judgment reason part.

[0101] S24: Compare and analyze the documents of multi-trial cases, and based on timestamp information and semantic similarity calculation, mark the newly added case facts and judgment reasons in the second-instance or retrial documents to ensure that no information in the multi-trial documents is missed; through the detailed description of the above sub-steps, it is possible to effectively parse and extract the case facts and judgment reasons in the multi-trial legal documents, and by marking the newly added facts, ensure the completeness and accuracy of the multi-trial case information, which helps to increase the accuracy of subsequent case generation.

[0102] The classification and splitting in S3 specifically include:

[0103] S31: Obtain the document information extracted in S2, including the case facts and judgment reasons, and preliminarily classify the documents according to the case type, trial procedure and judgment basis. The specific classification is based on the nodes and edges in the judicial knowledge domain graph constructed by S1;

[0104] S32: Applying a segmented rule extraction algorithm to classify case facts, the segmented rule extraction algorithm is based on the fixed paragraph structure of the document and the legal terminology set for classification;

[0105] S33: Classify and store the paragraphs that meet the rules in S32 according to categories. The classified case fact information includes the time of occurrence of the case, the identity of the parties, and the elements of the litigation request, so as to generate a subsequent basic case summary;

[0106] S34: Split the classified case fact information into specific events, dispute focus and key points of evidence according to the case type and trial procedure. The splitting is based on the associated nodes and edges in the judicial knowledge domain graph in S1 to ensure that the case fact information of different types of cases is correctly divided;

[0107] S35: Based on the results of classification and splitting, a preliminary summary of the basic facts of the case is generated. The summary includes the main facts of the case, the focus of the dispute and the description of relevant evidence. The summary text is generated based on the classified and split case fact information and is formatted and output according to a standardized template. The above steps can effectively extract and classify the case facts in legal documents through the above-mentioned segmented rule extraction algorithm and the graph-based classification and splitting mechanism, and generate a basic case summary with a clear structure and clear key points, ensuring the accurate analysis and information organization of complex legal documents.

[0108] The application of the segmented rule extraction algorithm in S32 to classify case facts specifically includes:

[0109] S321: Define a set of rules, where each rule corresponds to a category of case facts, such as "case background", "dispute focus" or "evidence description", and these rules are composed of case types, legal terms and fixed paragraph structures of documents for subsequent classification;

[0110] S322: Convert the input document paragraph into a word vector representation by using word embedding technology. Specifically, convert the words in each paragraph into corresponding numerical representations, and average the vectors of all words in the paragraph to obtain the vector representation of the entire paragraph.

[0111] S323: for each paragraph, compare its vector representation with the vector representation of the predefined rule, and calculate the similarity between the paragraph and each rule. The higher the similarity, the closer the paragraph is to the rule.

[0112] S324: Based on the calculated similarity, the paragraphs are classified into the category corresponding to the rule with the highest similarity. The above steps can accurately classify the case fact information so that the document content matches the case type and rules. This process improves the classification accuracy and processing efficiency of the case facts.

[0113] The computational steps for applying the segmented rule extraction algorithm are as follows:

[0114] First, define a set of segmented rules R, where each rule r k Corresponding to specific categories of case facts (such as “case background”, “dispute focus”, and “evidence description”), let r k represents the kth rule in the rule set, where k represents the number of the rule;

[0115] Then, for the input document paragraph P i , the paragraph P is embedded in the word i Converted into word vector representation v(P i ), the specific formula is: Among them, P i represents the i-th paragraph of the input; v(P i ) indicates paragraph P i The word vector of j Indicates paragraph P i The jth word in v(w j ) represents the word vector of word wj, N is the number of words in paragraph Pi, and j represents the index of the word;

[0116] Next, calculate the paragraph P i With the rule r k The similarity of , using cosine similarity to calculate its similarity score, the formula is as follows: Among them, sim(P i , r k ) indicates paragraph P i With the rule r k The similarity of v(r k ) represents the rule r k The word vector of ||v(P i )|| means paragraph P i The modulus of the word vector; ||v(r k )|| represents the rule r k The modulus of the word vector; represents the vector dot product;

[0117] Finally, based on the similarity score, select i The rule with the highest similarity k , and change paragraph P i Classify into the corresponding category; the classification formula is: Category(P i ) = argmax k sim(P i , r k ), where Category(P i ) indicates paragraph P i The category to which it is classified, the rule corresponding to the maximum similarity r k Determine the category of the paragraph.

[0118] S4 specifically includes:

[0119] S41: pre-processing the text of the judgment reasons extracted in S2, and dividing the text into sentences and paragraphs;

[0120] S42: Input each sentence into the BERT model, encode the words in each sentence, and generate a word vector representation with context relevance; the core of the BERT model is that it can learn the relationship between each word and its context through a multi-layer Transformer mechanism and output a vector representation of the entire sentence. This vector can well express the semantic information of the sentence;

[0121] S43: The vector of each sentence generated by BERT is input into the LSTM model to capture the contextual dependencies between sentences. The LSTM model can remember the content of previous sentences and take this contextual information into account when processing subsequent sentences, thereby providing a coherent analysis of the entire judgment reasoning section.

[0122] S44: After LSTM processes all sentences, it uses the attention mechanism to generate a summary. The attention mechanism weights the hidden state of each sentence to identify the sentences that contribute more to the summary, thereby generating a more accurate and focused summary content.

[0123] S45: Generate a summary of the reasons for the judgment and the essence of the judgment based on the hidden state of the LSTM through the decoder.

[0124] The steps to use the BERT+LSTM model are as follows:

[0125] First, the text is divided into sentences and paragraphs to obtain a set of sentences. Let each sentence be S i , where i represents the index of the i-th sentence;

[0126] Then, each sentence S i Input the BERT model, use the BERT model to encode the sentence, and convert the words of each sentence into word vectors; for sentence S i Each word w in j , the BERT model generates a context-dependent word vector v(w j ); BERT captures contextual information through a multi-layer Transformer mechanism and outputs a vector representation v(S i ), the formula is: v(S i )=BERT(S i ), where v(S i ) represents the sentence v i The input of the BERT model is the sentence S i , the output is its corresponding vector;

[0127] Next, the sentence vector v(S i) is input into the LSTM model. LSTM is used to capture the contextual dependencies between sentences. Let the hidden state of LSTM be h t , the current time is t , then the state update formula of LSTM is: t =LSTM(v(v t ), h t-1 ), where h t is the LSTM hidden state at time t, v(S t ) is the input sentence vector at time t, h t-1 It is the hidden state of the previous moment. LSTM captures the dependencies in the sentence sequence by continuously updating the hidden state.

[0128] After LSTM has processed all the sentences, the last hidden state h T Generate a summary. The summary generation process is completed through the attention mechanism, focusing on each hidden state h t The words that generate the summary j The correlation between them, the calculation formula of attention score is: Among them, h t represents the hidden state of LSTM, s j is the jth word vector in the generated summary, Attention(h t ,s j ) represents the hidden state h t Generate words j The attention mechanism associates each hidden state with the summary words to improve the generation accuracy;

[0129] Finally, through the attention mechanism and decoder, a summary of the reasons for the judgment and the main points of the judgment is generated based on the final hidden state of the LSTM. The decoder weights the hidden state of the LSTM output according to the attention score, and gradually generates a concise summary containing the core legal points of the case.

[0130] S5 specifically includes:

[0131] S51: Preliminarily match the judgment reasons and judgment essence generated in S4 with the historical cases in the judicial knowledge domain map constructed in S1, and extract the historical case information related to the current case based on the nodes of case type, judgment basis and judgment result in the map;

[0132] S52: semantically match the judgment reasons generated by S4 with the judgment reasons of the same type of cases in the graph, and ensure that the generated judgment reasons are consistent with the existing legal provisions and historical judgment logic by calculating the similarity between the two; specifically, let R1 be the currently generated judgment reasons, and R2 be the historical judgment reasons in the graph, and calculate their similarity. The similarity calculation is based on the word vector model, and the formula is as follows: Among them, v(R1) and v(R2) are the word vector representations of the judgment reasons R1 and R2 respectively, and · represents the dot product of the vectors, ensuring that the generated judgment reasons are semantically consistent with the historical cases in the graph;

[0133] S53: Compare the judgment summary. Based on the historical case data extracted from the graph, use the semantic similarity algorithm to compare the generated judgment summary with the historical judgment summary to determine whether the judgment summary covers all legal points. Let Y1 be the generated judgment summary, and Y2 be the historical judgment summary extracted from the graph. Calculate the semantic similarity between the two based on the similarity algorithm. The formula is as follows: Among them, v(Y1) and v(Y2) are the word vector representations of the judgment essence Y1 and the historical judgment essence Y2 respectively. This calculation is used to determine whether the generated judgment essence covers all legal points;

[0134] S54: Verify the completeness and accuracy of the judgment reasons and judgment essence through the results of the correlation comparison. If any omissions or inconsistencies are found in the comparison, they will be marked and corresponding correction suggestions will be generated based on the legal provisions and historical case information in the map;

[0135] S55: Finally, a verification report is generated, and the comparison results and correction suggestions are output. The generated reasons for the judgment and the main points of the judgment are adjusted according to the verification results to ensure the compliance and integrity of the case content within the legal framework. Through the above-mentioned correlation comparison and verification steps, it can be ensured that the generated reasons for the judgment and the main points of the judgment are consistent with the historical cases in the judicial knowledge field map, and potential omissions or inconsistencies can be discovered and corrected in a timely manner, thereby improving the accuracy and legality of the generated content.

[0136] S6 specifically includes:

[0137] S61: Take the judgment summary generated in S5 and the correction suggestions after the associated comparison as input, and convert the judgment summary text into a sequence of word vectors through word embedding technology, where each word is converted into a corresponding numerical representation to form a series of word vectors, which are used for subsequent convolution operations;

[0138] S62: Input the word vector sequence into the convolution layer of the convolutional neural network. The convolution layer extracts local features of the word vector sequence through a sliding convolution kernel to identify important vocabulary combinations and expression patterns in the judgment essence. The convolution layer can effectively capture the short-distance dependency between words.

[0139] S63: The output of the convolution layer is processed by the pooling layer. The role of the pooling operation is to extract the most important features from the convolution result, reduce the data dimension, and retain the most critical information. The pooling layer helps to enhance the model's ability to capture important information;

[0140] S64: The pooled features are passed to the fully connected layer, where all local features are integrated into global features, and all extracted features are combined to generate the final feature representation of the judgment essence;

[0141] S65: Based on the final feature representation, determine whether the judgment summary complies with the legal logic and language norms. If it does not meet the standards, feedback modification suggestions will be given; if it meets the standards, the final optimized judgment summary text will be generated. Through the multi-level feature extraction and integration of the convolutional neural network model, the language expression and legal logic of the judgment summary can be deeply optimized to ensure that the generated judgment summary is not only complete but also logical.

[0142] The steps for optimizing the referee essence through the convolutional neural network model are as follows:

[0143] Assume that the text of the judgment is Y i , where each word w j is embedded as a word vector v(w j ), the text is represented as a vector sequence [v(w1), v(w2), ..., v(W n )];

[0144] Apply a convolution layer to the embedded word vector sequence, set the convolution kernel size to k, extract local features for each k word vector, and generate a feature map f(Y i ), the convolution operation is used to capture the characteristics of local word combinations in the judgment essence. The convolution calculation formula is: f(Y i )=ReLU(W·v(Y i )+b), where W is the convolution kernel weight, b is the bias, and v(Y i ) is a word vector sequence, and the convolution result is nonlinearly mapped through the activation function ReLU;

[0145] The pooling layer uses the maximum pooling operation to reduce the feature dimension by selecting the maximum value in each convolution window and retain the most important feature information. The feature after pooling is represented as p(Y i );

[0146] The pooled feature vector is input into the fully connected layer, combined with the weight W of the fully connected layer c and bias b c , further integrate the features of the judgment essence, and output the optimized feature vector representation v′(Y i ) for the final optimization of the referee's essence.

[0147] S7 specifically includes:

[0148] S71: Match the judgment essence optimized in S6 with the historical cases in the judicial knowledge domain map. First, decompose the optimized judgment essence into multiple key elements, including case type, legal basis, dispute focus, etc., and then map these elements to the corresponding nodes in the map for subsequent correlation analysis;

[0149] S72: Calculate the similarity between the optimized judgment summary and the historical case summary in the graph, convert the words in the judgment summary into vector representations based on the word vector model, and then calculate the semantic similarity between the summary text and the historical case text. The similarity calculation is used to determine the degree of association between the current case and the historical case, ensuring that the most relevant case is matched;

[0150] S73: Screen the matched relevant historical cases and evaluate the correlation strength between different historical cases based on the correlation weights in the graph. The correlation weights in the graph are established based on multi-dimensional factors such as case type, legal provisions, and historical precedents. Each historical case has a different weight to measure its relevance to the current case.

[0151] S74: Generate a correlation index, specifically associate the current adjudication essence with the screened historical cases, and display the case type, adjudication result and related information of each historical case in the graph through the correlation index; the generation of the correlation index ensures that historical cases similar to the current case can be quickly retrieved in subsequent queries;

[0152] S75: Based on the generated association index, an index report is output, which includes the association details between the current case and the historical cases, the correlation score and the summary information of the historical cases, so as to help legal workers quickly understand the historical relevance of the current case; through semantic matching and correlation weight analysis, the step can effectively identify the association between the current case and the historical cases, and generate a structured association index, which provides a reference for legal workers and makes the handling of complex cases more organized and logical.

[0153] S8 specifically includes:

[0154] S81: Obtain the basic case summary from S3, extract the core facts of the case, including the case background, persons involved, focus of dispute and the course of the case, and perform standardization to form a basic facts paragraph as the opening part of the document;

[0155] S82: Integrate the judgment reasons and judgment summary generated in S4 into the document according to the predetermined format, ensuring that the judgment reasons are placed immediately after the basic facts of the case. The judgment reasons must include the court's legal analysis of the case, applicable legal provisions, and evaluation of the evidence. The judgment summary serves as a summary paragraph of the judgment reasons, which is used to briefly summarize the court's legal views and basis for the judgment.

[0156] S83: Based on the optimized judgment summary in S6, confirm whether its content complies with legal logic and language specifications, and insert the optimized judgment summary into the corresponding position of the document to ensure that the key points of the case are expressed concisely;

[0157] S84: Attach the associated index generated in S7 to the end of the case document. The associated index contains historical precedents and legal provisions related to the current case, helping readers to quickly refer to historical judgments and legal basis. The content of the associated index should be consistent with the main content of the document and presented in a structured form.

[0158] S85: Format the summarized documents to ensure that the document layout conforms to the standard format of judicial documents, including adding headers, paragraph numbers, titles and footers, and finally forming a complete case document. The above steps can generate a complete and clear case document by summarizing and splicing the contents of each part. The document content covers the facts of the case, reasons for the judgment, the essence of the judgment and related historical cases, ensuring that the document format conforms to the standard requirements of judicial documents and is convenient for subsequent use and reference.

[0159] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0160] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The case intelligent generation method based on judicial knowledge domain graph is characterized by: The following steps are involved: S1: constructing a judicial knowledge domain map, which includes information on case types, trial procedures, adjudication basis, and judgment results; S2: Parse the input multi-trial legal documents, extract the case facts and judgment reasons of the documents, and mark the newly added facts in the multi-trial cases; S3: Based on the judicial knowledge domain map constructed in S1 and the document information extracted in S2, the segmented rule extraction algorithm is used to classify and split the case facts in the documents to generate a preliminary summary of the basic case facts; S4: For the judgment reasons extracted from S2, the BERT+LSTM model is used to perform deep learning generative summarization to obtain the judgment reasons and judgment essence. S5: Combine the judicial knowledge domain map of S1 with the judgment reasons and judgment essence generated by S4 to compare the contents and verify their completeness and accuracy; S6: Using the results of the correlation comparison in S5, the convolutional neural network model is used to optimize the judgment essence, and enhance the legal logic of the judgment essence and the applicability of the case type; S7: Based on the judgment essence optimized by S6, the correlation index with other historical cases is automatically generated according to the correlation analysis in the judicial knowledge domain map; S8: Summarize the results of S3, S4, S6 and S7 into a complete case document and format it to meet the standard output requirements of judicial documents.

2. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S1 specifically includes: S11: Obtain historical judgment documents from an existing judicial database, wherein the historical judgment documents include first-instance judgments, second-instance judgments, and retrial judgments, wherein the data source is a public judicial database; S12: preprocessing the obtained judgment documents, including standardization of data format, cleaning of redundant information and elimination of abnormal data; S13: Based on the preprocessed data, a graph structure algorithm is applied to construct a judicial knowledge domain graph, with case types as nodes and case trial procedures, adjudication basis and judgment results as edges, forming a directed graph connecting different case types; S14: Label the nodes and edges in the graph, wherein the labels include the case type corresponding to each node, the trial procedure corresponding to each edge, the basis for adjudication, and relevant information of the judgment result.

3. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S2 specifically includes: S21: Perform text segmentation on the input multi-review legal documents, and preliminarily divide the case facts part and the judgment reasons part in the documents according to the fixed legal document format based on natural language processing technology. The division criteria include paragraph markers, legal terms and common document structures; S22: Use named entity recognition technology to extract predetermined elements from documents, including the persons involved, time, location, case type and case facts of the dispute; S23: Based on word frequency analysis and semantic parsing, extract the content of the judgment reasons, including the basis for the judgment, legal analysis and key points of the judgment; S24: Compare and analyze documents of multi-trial cases, and mark newly added case facts and judgment reasons in the second-instance or retrial documents based on timestamp information and semantic similarity calculations to ensure that no information in multi-trial documents is missed.

4. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S3 specifically includes: S31: Obtain the document information extracted in S2, including the case facts and judgment reasons, and preliminarily classify the documents according to the case type, trial procedure and judgment basis. The specific classification is based on the nodes and edges in the judicial knowledge domain graph constructed by S1; S32: applying a segmented rule extraction algorithm to classify case facts, wherein the segmented rule extraction algorithm performs classification based on a fixed paragraph structure of the document and a set of legal terms; S33: Classify and store the paragraphs that meet the rules in S32 according to categories, wherein the classified case fact information includes the time of occurrence of the case, the identities of the parties, and the elements of the litigation request; S34: Split the classified case fact information into specific events, dispute focus and key points of evidence according to the case type and trial procedure; S35: Based on the results of classification and splitting, a preliminary basic case summary is generated, which includes the main facts of the case, the focus of the dispute and the description of relevant evidence, and is formatted and output according to a standardized template.

5. The case intelligent generation method based on judicial knowledge domain graph according to claim 4 is characterized in that: The S32 specifically includes: S321: Define a set of rules, where each rule corresponds to a class of case facts; S322: Convert the input document paragraph into a word vector representation by using word embedding technology. Specifically, convert the words in each paragraph into corresponding numerical representations, and average the vectors of all words in the paragraph to obtain the vector representation of the entire paragraph. S323: for each paragraph, compare its vector representation with the vector representation of the predefined rule, and calculate the similarity between the paragraph and each rule. The higher the similarity, the closer the paragraph is to the rule. S324: According to the calculated similarity, classify the paragraph into the category corresponding to the rule with the highest similarity.

6. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S4 specifically includes: S41: pre-processing the text of the judgment reasons extracted in S2, and dividing the text into sentences and paragraphs; S42: Input each sentence into the BERT model, encode the words in each sentence, and generate a word vector representation with contextual relevance; S43: Input the vector of each sentence generated by BERT into the LSTM model to capture the contextual dependencies between sentences; S44: After LSTM processes all the sentences, an attention mechanism is used to generate a summary. The attention mechanism weights the hidden state of each sentence to identify the sentences that contribute more to the summary, thereby generating summary content. S45: Generate a summary of the reasons for the judgment and the essence of the judgment based on the hidden state of the LSTM through the decoder.

7. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S5 specifically includes: S51: Preliminarily match the judgment reasons and judgment essence generated in S4 with the historical cases in the judicial knowledge domain map constructed in S1, and extract the historical case information related to the current case based on the nodes of case type, judgment basis and judgment result in the map; S52: semantically match the judgment reasons generated by S4 with the judgment reasons of the same type of cases in the graph, and calculate the similarity between the two to ensure that the generated judgment reasons are consistent with the existing legal provisions and historical judgment logic; S53: Compare the judgment summary. Based on the historical case data extracted from the graph, use the semantic similarity algorithm to compare the generated judgment summary with the historical judgment summary to determine whether the judgment summary covers all the legal points. S54: Verify the completeness and accuracy of the reasons for the judgment and the essence of the judgment through the results of the correlation comparison. If any omissions or inconsistencies are found in the comparison, they will be marked and corresponding correction suggestions will be generated based on the legal provisions and historical case information in the map; S55: Finally, a verification report is generated, the comparison results and correction suggestions are output, and the generated reasons for judgment and judgment summary are adjusted according to the verification results.

8. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S6 specifically includes: S61: taking the judgment summary generated in S5 and the correction suggestions after the associated comparison as input, and converting the judgment summary text into a sequence of word vectors through word embedding technology, wherein each word is converted into a corresponding numerical representation to form a series of word vectors; S62: Input the word vector sequence into the convolution layer of the convolutional neural network. The convolution layer extracts local features of the word vector sequence through a sliding convolution kernel to identify important vocabulary combinations and expression patterns in the judgment essence. S63: The output of the convolution layer is processed by the pooling layer. The role of the pooling operation is to extract the most important features from the convolution result; S64: The pooled features are passed to the fully connected layer, where all local features are integrated into global features, and all extracted features are combined to generate the final feature representation of the judgment essence; S65: Based on the final feature representation, determine whether the judgment summary complies with the legal logic and language norms. If it does not meet the standards, feedback modification suggestions will be provided; if it meets the standards, the final optimized judgment summary text will be generated.

9. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S7 specifically includes: S71: Match the judgment essence optimized in S6 with the historical cases in the judicial knowledge domain map; S72: Calculate the similarity between the optimized judgment summary and the historical case summary in the graph, convert the words in the judgment summary into vector representations based on the word vector model, and then calculate the semantic similarity between the summary text and the historical case text; S73: Screen the matched relevant historical cases, and evaluate the correlation strength between different historical cases according to the correlation weights in the graph; S74: Generate a correlation index, specifically associate the current adjudication essence with the screened historical cases, and display the case type, adjudication result and related information of each historical case in the graph through the correlation index; S75: Output an index report based on the generated association index, which includes association details between the current case and the historical case, the correlation score, and summary information of the historical case.

10. The case intelligent generation method based on judicial knowledge domain graph according to claim 1 is characterized in that: The S8 specifically includes: S81: Obtain the basic case summary from S3, extract the core facts of the case, including the case background, persons involved, focus of dispute and the course of the case, and perform standardization to form a basic facts paragraph as the opening part of the document; S82: Integrate the judgment reasons and judgment summary generated in S4 into the document according to the predetermined format. The judgment summary serves as the summary paragraph of the judgment reasons, which briefly summarizes the court's legal opinions and judgment basis; S83: Based on the optimized judgment summary in S6, confirm whether its content complies with legal logic and language specifications, and insert the optimized judgment summary into the corresponding position of the document; S84: append the associated index generated in S7 to the end of the case document; S85: Format the summarized documents to ensure that the layout of the documents conforms to the standard format of judicial documents, including adding headers, paragraph numbers, titles and footers, and finally forming a complete case document.

Images