Legal text semantic reasoning report construction method and system

Abstract

The application relates to the technical field of natural language processing, and discloses a legal text semantic reasoning report construction method and system, which comprises the following steps: performing information decomposition processing on a legal text to obtain text sequence and text structure information; analyzing structure characteristic elements corresponding to effective corpus to extract relationship embedding vectors from initial structure information, analyzing relationship triplets corresponding to the relationship embedding vectors to generate a structured abstract corresponding to effective structure information; performing text segmentation processing on the effective corpus to obtain text segments, performing segmented iteration optimization on the effective corpus to obtain optimized segmented text; generating text target content corresponding to the legal text, calculating element correlation coefficients between legal element identifiers to determine correlation aggregation labels in semantic labels; and analyzing legal semantic intentions corresponding to the legal text to generate a reasoning report corresponding to the legal text. The application can improve the accuracy of legal text semantic reasoning report construction.

Description

Technical Field

[0001] This invention relates to a method and system for constructing semantic reasoning reports of legal texts, belonging to the field of natural language processing technology. Background Technology

[0002] In the legal field, various documents often exhibit highly specialized and structurally complex characteristics due to differences in historical evolution, professional habits, and expressive intentions. Although these texts contain core elements such as information about the parties involved, rights and obligations, conditions of conduct, and legal effects, their analysis presents significant challenges.

[0003] Existing methods mostly rely on keyword matching or statistical machine learning methods for content parsing. Keyword matching methods require the pre-compilation of legal terminology dictionaries, which are difficult to adapt to the increasingly refined judicial opinions and emerging legal concepts (such as "data ownership") in judicial practice, and are prone to parsing biases due to the lag in dictionary updates. Traditional machine learning methods can only extract surface semantics and lack in-depth exploration of the logical connections between legal elements (such as conditional relationships, causal relationships, or obligation-directed relationships), and cannot accurately construct the attribution structure between clauses and responsibilities. This results in logical inconsistencies or missing information in the final semantic reasoning report, thereby reducing the completeness and accuracy of legal text parsing. Summary of the Invention

[0004] This invention provides a method and system for constructing legal text semantic reasoning reports, the main purpose of which is to improve the accuracy of constructing legal text semantic reasoning reports.

[0005] To achieve the above objectives, the present invention provides a method for constructing a legal text semantic reasoning report, comprising:

[0006] The legal text to be processed is obtained, and the legal text is decomposed to obtain a text sequence and text structure information. The effective corpus and effective structural information in the text sequence and text structure information are then filtered out.

[0007] The structural feature elements corresponding to the effective corpus are analyzed to extract relation embedding vectors from the initial structural information, and the relation triples corresponding to the relation embedding vectors are analyzed to generate a structured summary corresponding to the effective structural information.

[0008] The effective corpus is processed into text segments to obtain text fragments. The segmentation error rate corresponding to the text fragments is calculated to perform segmentation iterative optimization on the effective corpus and obtain optimized segmented text.

[0009] Combining the structured summary and the optimized segmented text, the target text content corresponding to the legal text is generated, the legal element identifiers corresponding to the target text content are extracted, and the element correlation coefficients between the legal element identifiers are calculated to determine the associated aggregation tags in the semantic tags;

[0010] Based on the associated aggregation tags, the legal semantic intent corresponding to the legal text is analyzed, and a reasoning report corresponding to the legal text is generated by combining the legal semantic intent and the element association coefficient.

[0011] Optionally, the step of performing information decomposition processing on the legal text to obtain text sequence and text structure information includes:

[0012] The legal text is parsed to obtain the original text stream and a set of structural elements;

[0013] The original text stream is cleaned to obtain the target text stream;

[0014] Based on preset document type requirements, the target text stream is processed into sentence segments to obtain a text sequence;

[0015] Extract the logical association features corresponding to the text sequence, and combine the set of structural elements and the logical association features to analyze the text structure information.

[0016] Optionally, the step of filtering out effective corpus and effective structural information from the text sequence and the text structure information includes:

[0017] The text sequence is subjected to important annotation processing to obtain an annotated text sequence;

[0018] Sensitive information processing is performed on the labeled text sequence to obtain a processed text sequence;

[0019] The processed text sequence is then subjected to appropriate adjustment processing to obtain effective corpus;

[0020] Based on the effective corpus, the text structure information is checked and processed to obtain effective structure information.

[0021] Optionally, extracting the relation embedding vector from the initial structural information includes:

[0022] The structural feature elements are semantically encoded to obtain a feature semantic vector;

[0023] Identify the structural units in the initial structural information and establish a set of structural units;

[0024] Analyze the correlation between the feature semantic vector and the structural unit;

[0025] Based on the correlation, the associated structural units corresponding to the structural feature elements are selected from the structural units;

[0026] Based on the associated structural units, structural association information is extracted from the initial structural information;

[0027] The structural association information is vectorized to obtain the relation embedding vector.

[0028] Optionally, calculating the segmentation error rate corresponding to the text fragment includes:

[0029] The text fragments are analyzed for correspondence to obtain a set of correspondences;

[0030] Based on the set of correspondences, analyze the structural semantic features corresponding to the text fragments;

[0031] Based on the aforementioned structural semantic features, the correctness of the text segment is determined to obtain correct and incorrect segmentation.

[0032] The number of units corresponding to the correct segments and the incorrect segments are counted to obtain the number of correct segments and the number of incorrect segments;

[0033] The segmentation error rate corresponding to the text segment is calculated by combining the number of correct segments, the structural semantic features, and the number of incorrect segments.

[0034] Optionally, calculating the segmentation error rate corresponding to the text segment by combining the number of correct segments, the structural semantic features, and the number of incorrect segments includes:

[0035] Based on the structural semantic features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated;

[0036] By combining the number of correct segments and the number of incorrect segments, the error ratio corresponding to the text segment is calculated;

[0037] Based on the error ratio, set the correction coefficient corresponding to the text segment;

[0038] Combining the correction coefficient, the correct segmentation probability, the number of correct segments, and the number of incorrect segments, the segmentation error rate corresponding to the text fragment is calculated using the following formula:

[0039]

[0040] Where E represents the segmentation error rate corresponding to the text segment, α represents the correction coefficient, m represents the number of correct segments, n represents the number of incorrect segments, p_i represents the correct segmentation probability corresponding to the i-th segmentation unit in the text segment, and i represents the segmentation unit sequence number in the text segment.

[0041] Optionally, calculating the correct segmentation probability corresponding to each segmentation unit in the text segment based on the structural semantic features includes:

[0042] The structural semantic features are processed by constructing an association graph to obtain a structural meaning association graph;

[0043] The structural meaning association graph is subjected to feature integration processing to obtain an integrated feature matrix;

[0044] The integrated feature matrix is ​​smoothed and normalized to obtain a smoothed feature matrix;

[0045] The smooth feature matrix is ​​subjected to multidimensional fusion processing to obtain fused stable features;

[0046] Analyze the hierarchical levels of the fusion stable features and calculate the importance weights corresponding to the hierarchical levels;

[0047] By combining the importance weights and the fusion stability features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated.

[0048] Optionally, calculating the element association coefficient between the legal element identifiers includes:

[0049] Calculate the co-occurrence association strength among the legal element identifiers and determine the weight value corresponding to each legal element identifier;

[0050] Combining the co-occurrence association strength and the weight value, the element association coefficient between the legal element identifiers is calculated using the following formula:

[0051]

[0052] Where C represents the element correlation coefficient between legal element identifiers, w k S represents the weight value of the k-th legal element identifier. k，k+1 This represents the co-occurrence strength between the k-th legal element identifier and the (k+1)-th legal element identifier, where k and k+1 represent the sequence numbers corresponding to the legal element identifiers, M represents the total number of legal element identifiers, and N represents the number of other identifiers that co-occur with the k-th legal element identifier.

[0053] Optionally, the step of analyzing the legal semantic intent corresponding to the legal text based on the associated aggregated tags includes:

[0054] Analyze the meaning of the identifiers corresponding to the associated aggregation tags, and construct a meaning association network for the associated aggregation tags based on the identifier meanings;

[0055] The meaning association network is subjected to hierarchical grouping processing to obtain a hierarchical grouping network;

[0056] Extract the intent-driven elements from the hierarchical grouping network, perform intent comparison on the core elements of the network, and obtain preliminary semantic intent;

[0057] Calculate the logical density of the intent-driven element in the hierarchical grouping network;

[0058] Based on the logical tightness, the preliminary semantic intent is filtered to obtain the legal semantic intent corresponding to the legal text.

[0059] To address the aforementioned problems, this invention also provides a legal text semantic reasoning report construction system, the system comprising:

[0060] The text processing module is used to acquire the legal text to be processed, perform information decomposition processing on the legal text to obtain text sequence and text structure information, and filter out the effective corpus and effective structure information in the text sequence and text structure information.

[0061] The abstract extraction module is used to parse the structural feature elements corresponding to the effective corpus, extract relation embedding vectors from the initial structural information, analyze the relation triples corresponding to the relation embedding vectors, and generate a structured abstract corresponding to the effective structural information.

[0062] The segmentation iterative optimization module is used to perform text segmentation processing on the effective corpus to obtain text fragments, calculate the segmentation error rate corresponding to the text fragments, and perform segmentation iterative optimization on the effective corpus to obtain optimized segmented text.

[0063] The tag filtering module is used to combine the structured summary and the optimized segmented text to generate the text target content corresponding to the legal text, extract the legal element identifiers corresponding to the text target content, and calculate the element correlation coefficients between the legal element identifiers to determine the associated aggregation tags in the semantic tags;

[0064] The reasoning report generation module is used to analyze the legal semantic intent corresponding to the legal text based on the associated aggregation tags, and generate a reasoning report corresponding to the legal text by combining the legal semantic intent and the element association coefficient.

[0065] Compared to the problems described in the background art, this invention, by performing information decomposition processing on the legal text, can break down complex legal documents into information of different dimensions, providing a data foundation for subsequent effective information screening and effectively improving the efficiency of legal text information extraction. Furthermore, by analyzing the structural feature elements corresponding to the effective corpus, this invention can transform the complex legal structural information contained in the text into identifiable and processable key elements, making the structural features of the text information clearer and more operable, facilitating the discovery of the hidden legal logic and relationships behind the text. Moreover, by performing text segmentation processing on the effective corpus to obtain text fragments, this invention can initially divide continuous legal text into meaningful paragraph units, laying the foundation for subsequent processing, and through calculation... Calculating the segmentation error rate corresponding to the text fragment allows us to understand the accuracy of the current segmentation result, facilitating subsequent iterative optimization. Furthermore, this invention generates the target text content corresponding to the legal text by combining the structured summary and the optimized segmented text, thereby obtaining the core essence and complete expression of the legal text. By extracting the legal element identifiers corresponding to the target text content, key legal component markers in the text can be obtained, providing a foundation for the subsequent calculation of the element correlation coefficients between the legal element identifiers. Finally, this invention analyzes the legal semantic intent corresponding to the legal text based on the associated aggregation tags, thereby obtaining the core purpose and deeper meaning of the legal text, facilitating the generation of accurate legal reasoning reports. Therefore, the legal text semantic reasoning report construction method and system provided by this invention can improve the accuracy of legal text semantic reasoning report construction. Attached Figure Description

[0066] Figure 1 A flowchart illustrating a method for constructing a legal text semantic reasoning report according to an embodiment of the present invention;

[0067] Figure 2 A schematic diagram of the segmented iterative optimization process in the legal text structuring method provided by the present invention;

[0068] Figure 3 This is a schematic diagram of a module of a legal text semantic reasoning report construction system provided in an embodiment of the present invention.

[0069] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0070] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0071] This application provides a method for constructing a legal text semantic reasoning report. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the legal text semantic reasoning report construction method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0072] Reference Figure 1 The diagram shown is a flowchart illustrating a method for constructing a legal text semantic reasoning report according to an embodiment of the present invention. In this embodiment, the method for constructing a legal text semantic reasoning report includes:

[0073] S1. Obtain the legal text to be processed, perform information decomposition processing on the legal text to obtain text sequence and text structure information, and filter out the effective corpus and effective structure information in the text sequence and text structure information.

[0074] This invention, by performing information decomposition processing on the legal text, can break down complex legal documents into information of different dimensions, providing a data foundation for subsequent effective information screening and effectively improving the efficiency of legal text information extraction. The legal text refers to the legal document to be processed, containing various types of legal information that has not yet undergone information extraction and processing, such as contract documents and legal provisions. The text sequence refers to the continuous text information sequence extracted from the legal text, covering the core descriptive content of the legal document, such as clause content, definitions, and conditions. The text structure information refers to the result of formally displaying the key structures and potential features in the legal text, such as document outlines, chapter divisions, and list structures.

[0075] As an embodiment of the present invention, the step of performing information decomposition processing on the legal text to obtain text sequence and text structure information includes:

[0076] The legal text is parsed to obtain the original text stream and a set of structural elements;

[0077] The original text stream is cleaned to obtain the target text stream;

[0078] Based on preset document type requirements, the target text stream is processed into sentence segments to obtain a text sequence;

[0079] Extract the logical association features corresponding to the text sequence, and combine the set of structural elements and the logical association features to analyze the text structure information.

[0080] Wherein, the original text stream and the set of structural elements are the original content of the legal text and the parsed structural tags, respectively; the clean text stream is the text after removing irrelevant characters and formatting; the document type requirement is a document specification used to guide text segmentation, and the text sequence is the result of the clean text stream after segmentation based on this requirement: for example, the requirement includes "segmentation by chapter" and "retaining list items", etc., and the clean text stream obtains a structured text sequence after segmentation; the logical association feature is the logical relationship feature between texts in the text sequence.

[0081] Furthermore, the legal text can be formatted using text parsing techniques to obtain the original text stream and a set of structural elements, such as using PDF parsing tools to extract text and title tags; the original text stream can be cleaned using text cleaning algorithms to obtain a clean text stream, such as removing headers, footers, and extra spaces; based on document type requirements, the clean text stream can be segmented using natural language processing methods to obtain a text sequence, such as using sentence segmentation and paragraph recognition methods to divide the text into logical paragraphs and sentences; logical analysis methods can be used to extract the logical association features corresponding to the text sequence, such as identifying dependencies or conditional statements between clauses; combining the set of structural elements and the logical association features, structural analysis algorithms can be used to analyze the text structure information, such as using a tree structure to represent the hierarchical relationship of the document.

[0082] This invention obtains effective corpus and effective structural information by effectively filtering the text sequence and the text structure information. It can identify valuable legal text information, eliminate irrelevant or redundant parts, and ensure that the information entering the subsequent processing flow is relevant. The effective corpus and the effective structural information are the text and structural information obtained after removing invalid content from the text sequence and the text structure information, respectively.

[0083] As an embodiment of the present invention, the step of filtering out effective corpus and effective structural information from the text sequence and the text structural information includes:

[0084] The text sequence is subjected to important annotation processing to obtain an annotated text sequence;

[0085] Sensitive information processing is performed on the labeled text sequence to obtain a processed text sequence;

[0086] The processed text sequence is then subjected to appropriate adjustment processing to obtain effective corpus;

[0087] Based on the effective corpus, the text structure information is checked and processed to obtain effective structure information.

[0088] The labeled text sequence is the output of the text sequence after being pre-labeled using content importance assessment and classification methods, supplementing implicit meanings and importance information, and marking key clauses or importance levels; the processed text sequence is the result of blurring sensitive content (such as personal names and company names) in the labeled text sequence through techniques such as information hiding and simplification, such as replacing names in labeled text containing personal information with general identifiers; the effective corpus is the final text output that meets the current usage requirements after the processed text sequence has undergone applicability adjustment processing and is compared and screened with a requirement library that dynamically adjusts weights according to the current legal environment. If a new law repeals certain clauses, processed text sequences containing such clauses will be removed, and the effective parts will be retained to form the effective corpus.

[0089] Furthermore, the text sequence can be labeled with importance using a natural language processing-based importance classification method to obtain a labeled text sequence. For example, text classification technology can be used to identify important clauses such as "disclaimer clauses" or "key obligations" and label them as "high importance." Sensitive information processing can be performed on the labeled text sequence using an information fuzzification algorithm to obtain a processed text sequence. For example, a replacement algorithm can be used to replace specific names in legal text with category labels. The processed text sequence can be adjusted for applicability using an adaptive adjustment method to obtain valid corpus. For example, when the law is updated, the adjustment method automatically ignores outdated clauses and retains valid clauses. Based on the valid corpus, a structural consistency check method can be used to verify the consistency of the text structure information to obtain valid structural information. For example, comparing whether the chapter titles in the structural information match the content in the valid corpus; if the structural information shows chapters but the corpus does not contain content, the structural information is adjusted or marked as missing.

[0090] S2. Parse the structural feature elements corresponding to the effective corpus to extract relation embedding vectors from the initial structural information, analyze the relation triples corresponding to the relation embedding vectors, and generate a structured summary corresponding to the effective structural information.

[0091] This invention, by analyzing the structural feature elements corresponding to the effective corpus, can transform the complex legal structural information contained in the text into identifiable and processable key elements, making the structural features of the text information clearer and more operable, facilitating the discovery of the hidden legal logic and relationships behind the text. The structural feature elements refer to the key components extracted from the effective corpus that reflect the core structure and logical relationships of the legal text, such as subject relationship elements, rights and obligations elements, and conditional constraint elements. For example, by identifying the correspondence between the rights and obligations of the parties in the clauses, the rights and obligations elements can be obtained, and the balance of the legal relationship can be analyzed. Optionally, the analysis of the structural feature elements corresponding to the effective corpus can be implemented using structural analysis algorithms in natural language processing, such as using dependency parsing to analyze the grammatical structure of the text and extract the key structural feature elements.

[0092] This invention extracts relationship embedding vectors from the initial structural information, enabling the quantitative representation of the structural information of legal texts and constructing a numerical representation reflecting the relationships between legal elements, thus facilitating subsequent analysis and processing. The relationship embedding vectors are feature data in the initial structural information that mathematically express the relationships between legal elements, represented in vector form, reflecting the inherent connections between different elements in the legal text.

[0093] As an embodiment of the present invention, the step of extracting the relation embedding vector from the initial structural information includes:

[0094] The structural feature elements are semantically encoded to obtain a feature semantic vector;

[0095] Identify the structural units in the initial structural information and establish a set of structural units;

[0096] Analyze the correlation between the feature semantic vector and the structural unit;

[0097] Based on the correlation, the associated structural units corresponding to the structural feature elements are selected from the structural units;

[0098] Based on the associated structural units, structural association information is extracted from the initial structural information;

[0099] The structural association information is vectorized to obtain the relation embedding vector.

[0100] Wherein, the feature semantic vector is a numerical vector form obtained by semantic encoding of the structural feature elements, for example, encoding "Party A's obligations" into a multi-dimensional vector representation; the structural unit is the basic component in the initial structural information, corresponding to the structural blocks of the text, such as "chapter title" and "clause item"; the relevance is a measure of the degree of association between the feature semantic vector and the structural unit, for example, calculating the relevance between the feature vector and the content of the structural unit through similarity calculation; the associated structural unit is a unit selected from the structural units that has a strong association with the structural feature elements, for example, a structural unit containing a description of specific rights and obligations; based on the associated structural units, the connection relationship information between these units is extracted from the initial structural information, i.e., structural association information; the relationship embedding vector is a numerical representation obtained after vectorization processing of the structural association information.

[0101] Furthermore, semantic encoding of the structural feature elements can be performed using word embedding technology to obtain feature semantic vectors; structural units in the initial structural information can be identified using structural parsing methods to establish a set of structural units; the correlation between the feature semantic vectors and the structural units can be analyzed using attention calculation; based on the correlation, a threshold is set to filter out the associated structural units corresponding to the structural feature elements from the structural units; based on the associated structural units, structural association information can be extracted from the initial structural information using graph structure extraction methods; and the structural association information can be vectorized using a neural network encoder to obtain relation embedding vectors.

[0102] This invention analyzes the relation triples corresponding to the relation embedding vectors to generate structured summaries corresponding to the effective structural information. This transforms numerical relation representations into understandable structured language descriptions, more clearly showcasing the structured information of the text. The relation triples refer to subject-relationship-object triples identified through structural analysis techniques based on the relation embedding vectors. The structured summary is a textual description that summarizes the structural features and relationships of legal texts. Furthermore, the invention analyzes the relation triples corresponding to the relation embedding vectors, selects core triples, and organizes their logical connections, integrating them according to the framework format of effective structural information to generate structured summaries corresponding to the effective structural information.

[0103] S3. Perform text segmentation processing on the effective corpus to obtain text segments, calculate the segmentation error rate corresponding to the text segments, and perform segmentation iterative optimization on the effective corpus to obtain optimized segmented text.

[0104] This invention segments effective corpus into text fragments, which can initially divide continuous legal text into meaningful paragraph units, laying the foundation for subsequent processing. Furthermore, by calculating the segmentation error rate corresponding to the text fragments, the accuracy of the current segmentation results can be understood, thus facilitating subsequent iterative optimization.

[0105] The text fragments are independent content units focusing on a single legal theme, formed by splitting the effective corpus according to legal semantic logic and format specifications. The segmentation error rate represents the quantitative degree of logical incompleteness and semantic fragmentation corresponding to the text fragments, reflecting the degree of consistency between the segmentation results and the reasonable logic of the legal text. Furthermore, the effective corpus can be segmented using text segmentation methods to obtain text fragments. The text segmentation method refers to a processing method that divides continuous text into relatively independent units based on semantic integrity, grammatical structure, or specific domain requirements. It includes various segmentation criteria and processing methods, and the segmentation standard can be flexibly selected according to the characteristics of the text. For example, for legal texts, the text segmentation method can divide them into independent clause units or semantic paragraphs based on information such as clause markers, semantic integrity, and logical structure.

[0106] As an embodiment of the present invention, calculating the segmentation error rate corresponding to the text fragment includes:

[0107] The text fragments are analyzed for correspondence to obtain a set of correspondences;

[0108] Based on the set of correspondences, analyze the structural semantic features corresponding to the text fragments;

[0109] Based on the aforementioned structural semantic features, the correctness of the text segment is determined to obtain correct and incorrect segmentation.

[0110] The number of units corresponding to the correct segments and the incorrect segments are counted to obtain the number of correct segments and the number of incorrect segments;

[0111] The segmentation error rate corresponding to the text segment is calculated by combining the number of correct segments, the structural semantic features, and the number of incorrect segments.

[0112] The correspondence set is the set of associations between the organizational form and the meaning expression of the text fragment after correspondence analysis. For example, after analysis, "Party A shall pay the amount within three working days" presents a conditional adverbial clause plus a main clause, and the meaning expression reflects the correspondence between time conditions and behavioral obligations. The structural semantic features are the matching and coordination features between the structural organizational form and the meaning expression of the text fragment. The correct segmentation and the incorrect segmentation are the segmentation units of the text fragment that meet or do not meet the structural meaning coordination standard after the correctness judgment is made based on the structural semantic features. For example, a segment with a clear structure and complete meaning is a correct segmentation, while a segment with a chaotic structure and broken meaning is an incorrect segmentation.

[0113] Furthermore, the text fragments can be analyzed using language analysis tools to determine the correspondence between structure and meaning, resulting in a set of correspondences. For example, a grammar analyzer can be used to analyze sentence components, combined with semantic analysis tools to segment semantic units. Based on the set of correspondences, feature extraction and matching algorithms can be used to analyze the structural semantic features corresponding to the text fragments, such as determining the semantic coherence and structural integrity at segmentation boundaries. Based on the structural semantic features, classification methods can be used to determine the correctness of the text fragments, resulting in correct and incorrect segmentation.

[0114] Furthermore, as an optional embodiment of the present invention, the step of calculating the segmentation error rate corresponding to the text segment by combining the number of correct segments, the structural semantic features, and the number of incorrect segments includes:

[0115] Based on the structural semantic features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated;

[0116] By combining the number of correct segments and the number of incorrect segments, the error ratio corresponding to the text segment is calculated;

[0117] Based on the error ratio, set the correction coefficient corresponding to the text segment;

[0118] The segmentation error rate corresponding to the text segment is calculated by combining the correction coefficient, the correct segmentation probability, the number of correct segments, and the number of incorrect segments.

[0119] The error ratio is the ratio of the number of erroneous segments to the number of correct segments corresponding to the text fragment, reflecting the degree of error in the segmentation result; the correction coefficient is an adjustment parameter set for the text fragment to balance the influence of different segmentation error types; furthermore, based on the error ratio, a basic coefficient can be mapped through a mathematical function to set the correction coefficient corresponding to the text fragment.

[0120] Furthermore, as another embodiment of the present invention, the segmentation error rate corresponding to the text segment is calculated using the following formula, combining the correction coefficient, the correct segmentation probability, the number of correct segments, and the number of incorrect segments:

[0121]

[0122] Where E represents the segmentation error rate corresponding to the text segment, α represents the correction coefficient, m represents the number of correct segments, n represents the number of incorrect segments, p_i represents the correct segmentation probability corresponding to the i-th segmentation unit in the text segment, and i represents the segmentation unit sequence number in the text segment.

[0123] Furthermore, as an optional embodiment of the present invention, calculating the correct segmentation probability corresponding to each segmentation unit in the text segment based on the structural semantic features includes:

[0124] The structural semantic features are processed by constructing an association graph to obtain a structural meaning association graph;

[0125] The structural meaning association graph is subjected to feature integration processing to obtain an integrated feature matrix;

[0126] The integrated feature matrix is ​​smoothed and normalized to obtain a smoothed feature matrix;

[0127] The smooth feature matrix is ​​subjected to multidimensional fusion processing to obtain fused stable features;

[0128] Analyze the hierarchical levels of the fusion stable features and calculate the importance weights corresponding to the hierarchical levels;

[0129] By combining the importance weights and the fusion stability features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated.

[0130] Wherein, the structural meaning association graph is a graphical representation of the structural semantic features; the integrated feature matrix is ​​the feature integration result of the structural meaning association graph; the smoothed feature matrix is ​​a robust feature representation of the integrated feature matrix after smoothing and normalization; the fused stable feature is the multi-dimensional information fusion output of the stable feature matrix; and the hierarchical level is the different abstract levels of the fused stable feature.

[0131] Furthermore, the structural semantic features are processed by a graph construction method based on element co-occurrence relationships to obtain a structural meaning association graph. For example, the subjects, rights, and obligations in legal clauses are used as nodes, and the graph structure is constructed with "association" and "constraint" relationships as edges. The structural meaning association graph is then processed by a graph convolutional network to obtain an integrated feature matrix. For example, node features and adjacent node features are convolved and fused to generate a matrix containing global association information. The stable feature matrix is ​​then processed by a multi-dimensional fusion technique to obtain a fused stable feature. For example, syntactic features and semantic features are combined. After feature splicing, the features are weighted and fused using an attention mechanism. The hierarchical analysis module analyzes the hierarchical level of the fused stable features, such as dividing the features into chapter-section-item levels. The importance weights corresponding to the hierarchical levels are determined by weighted calculation, such as assigning values ​​according to the hierarchical influence: 0.4 for chapter level, 0.3 for section level, and 0.3 for item level. Finally, combining the importance weights and the fused stable features, the correct segmentation probability corresponding to each segment unit in the text segment is obtained through a calculation function. For example, the weighted fused features are input into the Sigmoid function, and the output is a probability value between 0 and 1.

[0132] This invention optimizes the effective corpus by segmenting iteratively based on the segmentation error rate to obtain optimized segmented text. This effectively improves the structural rationality and semantic integrity of text segmentation and reduces erroneous segmentation. The optimized segmented text is the optimal segmentation result obtained by iteratively adjusting the segmentation boundaries based on the segmentation error rate to maximize the structural and semantic consistency of the segmentation results. Further, for a clearer explanation of the segmentation iterative optimization process of the effective corpus, please refer to the following... Figure 2 This is a schematic diagram of the segmented iterative optimization process in the legal text structuring processing method provided by the present invention. The process is based on the segmentation error rate. On the one hand, it performs forward calculation of segmentation units to update features, and on the other hand, it optimizes the segmentation granularity standard and performs context consistency verification. Then, the segmentation processor after the above operations is used to reprocess the text, and finally the optimized segmented text is generated. Through multi-level optimization (feature update, standard optimization, consistency verification), the iterative improvement of text segmentation is achieved.

[0133] S4. Combining the structured summary and the optimized segmented text, generate the text target content corresponding to the legal text, extract the legal element identifiers corresponding to the text target content, and calculate the element correlation coefficients between the legal element identifiers to determine the associated aggregation tags in the semantic tags.

[0134] This invention generates the target text content corresponding to the legal text by combining the structured summary and the optimized segmented text, thereby obtaining the core essence and complete expression of the legal text. By extracting the legal element identifiers corresponding to the target text content, key legal component markers in the text can be obtained, providing a basis for the subsequent calculation and processing of the element correlation coefficients between the legal element identifiers.

[0135] The target text content is the complete content of the legal text obtained by integrating the structured summary and the optimized segmented text. The legal element identifier is the key legal component marker corresponding to the target text content. Furthermore, combining the structured summary and the optimized segmented text, the target text content corresponding to the legal text can be generated through content fusion and logical connection methods. If the structured summary is "An overview of the rights and obligations of both parties to the contract and the clauses on liability for breach of contract," and the optimized segmented text is the detailed content of the clauses after fine division, through content fusion and logical connection, the summary key points and detailed clauses are integrated to generate the target text content "This contract clearly stipulates the payment obligations of Party A and the delivery responsibilities of Party B, and the specific clauses include..."; the legal element identifier corresponding to the target text content can be extracted using an identifier extraction method.

[0136] This invention calculates the element correlation coefficient between the legal element identifiers to understand the degree of correlation between them, providing a quantitative basis for determining the associated aggregation tags in the subsequent legal element identifiers. The element correlation coefficient represents the strength of the inherent connection and the degree of logical association between the legal element identifiers in the legal text context.

[0137] As an embodiment of the present invention, calculating the element association coefficient between the legal element identifiers includes:

[0138] Calculate the co-occurrence association strength among the legal element identifiers and determine the weight value corresponding to each legal element identifier;

[0139] By combining the co-occurrence association strength and the weight value, the element association coefficient between the legal element identifiers is calculated.

[0140] Wherein, the co-occurrence association strength represents the degree of association between the legal element identifiers when they co-occur in the same context; the weight value represents the numerical value of the importance of the legal element identifier in the text. Furthermore, the co-occurrence association strength between the legal element identifiers can be calculated using statistical analysis methods, such as calculating point mutual information (PMI) or cosine similarity; the weight value corresponding to the legal element identifier can be determined through weighted calculation, such as calculating the weight based on the term frequency-inverse document frequency (TF-IDF) method.

[0141] Furthermore, as another embodiment of the present invention, the element association coefficient between the legal element identifiers is calculated using the following formula, combining the co-occurrence association strength and the weight value:

[0142]

[0143] Where C represents the element correlation coefficient between legal element identifiers, w k S represents the weight value of the k-th legal element identifier. k，k+1 This represents the co-occurrence strength between the k-th legal element identifier and the (k+1)-th legal element identifier, where k and k+1 represent the sequence numbers corresponding to the legal element identifiers, M represents the total number of legal element identifiers, and N represents the number of other identifiers that co-occur with the k-th legal element identifier.

[0144] This invention determines the associated aggregation tags in the legal element identifiers based on the element correlation coefficient, which can filter out logically close identifier combinations that meet the requirements of legal text specifications, thereby improving the efficiency and accuracy of subsequent text processing. The associated aggregation tags are a set of identifiers with strong logical correlation selected from the legal element identifiers by the element correlation coefficient. Furthermore, when the element correlation coefficient is greater than a preset correlation threshold, the associated aggregation tags in the legal element identifiers are determined. The preset correlation threshold can be set according to the actual legal text processing needs.

[0145] S5. Based on the associated aggregation tags, analyze the legal semantic intent corresponding to the legal text, and combine the legal semantic intent with the element association coefficient to generate a reasoning report corresponding to the legal text.

[0146] This invention analyzes the legal semantic intent corresponding to the legal text based on the associated aggregation tags, thereby obtaining the core purpose and deeper meaning of the legal text, so as to generate an accurate legal reasoning report in the future. The legal semantic intent is the core legal purpose and value orientation corresponding to the legal text.

[0147] As an embodiment of the present invention, the step of analyzing the legal semantic intent corresponding to the legal text based on the associated aggregation tags includes:

[0148] Analyze the meaning of the identifiers corresponding to the associated aggregation tags, and construct a meaning association network for the associated aggregation tags based on the identifier meanings;

[0149] The meaning association network is subjected to hierarchical grouping processing to obtain a hierarchical grouping network;

[0150] Extract the intent-driven elements from the hierarchical grouping network, perform intent comparison on the core elements of the network, and obtain preliminary semantic intent;

[0151] Calculate the logical density of the intent-driven element in the hierarchical grouping network;

[0152] Based on the logical tightness, the preliminary semantic intent is filtered to obtain the legal semantic intent corresponding to the legal text.

[0153] Wherein, the meaning association network is the meaning association structured network corresponding to the association aggregation tag; the hierarchical grouping network is the hierarchical structure network after the meaning association network has undergone hierarchical grouping processing; the intent-dominant element is the core meaning element in the hierarchical grouping network; the preliminary semantic intent is the preliminary intent judgment obtained by the network core elements through intent comparison; and the logical tightness is the degree of semantic and structural connection between the intent-dominant element in the hierarchical grouping network.

[0154] Furthermore, the meaning of the associated aggregated tags can be analyzed using semantic parsing methods; based on the meaning of the tags, a meaning association network corresponding to the associated aggregated tags can be constructed using network construction methods; the meaning association network can be hierarchically grouped using grouping clustering methods to obtain a hierarchical grouping network; the intent-dominant elements in the hierarchical grouping network can be extracted using element extraction techniques; the core elements of the network can be compared using intent comparison methods to obtain preliminary semantic intent, such as constructing a legal intent classification system and performing similarity calculations; when the logical tightness is lower than a preset tightness, the preliminary semantic intent is filtered to obtain the legal semantic intent corresponding to the legal text.

[0155] Furthermore, as an optional embodiment of the present invention, calculating the logical density of the intent-dominant element in the hierarchical grouping network includes:

[0156] Calculate the importance coefficient of the intent-driven element in the hierarchical grouping network;

[0157] The statistical analysis of the element hierarchy depth of the intended dominant element in the hierarchical grouping network;

[0158] By combining the importance coefficient and the element hierarchy depth, the logical density of the intention-dominant element in the hierarchical grouping network is calculated.

[0159] The importance coefficient represents the degree of quantification of the core position or connection criticality of the intention-driven element in the hierarchical grouping network; the element hierarchy depth represents the hierarchical position of the intention-driven element in the hierarchical grouping network or the path length from the core node.

[0160] Furthermore, the importance coefficient of the intention-dominant element in the hierarchical grouping network can be calculated using a centrality calculation method; the element hierarchy depth of the intention-dominant element in the hierarchical grouping network can be calculated by statistically analyzing the path length from element nodes to core nodes.

[0161] Furthermore, as another embodiment of the present invention, the logical density of the intended dominant element in the hierarchical grouping network is calculated by combining the importance coefficient and the element hierarchy depth using the following formula:

[0162]

[0163] Where T represents the logical tightness of the intention-dominant element in the hierarchical grouping network, and I... j G represents the importance coefficient of the j-th element among the dominant elements, β represents the hierarchical influence factor, and G represents the importance coefficient of the j-th element among the dominant elements. j Q represents the element hierarchy depth of the j-th element in the intent-dominant elements, where j represents the sequence number of the intent-dominant elements and Q represents the number of intent-dominant elements.

[0164] Furthermore, this formula uses an exponential decay function. The hierarchy depth reflects the attenuation effect of element importance. The deeper the element is, the smaller its contribution to the overall logical tightness. The sum of the products of the importance coefficient and the attenuation factor is taken as the average value to obtain the overall logical tightness measure. The hierarchy influence factor is used to adjust the degree of influence of hierarchy depth on importance attenuation. The hierarchy influence factor can be determined through empirical setting or statistical analysis.

[0165] This invention generates a reasoning report corresponding to the legal text by combining the legal semantic intent and the element correlation coefficient, thereby obtaining analytical conclusions and suggestions regarding the legal text and enhancing the depth of understanding and application value of the legal text. Furthermore, the reasoning report generated by combining the legal semantic intent and the element correlation coefficient first clarifies the core purpose of the text (such as confirmation of rights or definition of liability) based on the legal semantic intent, then checks the completeness of text elements (such as essential elements like subject, object, and content) and logical consistency based on the element correlation coefficient; finally, it combines the intent determination with the logical check results to generate a reasoning report containing a description of the text's purpose, an assessment of logical completeness, and application suggestions, such as "The text intent is for determining liability for breach of contract; the element correlation coefficient is 0.88; it is recommended to focus on proving impossibility of performance and fault."

[0166] Compared to the problems described in the background art, this invention, by performing information decomposition processing on the legal text, can break down complex legal documents into information of different dimensions, providing a data foundation for subsequent effective information screening and effectively improving the efficiency of legal text information extraction. Furthermore, by analyzing the structural feature elements corresponding to the effective corpus, this invention can transform the complex legal structural information contained in the text into identifiable and processable key elements, making the structural features of the text information clearer and more operable, facilitating the discovery of the hidden legal logic and relationships behind the text. Moreover, by performing text segmentation processing on the effective corpus to obtain text fragments, this invention can initially divide continuous legal text into meaningful paragraph units, laying the foundation for subsequent processing, and through calculation... Calculating the segmentation error rate corresponding to the text fragment allows us to understand the accuracy of the current segmentation result, facilitating subsequent iterative optimization. Furthermore, this invention generates the target text content corresponding to the legal text by combining the structured summary and the optimized segmented text, thereby obtaining the core essence and complete expression of the legal text. By extracting the legal element identifiers corresponding to the target text content, key legal component markers in the text can be obtained, providing a foundation for the subsequent calculation of the element correlation coefficients between the legal element identifiers. Finally, this invention analyzes the legal semantic intent corresponding to the legal text based on the associated aggregation tags, thereby obtaining the core purpose and deeper meaning of the legal text, facilitating the generation of accurate legal reasoning reports. Therefore, the legal text semantic reasoning report construction method and system provided by this invention can improve the accuracy of legal text semantic reasoning report construction.

[0167] like Figure 3 The diagram shown is a functional module diagram of the legal text semantic reasoning report construction system of the present invention.

[0168] The legal text semantic reasoning report construction system 200 of this invention can be installed in an electronic device. Depending on the functions implemented, the legal text semantic reasoning report construction system may include a text processing module 201, a summary extraction module 202, a segmented iterative optimization module 203, a tag filtering module 204, and a reasoning report generation module 205. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0169] In this embodiment of the invention, the functions of each module / unit are as follows:

[0170] The text processing module 201 is used to acquire the legal text to be processed, perform information decomposition processing on the legal text to obtain text sequence and text structure information, and filter out the effective corpus and effective structure information in the text sequence and text structure information.

[0171] The summary extraction module 202 is used to parse the structural feature elements corresponding to the effective corpus, extract relation embedding vectors from the initial structural information, analyze the relation triples corresponding to the relation embedding vectors, and generate a structured summary corresponding to the effective structural information.

[0172] The segmented iterative optimization module 203 is used to perform text segmentation processing on the effective corpus to obtain text fragments, calculate the segmentation error rate corresponding to the text fragments, and perform segmented iterative optimization on the effective corpus to obtain optimized segmented text.

[0173] The tag filtering module 204 is used to combine the structured summary and the optimized segmented text to generate the text target content corresponding to the legal text, extract the legal element identifiers corresponding to the text target content, and calculate the element correlation coefficients between the legal element identifiers in order to determine the associated aggregation tags in the semantic tags;

[0174] The reasoning report generation module 205 is used to analyze the legal semantic intent corresponding to the legal text based on the associated aggregation tags, and generate a reasoning report corresponding to the legal text by combining the legal semantic intent and the element association coefficient.

[0175] In detail, the modules in the legal text semantic reasoning report construction system 200 described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the one described in the article for constructing legal text semantic reasoning reports, and it can produce the same technical effect, so it will not be elaborated here.

[0176] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0177] Finally, it should be noted that in the above embodiments, each embodiment can be combined with each other or independent. Deleting any one of them will not affect the technical implementation of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a semantic reasoning report of legal texts, characterized in that: The method includes: The legal text to be processed is obtained, and the legal text is decomposed to obtain a text sequence and text structure information. The effective corpus and effective structural information in the text sequence and text structure information are then filtered out. The structural feature elements corresponding to the effective corpus are analyzed to extract relation embedding vectors from the text structure information. The relation triples corresponding to the relation embedding vectors are analyzed to generate a structured summary corresponding to the effective structural information. The step of extracting the relation embedding vector from the text structure information includes: The structural feature elements are semantically encoded to obtain a feature semantic vector; Identify structural units in the text structure information and establish a set of structural units; Analyze the correlation between the feature semantic vector and the structural unit; Based on the correlation, the associated structural units corresponding to the structural feature elements are selected from the structural units; Based on the associated structural units, structural association information is extracted from the text structural information; The structural association information is vectorized to obtain the relation embedding vector; The effective corpus is processed into text segments to obtain text fragments. The segmentation error rate corresponding to the text fragments is calculated to perform segmentation iterative optimization on the effective corpus and obtain optimized segmented text. The calculation of the segmentation error rate corresponding to the text segment includes: The text fragments are analyzed for correspondence to obtain a set of correspondences; Based on the set of correspondences, analyze the structural semantic features corresponding to the text fragments; Based on the aforementioned structural semantic features, the correctness of the text segment is determined to obtain correct and incorrect segmentation. The number of units corresponding to the correct segments and the incorrect segments are counted to obtain the number of correct segments and the number of incorrect segments; The segmentation error rate corresponding to the text segment is calculated by combining the number of correct segments, the structural semantic features, and the number of incorrect segments. The step of calculating the segmentation error rate corresponding to the text segment by combining the number of correct segments, the structural semantic features, and the number of incorrect segments includes: Based on the structural semantic features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated; By combining the number of correct segments and the number of incorrect segments, the error ratio corresponding to the text segment is calculated; Based on the error ratio, set the correction coefficient corresponding to the text segment; Combining the correction coefficient, the correct segmentation probability, the number of correct segments, and the number of incorrect segments, the segmentation error rate corresponding to the text fragment is calculated using the following formula: Where E represents the segmentation error rate corresponding to the text segment. This represents the correction factor, m represents the number of correct segments, and n represents the number of incorrect segments. This represents the correct segmentation probability corresponding to the i-th segmentation unit in the text segment, where i represents the segmentation unit sequence number in the text segment; By combining the structured summary and the optimized segmented text, the text target content corresponding to the legal text is generated, the legal element identifiers corresponding to the text target content are extracted, and the element correlation coefficients between the legal element identifiers are calculated to determine the association aggregation identifiers in the legal element identifiers; Based on the associated aggregation identifier, the legal semantic intent corresponding to the legal text is analyzed, and a reasoning report corresponding to the legal text is generated by combining the legal semantic intent and the element association coefficient.

2. The method for constructing a legal text semantic reasoning report as described in claim 1, characterized in that, The information decomposition process of the legal text to obtain text sequence and text structure information includes: The legal text is parsed to obtain the original text stream and a set of structural elements; The original text stream is cleaned to obtain the target text stream; Based on preset document type requirements, the target text stream is processed into sentence segments to obtain a text sequence; Extract the logical association features corresponding to the text sequence, and combine the set of structural elements and the logical association features to analyze the text structure information.

3. The method for constructing a legal text semantic reasoning report as described in claim 1, characterized in that, The process of filtering out effective corpus and effective structural information from the text sequence and the text structural information includes: The text sequence is subjected to important annotation processing to obtain an annotated text sequence; Sensitive information processing is performed on the labeled text sequence to obtain a processed text sequence; The processed text sequence is then subjected to appropriate adjustment processing to obtain effective corpus; Based on the effective corpus, the text structure information is checked and processed to obtain effective structure information.

4. The method for constructing a legal text semantic reasoning report as described in claim 1, characterized in that, The step of calculating the correct segmentation probability corresponding to each segmentation unit in the text segment based on the structural semantic features includes: The structural semantic features are processed by constructing an association graph to obtain a structural meaning association graph; The structural meaning association graph is subjected to feature integration processing to obtain an integrated feature matrix; The integrated feature matrix is ​​smoothed and normalized to obtain a smoothed feature matrix; The smooth feature matrix is ​​subjected to multidimensional fusion processing to obtain fused stable features; Analyze the hierarchical levels of the fusion stable features and calculate the importance weights corresponding to the hierarchical levels; By combining the importance weights and the fusion stability features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated.

5. The method for constructing a legal text semantic reasoning report as described in claim 1, characterized in that, The calculation of the element association coefficient between the legal element identifiers includes: Calculate the co-occurrence association strength among the legal element identifiers and determine the weight value corresponding to each legal element identifier; Combining the co-occurrence association strength and the weight value, the element association coefficient between the legal element identifiers is calculated using the following formula: Where C represents the element association coefficient between legal element identifiers. This represents the weight value of the k-th legal element identifier. This represents the co-occurrence strength between the k-th legal element identifier and the (k+1)-th legal element identifier, where k and k+1 represent the sequence numbers corresponding to the legal element identifiers, M represents the total number of legal element identifiers, and N represents the number of other identifiers that co-occur with the k-th legal element identifier.

6. The method for constructing a legal text semantic reasoning report as described in claim 1, characterized in that, The step of analyzing the legal semantic intent corresponding to the legal text based on the associated aggregation identifier includes: Analyze the meaning of the associated aggregation identifier, and construct the meaning association network corresponding to the associated aggregation identifier based on the identifier meaning; The meaning association network is subjected to hierarchical grouping processing to obtain a hierarchical grouping network; Extract the intent-dominant elements from the hierarchical grouping network, perform intent comparison on the intent-dominant elements, and obtain the preliminary semantic intent; Calculate the logical density of the intent-driven element in the hierarchical grouping network; Based on the logical tightness, the preliminary semantic intent is filtered to obtain the legal semantic intent corresponding to the legal text.

7. A legal text semantic reasoning report construction system, characterized in that, The system includes: The text processing module is used to acquire the legal text to be processed, perform information decomposition processing on the legal text to obtain text sequence and text structure information, and filter out the effective corpus and effective structure information in the text sequence and text structure information. The abstract extraction module is used to parse the structural feature elements corresponding to the effective corpus, extract relation embedding vectors from the text structure information, analyze the relation triples corresponding to the relation embedding vectors, and generate a structured abstract corresponding to the effective structural information. The step of extracting the relation embedding vector from the text structure information includes: The structural feature elements are semantically encoded to obtain a feature semantic vector; Identify structural units in the text structure information and establish a set of structural units; Analyze the correlation between the feature semantic vector and the structural unit; Based on the correlation, the associated structural units corresponding to the structural feature elements are selected from the structural units; Based on the associated structural units, structural association information is extracted from the text structural information; The structural association information is vectorized to obtain the relation embedding vector; The segmentation and iterative optimization module is used to perform text segmentation processing on the effective corpus to obtain text segments, calculate the segmentation error rate corresponding to the text segments, and perform segmentation and iterative optimization on the effective corpus to obtain optimized segmented text. The calculation of the segmentation error rate corresponding to the text segment includes: The text fragments are analyzed for correspondence to obtain a set of correspondences; Based on the set of correspondences, analyze the structural semantic features corresponding to the text fragments; Based on the aforementioned structural semantic features, the correctness of the text segment is determined to obtain correct and incorrect segmentation. The number of units corresponding to the correct segments and the incorrect segments are counted to obtain the number of correct segments and the number of incorrect segments; The segmentation error rate corresponding to the text segment is calculated by combining the number of correct segments, the structural semantic features, and the number of incorrect segments. The step of calculating the segmentation error rate corresponding to the text segment by combining the number of correct segments, the structural semantic features, and the number of incorrect segments includes: Based on the structural semantic features, the probability of correct segmentation corresponding to each segmentation unit in the text segment is calculated; By combining the number of correct segments and the number of incorrect segments, the error ratio corresponding to the text segment is calculated; Based on the error ratio, set the correction coefficient corresponding to the text segment; Combining the correction coefficient, the correct segmentation probability, the number of correct segments, and the number of incorrect segments, the segmentation error rate corresponding to the text fragment is calculated using the following formula: Where E represents the segmentation error rate corresponding to the text segment. This represents the correction factor, m represents the number of correct segments, and n represents the number of incorrect segments. This represents the correct segmentation probability corresponding to the i-th segmentation unit in the text segment, where i represents the segmentation unit sequence number in the text segment; The identifier filtering module is used to combine the structured summary and the optimized segmented text to generate the text target content corresponding to the legal text, extract the legal element identifiers corresponding to the text target content, and calculate the element correlation coefficients between the legal element identifiers to determine the association aggregation identifiers in the legal element identifiers; The reasoning report generation module is used to analyze the legal semantic intent corresponding to the legal text based on the associated aggregation identifier, and generate a reasoning report corresponding to the legal text by combining the legal semantic intent and the element association coefficient.

Images