Legal information analysis method and system, computer device, medium and product

Abstract

The invention discloses a legal affair information analysis method and system, a computer device, a medium and a product. The method comprises the following steps: analyzing legal affair information to be analyzed to generate an initial structured text; analyzing the initial reference law article based on the standard law article database, and updating when a law article analysis result does not meet requirements; logical reasoning verification is carried out on the initial logic information, and when a logic chain analysis result does not meet requirements, updating is carried out; comparing the initial analysis conclusion with a preset legal knowledge graph, determining text segments with potential errors, and performing confidence analysis to obtain a corresponding confidence analysis result; and according to the optimized initial reference law article, the optimized initial logic information and all confidence analysis results, generating a comprehensive, accurate and reliable analysis scheme. According to the method, data dependence can be reduced, the problems existing when a general large model is applied in the legal field are effectively solved, model output illusion is greatly relieved, and legal information analysis quality is improved.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a legal information analysis method and system, computer device, medium, and product. Background Technology

[0002] In recent years, large language model technology has shown rapid development. Open-source large models such as DeepSeek-R1, QWEN, and LLAMA have demonstrated powerful capabilities in natural language understanding and generation. However, the application of these general-purpose large models in specialized vertical fields such as law has been less than satisfactory. Specifically, this manifests in several ways: insufficient understanding of legal knowledge; while possessing some legal knowledge, the depth of understanding of legal terminology, concepts, and their complex relationships is insufficient, often leading to misinterpretations or inappropriate applications; a lack of legal reasoning ability; insufficient training in professional skills such as the application of legal provisions, case analysis, and legal reasoning; frequent professional errors; a tendency to generate "illusions" when dealing with legal issues, resulting in content that violates legal provisions; and poor adaptability, making it difficult to adapt to changes in legal systems across different countries and periods.

[0003] To address the challenges of applying general-purpose large models in the legal field, various solutions have been proposed. Full-scale fine-tuning involves collecting a large legal corpus to fine-tune all parameters of the model, as seen in models like ChatLaw. However, this method requires substantial labeled data and computational resources, resulting in extremely high costs. Retrieval-enhanced generation (RAG) leverages external knowledge bases to enhance model output, but it relies heavily on retrieval quality and struggles to handle complex legal reasoning. Expert system integration combines traditional legal expert systems with large models; however, rule maintenance is costly, and it struggles to handle boundary cases. Efficient parameter fine-tuning employs techniques like LoRA to fine-tune only a subset of parameters to reduce resource consumption, but simply using efficient parameter fine-tuning fails to meet the specialized needs of the legal field.

[0004] Current technologies still have several major shortcomings. First, obtaining high-quality labeled data in the legal field is difficult and costly, making it difficult for model training to overcome the data bottleneck. Second, legal knowledge has structured characteristics and strict logical relationships, while most existing methods use unstructured text for direct training, failing to fully utilize the structural characteristics of legal knowledge. Third, in scenarios where the model is fine-tuned with a small amount of data, it is prone to overfitting the training examples or forgetting its original capabilities. Fourth, legal tasks are diverse, covering consultation, drafting, review, dispute resolution, etc., but existing methods are difficult to adapt to multiple legal tasks simultaneously under low-resource conditions. Summary of the Invention

[0005] In view of this, the present disclosure provides a legal information analysis method and system, computer device, medium, and product, which can solve the problems in the prior art such as the difficulty and high cost of obtaining high-quality labeled data in the legal field, which leads to data bottlenecks in training, the failure to fully utilize the characteristics of legal knowledge structure, the tendency of fine-tuning with a small amount of data to overfit or forget the original capabilities, and the difficulty in adapting to diverse legal tasks under low resource conditions.

[0006] In a first aspect, embodiments of this disclosure provide a legal information analysis method, including: The legal information to be analyzed is based on a legal big data model to generate initial structured text; the initial structured text includes initial cited legal provisions, initial logical information, and initial analysis conclusions; The initial cited legal provisions are analyzed based on the standard legal provisions database to obtain legal provisions analysis results. When the legal provisions analysis results do not meet the preset legal provisions requirements, the corresponding legal provisions update strategy is invoked to optimize the initial cited legal provisions. Logical reasoning is performed on the initial logical information to obtain the logical chain analysis results; When the logic chain analysis result does not meet the preset logic requirements, the corresponding logic update strategy is invoked to optimize the initial logic information. The initial analysis conclusions are compared with a preset legal knowledge graph to identify text fragments with potential errors; A confidence analysis is performed on the text fragments with potential errors to obtain the corresponding confidence analysis results; An analysis scheme is generated based on the optimized initial references, the optimized initial logic information, and all the confidence analysis results.

[0007] Secondly, this disclosure also provides a legal information analysis system, including: The legal information analysis module is used to analyze the legal information to be analyzed based on the legal big data model and generate initial structured text; the initial structured text includes initial cited legal provisions, initial logical information, and initial analysis conclusions; The legal provision analysis and optimization module is used to analyze the initial referenced legal provisions based on the standard legal provision database, obtain the legal provision analysis results, and when the legal provision analysis results do not meet the preset legal provision requirements, call the corresponding legal provision update strategy to optimize the initial referenced legal provisions. The logic reasoning verification module is used to perform logic reasoning verification on the initial logic information to obtain the logic chain analysis result; when the logic chain analysis result does not meet the preset logic requirements, the corresponding logic update strategy is invoked to optimize the initial logic information. The conclusion comparison module is used to compare the initial analysis conclusions with a preset legal knowledge graph to identify text fragments with potential errors. The confidence analysis module is used to perform confidence analysis on the text fragments with potential errors and obtain the corresponding confidence analysis results. The analysis scheme generation module is used to generate an analysis scheme based on the optimized initial referenced legal provisions, the optimized initial logical information, and all the confidence analysis results.

[0008] Thirdly, this disclosure also provides a computer device, which adopts the following technical solution: The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the legal information analysis methods described above.

[0009] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing computer instructions for causing a computer to execute any of the legal information analysis methods described above.

[0010] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0011] The legal information analysis method disclosed in this application first analyzes the legal information to be analyzed based on a large-scale legal model to generate initial structured text. Then, it analyzes the initial cited legal provisions based on a standard legal provisions database to obtain legal provisions analysis results. If the legal provisions analysis results do not meet preset legal provisions requirements, the corresponding legal provisions update strategy is invoked to optimize the initial cited legal provisions. Next, logical reasoning verification is performed on the initial logical information to obtain logical chain analysis results. If the logical chain analysis results do not meet preset logical requirements, the corresponding logical update strategy is invoked to optimize the initial logical information. Finally, the initial analysis conclusions are compared with a preset legal knowledge graph to identify potentially erroneous text. This method performs confidence analysis on text fragments with potential errors, obtaining corresponding confidence analysis results. Finally, based on the optimized initial legal citations, optimized initial logical information, and all confidence analysis results, an analysis plan is generated. This method does not rely on a large amount of labeled data, effectively reducing the dependence on high-quality labeled data and thus effectively alleviating the data bottleneck problem. At the same time, it makes full use of the characteristics of legal knowledge structure to discover and correct potential errors. Moreover, it does not simply rely on model fine-tuning to improve performance, but uses a multi-dimensional verification and optimization mechanism to ensure the accuracy of the analysis results. It is highly professional, highly adaptable, and effectively reduces model output illusions.

[0012] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating the legal information analysis method provided in this embodiment of the disclosure.

[0015] Figure 2 A flowchart illustrating the optimized method for obtaining initial legal references provided in this embodiment of the disclosure.

[0016] Figure 3 This is a flowchart illustrating the method for obtaining logic chain analysis results provided in this embodiment of the disclosure.

[0017] Figure 4 A flowchart illustrating a method for determining potentially erroneous text fragments provided in embodiments of this disclosure.

[0018] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0019] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0020] Reference Figure 1 This application discloses a method for analyzing legal information, including: S100 analyzes the legal information to be analyzed based on the legal big data model and generates initial structured text; the initial structured text includes initial cited legal provisions, initial logical information, and initial analysis conclusions.

[0021] Transforming unstructured legal information into structured text facilitates subsequent processing and analysis. Through the processing of large-scale legal models, key information can be quickly extracted, providing a foundation for subsequent steps and improving work efficiency.

[0022] S200: Analyze the initial referenced legal provisions based on the standard legal provisions database to obtain the legal provisions analysis results. When the legal provisions analysis results do not meet the preset legal provisions requirements, call the corresponding legal provisions update strategy to optimize the initial referenced legal provisions.

[0023] Ensure that the cited legal provisions are accurate and applicable, avoid deviations in analysis results due to the use of outdated or incorrect legal provisions, improve the accuracy and reliability of legal application, and enhance the persuasiveness of analysis results.

[0024] S300: Perform logical reasoning verification on the initial logical information to obtain the logical chain analysis results; When the logic chain analysis result does not meet the preset logic requirements, the corresponding logic update strategy is invoked to optimize the initial logic information.

[0025] Ensuring the rigor of logical reasoning and avoiding logical loopholes or erroneous inferences is crucial. Verifying logical chains can uncover potential problems in the analysis process, improving the logicality and credibility of the results. Timely correction of logical errors makes the analysis process more reasonable and accurate, and optimized logical information can more objectively reflect the facts of the case, providing more reliable support for the final analysis plan.

[0026] S400 compares the initial analysis conclusions with the preset legal knowledge graph to identify text fragments with potential errors.

[0027] By leveraging the rich information in legal knowledge graphs, a comprehensive review of initial analytical conclusions can be conducted to identify potential errors or inconsistencies. Legal knowledge graphs can provide broader legal basis and references, helping to improve the accuracy of analytical conclusions.

[0028] S500 performs confidence analysis on text fragments with potential errors and obtains the corresponding confidence analysis results.

[0029] Quantifying the reliability of potentially erroneous text fragments provides a more specific reference for subsequent decision-making; confidence analysis can provide a clearer understanding of the problematic parts of the analysis conclusions, which helps to make targeted adjustments.

[0030] S600 generates an analysis scheme based on the optimized initial references, optimized initial logic information, and all confidence analysis results.

[0031] Taking into account the accuracy of legal provisions, the rigor of logic, and the reliability of analytical conclusions, the generated analysis plan is more comprehensive, objective, and accurate. At the same time, it can provide legal professionals with more valuable suggestions to help them make more reasonable decisions.

[0032] This embodiment reduces analytical bias caused by legal errors, logical loopholes, or unreasonable conclusions by comprehensively checking and optimizing legal provisions, logic, and analytical conclusions, making the analysis results more consistent with legal provisions and actual conditions. Each step has corresponding verification and optimization mechanisms, analyzing and evaluating legal information from different perspectives to ensure the rigor and reliability of the entire analysis process, effectively preventing legal errors and "illusions." The analysis scheme generated by this method comprehensively considers multiple factors, providing legal personnel with more comprehensive and accurate information, helping them to make more scientific and reasonable decisions. Based on a standard legal provisions database and legal knowledge graph, it can be updated and adapted to legal changes in a timely manner, ensuring that the analysis results always meet the latest legal requirements.

[0033] The legal information analysis method disclosed in this application does not simply rely on a large amount of labeled data for model training to ensure the accuracy of the analysis. Instead, after the initial structured text is generated through preliminary analysis by the large legal model, a series of subsequent verification and optimization steps are used to improve the quality of the analysis results. For example, the initial cited legal provisions are analyzed and optimized based on a standard legal provisions database, the initial logical information is verified based on logical reasoning, and the initial analysis conclusions are compared with a pre-set legal knowledge graph. These processes do not rely on a large amount of labeled data, reducing the dependence on high-quality labeled data and thus alleviating the data bottleneck problem.

[0034] The legal information to be analyzed is based on a large-scale legal model, generating initial structured text, including initial cited legal provisions, initial logical information, and initial analysis conclusions. This structured output fully reflects the structured characteristics of legal knowledge, organizing legal information according to different categories and relationships, facilitating subsequent analysis and processing. In subsequent steps, structured resources such as a standard legal provision database and a pre-set legal knowledge graph are utilized. The standard legal provision database stores legal provisions with strict structure and logical relationships. Through analysis and optimization of the initial cited legal provisions, the accuracy of the cited provisions and their conformity to the structural characteristics of legal knowledge are ensured. The pre-set legal knowledge graph contains rich legal concepts, rules, and case information, reflecting the complex relationships of legal knowledge. Comparing the initial analysis conclusions with this graph allows for better utilization of the structural characteristics of legal knowledge to identify and correct potential errors.

[0035] This application does not rely solely on model fine-tuning to improve performance, but rather uses a multi-dimensional verification and optimization mechanism to ensure the accuracy of the analysis results. The initial cited legal provisions, initial logical information, and initial analysis conclusions are verified and optimized separately. Each step has an independent judgment and adjustment mechanism. Even with a small amount of data for fine-tuning, these verification and optimization steps can discover and correct possible errors in the model, reduce the impact of overfitting, and at the same time retain the model's original natural language processing capabilities.

[0036] This method constructs a universal legal information analysis framework that can be used to process various legal tasks, including consultation, drafting, review, and dispute resolution. By generating initial structured text, it uniformly verifies and optimizes legal provisions, logic, and analytical conclusions, adapting to the needs of different legal tasks. For example, in contract review, it can verify the accuracy of cited contract-related legal provisions and the rationality of logical reasoning; similarly, in dispute resolution, these steps can be used to analyze the legal basis and logical relationships of a case. This universal analytical framework can flexibly handle various legal tasks under low-resource conditions, improving the method's adaptability and versatility.

[0037] The methods for constructing a pre-defined legal knowledge graph include: S110, Determine the constituent elements of the legal field's map.

[0038] The graph consists of elements related to legal principles, legal provisions, and cases. Elements related to legal principles include basic concepts in the legal field, elements related to legal provisions include legal provisions in the legal field, and elements related to cases include related cases in the legal field and the relationships between legal subjects in the legal field.

[0039] Specifically, basic concepts in the legal field include legal terminology, legal subjects (natural persons, legal persons, etc.), and legal objects (property rights, creditor's rights, etc.); legal provisions in the legal field include the Constitution, laws, administrative regulations, local regulations, judicial interpretations, and attribute information for each provision; attribute information includes one or more of the following: the effective date, level of validity, and revision history of the provision; related cases in the legal field include typical cases, judicial precedents, and other practical application scenarios; and legal subject relationships in the legal field include one or more of the following: application relationship, interpretation relationship, reference relationship, and priority relationship.

[0040] This step clearly defines the scope of the legal knowledge graph, providing a clear direction for subsequent knowledge architecture construction. It also covers elements at three levels: legal theory, legal provisions, and cases, and can comprehensively and systematically reflect the knowledge system of the legal field.

[0041] S120 constructs a jurisprudential knowledge architecture based on basic concepts in the legal field. The jurisprudential knowledge architecture includes several basic concept knowledge nodes and the relationship edges between different basic concept knowledge nodes.

[0042] For example, basic legal concepts collected can be used as knowledge nodes, such as "property rights," "ownership," and "usufruct rights." Each knowledge node contains information such as the name, definition, and related attributes of the concept. The relationships between different basic concepts can be analyzed. For example, since "ownership" is a specific type of "property rights," a "belonging" relationship edge can be established between the two knowledge nodes of "ownership" and "property rights." In this way, a network structure that reflects the logical relationships of legal principles can be constructed.

[0043] This step graphically illustrates the inherent logical relationships between basic legal concepts, which helps to deepen the understanding of the legal theory system and provides a foundation for legal reasoning. For example, relevant legal principles and rules can be inferred from the relationships between concepts.

[0044] S130, construct a legal provision-level knowledge architecture based on legal provisions in the legal field. The legal provision-level knowledge architecture includes several legal provision-type knowledge nodes and the relationship edges between different legal provision-type knowledge nodes.

[0045] Specifically, each legal provision is treated as a knowledge node, containing information such as its number, name, and specific content. The relationships between the provisions are analyzed; for example, some provisions may supplement, interpret, or restrict others. For instance, the provisions on special contracts in the Civil Code may supplement the provisions on general contracts, thus establishing a corresponding relationship between the knowledge nodes of these two provisions.

[0046] This step clearly demonstrates the interrelationships between legal provisions, making it easier to find and understand relevant provisions. When dealing with legal issues, it allows for the quick location and application of relevant provisions based on the relationships between them.

[0047] S140. Construct a case-level knowledge architecture based on related cases in the legal field and the relationships between legal subjects in the legal field. The case-level knowledge architecture includes several case-type knowledge nodes and the relationship edges between different case-type knowledge nodes.

[0048] Specifically, each related legal case is treated as a knowledge node, containing basic case information such as case number, parties, cause of action, and judgment. The similarities and connections between different cases are analyzed. For example, two cases involving the same legal issues and relationships between legal entities can establish a "similar" relationship edge. Simultaneously, based on the relationships between legal entities in the cases, connections are established between the case knowledge nodes and related entity knowledge nodes.

[0049] By examining the relationships between cases, we can summarize some common legal problem-solving patterns and adjudication rules; when dealing with new legal issues, we can quickly find relevant cases for reference and learning.

[0050] S150 is a legal domain knowledge graph based on a multi-layered organizational knowledge architecture encompassing jurisprudential, statute, and case-level knowledge structures.

[0051] Specifically, the jurisprudential knowledge architecture, the legal provision-level knowledge architecture, and the case-level knowledge architecture are integrated to establish connections between knowledge nodes at different levels. For example, jurisprudential concepts are associated with relevant legal provisions, and legal provisions are associated with cases involving those provisions, forming a knowledge graph with a multi-level structure. The top layer can be jurisprudential knowledge, the middle layer is legal provision-level knowledge, and the bottom layer is case-level knowledge. Different levels are interconnected through relational edges to form an organic whole.

[0052] Furthermore, the legal-related elements in the legal theory-level knowledge architecture, the legal-related elements in the legal provision-level knowledge architecture, and the case-related elements in the case-level knowledge architecture are each treated as independent knowledge nodes. Knowledge nodes at different levels are interconnected through relational edges, and each relational edge corresponds to a correlation weight, which is used to describe the relationship between different nodes.

[0053] By integrating knowledge from three levels—legal theory, legal provisions, and case law—a complete and systematic legal knowledge framework is formed. Users can query the knowledge graph from different dimensions, such as searching for relevant legal provisions and cases based on legal concepts, or searching for related legal theories and provisions based on cases. Simultaneously, cross-level knowledge reasoning is also possible, improving the efficiency of resolving legal issues.

[0054] The method disclosed in this embodiment systematically organizes and integrates complex legal knowledge by constructing a multi-layered organizational knowledge architecture legal domain knowledge graph, making the organization and management of legal knowledge more efficient. The clear relationship structure and rich knowledge information in the knowledge graph provide strong support for the analysis and resolution of legal issues, enabling rapid location of relevant legal principles, provisions, and cases, and improving the accuracy of legal reasoning and decision-making. The knowledge graph presents legal knowledge in a visual way, facilitating sharing and communication among different users, and contributing to the popularization and dissemination of legal knowledge. It also provides a foundation for intelligent applications in the legal field, such as intelligent legal consultation and automatic generation of legal documents, promoting the digital transformation of the legal industry.

[0055] Furthermore, this application also includes: establishing a knowledge acquisition and updating mechanism. Specifically, this includes: performing automated extraction, using natural language processing technology to automatically extract structured knowledge from publicly available data such as legal texts and judicial judgments.

[0056] Furthermore, expert review can be implemented, with legal experts reviewing and supplementing the automatically extracted knowledge to ensure its accuracy. In the expert review step, legal experts review and supplement the automatically extracted knowledge based on laws and regulations, legal principles, and judicial practice experience.

[0057] Furthermore, dynamic updates can be designed, and an incremental update mechanism can be built. When laws and regulations change, relevant knowledge nodes and relationships are automatically updated. In the dynamic update process, the official release channels of laws and regulations are monitored, and the incremental update mechanism is triggered when new laws and regulations are released or existing laws and regulations are revised.

[0058] Furthermore, version control can be implemented to maintain historical versions of the knowledge graph, in order to support legal consultations at specific points in time. In the version control process, version numbers are used to record different versions of the knowledge graph, and each version contains knowledge nodes and relationship information at the corresponding point in time.

[0059] Furthermore, multimodal representation can be performed to transform legal knowledge into various forms such as vectors, symbolic rules, and text descriptions, facilitating interaction with different levels of the model; deep learning models can be used to transform legal knowledge into vector form, and logical reasoning rules can be used to transform legal knowledge into symbolic rule form.

[0060] Furthermore, in the hierarchical organization process, legal principles can be used as the highest level, legal provisions as the intermediate level, and case studies as the lowest level for knowledge organization.

[0061] Furthermore, association strength quantification can be performed, assigning weights to the relationship edges in the knowledge graph to represent the association strength between different legal concepts. In the association strength quantification step, the weights of the relationship edges are determined based on factors such as the frequency of citations and the degree of dependence between legal concepts.

[0062] Reference Figure 2 The method S200, which "analyzes the initial cited legal provisions based on the standard legal provisions database to obtain the legal provisions analysis results, and calls the corresponding legal provisions update strategy to optimize the initial cited legal provisions when the legal provisions analysis results do not meet the preset legal provisions requirements," specifically includes the following: S210, Semantically encode the case description in the initial structured text to generate an initial cause-of-fact vector; S220, extract events from the initial structured text and generate initial element fingerprints; S230, based on the initial cause of action vector, the initial element fingerprint and the time-geographic key composed of the filing date and jurisdiction, a two-level search is performed in the standard legal provisions database to obtain candidate legal provisions texts; S240, Perform case description semantic encoding on the candidate legal text to generate candidate cause-of-case vectors; S250, extract events from candidate legal texts and generate candidate element fingerprints; S260, obtain the semantic similarity between the initial cause-of-fact vector and the candidate cause-of-fact vector; Obtain the fingerprint similarity between the initial requirement fingerprint and the candidate requirement fingerprint; Retrieve the validity of candidate legal provisions within the time and geographic key; The accuracy of the legal provisions is calculated based on semantic similarity, fingerprint similarity, and validity. S270, Calculate the applicability rate of the legal provision based on the validity status of the candidate legal text relative to the latest effective version; S280 constructs pseudo-legal provisions through a legal provision-counterexample generator, and calculates the log probabilities of the initial referenced legal provisions and pseudo-legal provisions using a legal big model. The game confidence score is obtained based on the difference in log probabilities after Sigmoid mapping. S290, when the accuracy rate of legal provisions, the applicability rate of legal provisions, and the confidence score of the game are all not lower than the corresponding thresholds, the candidate legal provisions texts are used as the optimized initial reference legal provisions.

[0063] For S210, assuming the initial structured text is a description of a sales contract dispute, including "Party A purchased a batch of electronic products from Party B, but Party B failed to deliver the goods on time as agreed, and Party A requested compensation for losses," a pre-trained legal language model (such as a legal-specific model based on BERT) is used to semantically encode this description, transforming the text into a fixed-length vector, such as a 768-dimensional vector, as the initial cause-of-action vector. Semantic encoding transforms the natural language case description into a computer-processable vector form, facilitating subsequent similarity calculations and retrieval operations. Different case descriptions, after encoding, can be compared in the vector space, helping to accurately match relevant legal provisions.

[0064] For S220, regarding the aforementioned sales contract dispute case description, event extraction technology is used to identify key events, such as "Party A and Party B signed a sales contract," "Party B failed to deliver the goods on time," and "Party A requested compensation for losses." These events are converted into binary codes, forming a fixed-length binary vector as the initial element fingerprint. For example, each event corresponds to a binary bit; if the event exists, the bit is 1, and if it does not exist, it is 0. The element fingerprint can extract the key elements of the case, representing the core information of the case in a concise form. In subsequent retrieval and comparison, comparing the element fingerprints can quickly determine the similarity between different cases, improving retrieval efficiency and accuracy.

[0065] For S230, for example, the time and location key is "October 1, 2024 - Haidian District, Beijing". First, a preliminary search is performed in the standard legal provisions database based on the initial cause of action vector to filter out a set of legal provisions with high similarity to the cause of action vector. Then, a secondary search is performed in this set based on the initial element fingerprint to further narrow down the scope and obtain candidate legal provisions. For example, the preliminary search yields 100 relevant legal provisions, and the secondary search selects 10 as candidate legal provisions. This two-layer search combines cause of action and element information, enabling more accurate filtering of legal provisions related to the case from the vast standard legal provisions database. The inclusion of the time and location key takes into account the timeliness and regional applicability of the law, ensuring that the candidate legal provisions are valid within a specific time and geographical scope.

[0066] For S240, for example, for the 10 selected candidate legal provisions, semantic encoding is performed using the same pre-trained legal language model as in step S210, converting each legal provision into a 768-dimensional vector as a candidate cause-of-case vector. Converting the candidate legal provisions into vector form facilitates similarity calculation with the initial cause-of-case vector. By comparing the similarity between vectors, the semantic matching degree between the candidate legal provisions and the case description can be evaluated.

[0067] For S250, for example, events are extracted from the texts of 10 candidate legal provisions, key events involved in each legal provision are identified, and these events are converted into binary codes to form candidate element fingerprints. Candidate element fingerprints can be used to compare with the initial element fingerprints to further evaluate the degree of matching between the candidate legal provisions and the case in terms of key elements. Combined with step S220, the relevance between the legal provisions and the case can be judged from different perspectives.

[0068] In S260, to obtain the semantic similarity between the initial cause of action vector and the candidate cause of action vector, the cosine similarity formula is used to calculate the similarity between the initial cause of action vector and the candidate cause of action vector, resulting in a value between 0 and 1. The closer the value is to 1, the more similar the semantics are.

[0069] To obtain the fingerprint similarity between the initial requirement fingerprint and the candidate requirement fingerprint, the fingerprint similarity is obtained by comparing the binary bits of the initial requirement fingerprint and the candidate requirement fingerprint and calculating the proportion of identical bits.

[0070] To obtain the validity of candidate legal texts within the time-domain key, specifically check whether the candidate legal texts are valid within the time-domain key formed by the filing date and jurisdiction. If valid, mark it as 1; if invalid, mark it as 0.

[0071] To calculate the accuracy of legal provisions based on semantic similarity, fingerprint similarity, and validity, specifically, semantic similarity, fingerprint similarity, and validity are considered comprehensively, with different weights assigned. For example, the weight of semantic similarity is 0.4, the weight of fingerprint similarity is 0.4, and the weight of validity is 0.2. The accuracy of the legal provisions is calculated by weighted summation. For example, if a candidate legal provision has a semantic similarity of 0.8, a fingerprint similarity of 0.7, and a validity of 1, then the accuracy of the legal provision = 0.8 × 0.4 + 0.7 × 0.4 + 1 × 0.2 = 0.76.

[0072] By comprehensively considering semantic similarity, fingerprint similarity, and validity, the matching degree between candidate legal provisions and cases can be fully evaluated. Different weight settings can be adjusted according to actual needs to highlight the importance of different factors and improve the accuracy of legal provision selection.

[0073] For S270, the applicability rate of a legal provision is calculated based on its validity status relative to the latest effective version, assuming that the standard legal provision database is updated in a timely manner with information on the latest effective versions of the provisions. For each candidate legal provision text, it is checked whether it is the latest effective version. If it is the latest version, the applicability rate is marked as 1; if it is an older version but still valid, an applicability rate value between 0 and 1 is assigned based on the degree of difference between it and the latest version (such as the number of clause amendments, their importance, etc.). For example, if a candidate legal provision is an older version but only a few clauses have been slightly amended, the applicability rate can be set to 0.8.

[0074] For S280, a pseudo-legal provision is constructed using a legal provision-counterexample generator. The logarithmic probabilities of the initial cited legal provision and the pseudo-legal provision are calculated using a large-scale legal model. The game-theoretic confidence score is obtained by mapping the difference in logarithmic probabilities using a Sigmoid function. The legal provision-counterexample generator can construct pseudo-legal provisions similar to the initial cited legal provision but containing errors or unreasonable aspects, based on a standard legal provision database and known legal rules. For example, for an initial cited legal provision on tort liability, the pseudo-legal provision might modify the liability determination conditions or compensation standards. A large-scale legal model (such as a specially trained legal language model) is used to process the initial cited legal provision and the pseudo-legal provision separately, calculating their logarithmic probabilities. Assume the logarithmic probability of the initial cited legal provision is -2, the logarithmic probability of the pseudo-legal provision is -3, and the difference between them is 1. This difference is mapped using a Sigmoid function to obtain the game-theoretic confidence score. The formula for the Sigmoid function is S(x) = 1 / (1+e^(-1 / ... −x Substituting the difference of 1, we obtain a game confidence score of approximately 0.73. The game confidence score can assess the reliability of the initially cited legal provision from another perspective. By comparing it with a pseudo-legal provision, it can reveal the advantages of the initially cited legal provision in terms of legal logic and rationality, providing additional basis for the selection of legal provisions and enhancing the reliability of the legal provision evaluation.

[0075] Considering the expiration date of legal provisions ensures that the selected provisions are up-to-date and applicable, avoiding the use of outdated or soon-to-expire provisions, thus improving the accuracy and reliability of legal application. Game theory confidence scores can assess the reliability of the initially cited legal provision from another perspective. By comparing it with pseudo-legal provisions, it can reveal the advantages of the initially cited provision in terms of legal logic and rationality, providing additional basis for judgment in the selection of legal provisions and enhancing the reliability of legal provision evaluation.

[0076] For S290, the thresholds for legal provision accuracy, applicability, and game theory confidence score were set at 0.7, 0.8, and 0.7, respectively. For the 10 candidate legal provisions, calculations and comparisons revealed that 3 candidate provisions had accuracy, applicability, and game theory confidence scores all above their respective thresholds. These candidate provisions can be used as the optimized initial references. By setting thresholds, the most suitable legal provisions can be selected, improving the quality of provision selection. Only when multiple evaluation indicators meet the requirements are candidate provisions used as optimized initial references, ensuring that the selected provisions have high levels of semantic matching, timeliness, and rationality.

[0077] By employing multi-dimensional evaluation indicators (semantic similarity, fingerprint similarity, validity, applicability, game-theoretic confidence score, etc.) and a multi-level retrieval mechanism, the system can more accurately select the most suitable legal provisions from the standard legal provisions database, reducing the possibility of incorrect legal provisions. It comprehensively considers multiple aspects such as the semantics, key elements, timeliness, and reasonableness of legal provisions, and introduces game-theoretic confidence scores to evaluate the reliability of legal provisions from another perspective, making the selection of legal provisions more comprehensive and reliable. Different weight settings and threshold settings can be adjusted according to the characteristics and needs of actual cases, improving the flexibility and adaptability of the solution. The automated legal provision screening and evaluation process can greatly reduce the time and workload of manually searching and judging legal provisions, improving the efficiency of legal application.

[0078] Reference Figure 3 The method for S300, "to perform logical reasoning verification on the initial logical information and obtain the logical chain analysis results," specifically includes the following methods for obtaining the logical chain analysis results: S310 performs chain-structured processing on the initial logical information to generate a set of conditional propositions, a set of essential propositions, and a conclusion proposition.

[0079] Structuring initial logical information can transform complex textual information into a clear set of propositions, facilitating subsequent logical analysis and processing. By clearly distinguishing between conditions, elements, and conclusions, logical relationships become more intuitive, helping to quickly locate and understand key information.

[0080] S320: Construct a directed causal graph based on the set of conditional propositions, the set of essential propositions, and the conclusion proposition, where nodes are propositions and edges are the reasoning relationships between propositions.

[0081] Directed causal graphs visually represent the logical relationships between propositions. The graphs clearly show how propositions are interconnected and derived, helping to identify potential logical flaws and reasoning paths, thus providing a visual foundation for subsequent testing.

[0082] S330 performs missing element detection, skip step detection, and causal inversion detection on the directed causal graph to obtain the missing element degree of the proposition, the skip step degree of the skip step reasoning, and the inversion degree of the reasoning edge with the wrong direction.

[0083] These three detection methods can comprehensively examine the logical chain and discover problems in logical reasoning from different perspectives; missing element detection can ensure that all necessary conditions and elements are considered; skip step detection can detect whether key steps are skipped in the reasoning process; and causal inversion detection can identify whether the reasoning direction is wrong, ensuring the correctness and completeness of logical reasoning.

[0084] S340 calculates the logic chain integrity score based on the degree of missing items, the degree of skipping steps, and the degree of inversion.

[0085] The logical chain integrity score is S: ,in, , , These are, respectively, the missing value, the skipped value, and the reversed value. , , These are the weights corresponding to the missing value, skip step value, and inversion value, respectively. For example, The preferred value is 0.5. The preferred value is 0.3. The preferred value is 0.2.

[0086] By comprehensively considering missing value, skipping degree, and inversion degree, a score is used to measure the completeness of the logical chain. This score can intuitively reflect the quality of the logical chain, making it easy to compare and evaluate different logical reasoning, and providing a quantitative basis for subsequent decision-making.

[0087] S350 When the logic chain integrity score is lower than the preset threshold, the logic chain analysis result is output. The logic chain analysis result includes a list of missing elements, a list of skip edges, and a list of reversed edges to trigger the subsequent logic update strategy.

[0088] When the logic chain integrity score is low, the output of detailed analysis results can help users quickly locate the problems in logical reasoning; the missing elements list can let users know which necessary conditions have not been considered; the skip edge list can point out the key steps skipped in the reasoning process; and the reverse edge list can help users correct the wrong reasoning direction, thereby improving and perfecting the logical reasoning.

[0089] The solution disclosed in this embodiment, through structured processing and multi-dimensional detection of initial logical information, can comprehensively identify various problems in logical reasoning, such as missing elements, skipped steps in reasoning, and causal inversion, thereby ensuring the accuracy of logical reasoning. The logical chain integrity score and detailed analysis results provide quantitative and visual basis for decision-making. In fields such as law and finance, accurate logical reasoning is crucial for decision-making. This solution can help decision-makers better evaluate the quality of logical reasoning and make more reliable decisions. The automated logical reasoning verification process can quickly complete the analysis of a large amount of logical information, avoiding the tediousness and error-proneness of manual inspection, greatly improving work efficiency. At the same time, detailed analysis results allow users to quickly locate problems, reducing the time and cost of error correction.

[0090] In S330, the integrity scan includes at least: a) missing element detection, used to identify missing element propositions and calculate the missing degree; b) skip step detection, used to identify skip step reasoning and calculate the skip step degree; c) causal inversion detection, used to identify reasoning edges with incorrect directions and calculate the inversion degree.

[0091] In the first embodiment, the method for obtaining the missing value includes: 1) determining the set of necessary component nodes based on a preset logical template; 2) matching the actual component nodes in the component proposition set with the component nodes in the set of necessary component nodes using a semantic similarity threshold to obtain the number of actual component nodes that meet the matching conditions; 3) taking the difference between the number of actual component nodes that meet the matching conditions and the total number of necessary component nodes in the set of necessary component nodes as the number of missing component nodes; 4) obtaining the missing value based on the number of missing component nodes and the total number of necessary component nodes in the set of necessary component nodes.

[0092] Suppose we are dealing with a legal analysis of a contract dispute case. The pre-defined logical template could be a set of rules based on relevant contract law provisions. For example, a valid contract requires elements such as the contracting parties, the subject matter of the contract, the contract price, the performance period, and the place and method of performance. Based on this template, these elements are explicitly listed as the set of necessary elements: {Contracting Parties, Subject Matter of the Contract, Contract Price, Performance Period, Place and Method of Performance}. This pre-defined logical template, developed based on professional legal knowledge and practical experience, provides a clear standard and framework for subsequent analysis. By identifying the set of necessary elements, the entire analysis process becomes systematic and ensures that we comprehensively examine the issue from a legal perspective, avoiding the omission of important legal elements.

[0093] To obtain the number of actual element nodes that meet the matching conditions by matching each actual element node in the set of element propositions with the element nodes in the set of necessary element nodes using a semantic similarity threshold, in the above contract dispute case, the set of element propositions we collected might be {Party A and Party B signed a contract involving a batch of electronic products, with a contract price of 100,000 yuan}. For "Party A and Party B signed a contract", we matched it with the set of necessary element nodes and found that it is semantically similar to "contract subject". When the semantic similarity threshold is set to 0.8, if the calculated semantic similarity between the actual element node and "contract subject" reaches 0.85, it is considered to meet the matching conditions. Similarly, "the contract involves a batch of electronic products" matches "contract subject matter", and "the contract price of 100,000 yuan" matches "contract price". Finally, the number of actual element nodes that meet the matching conditions is 3. Since the actual collected propositions may differ from the preset necessary elements in their wording, semantic similarity matching can more flexibly identify the correspondence between the actual elements and the necessary elements. The semantic similarity threshold can be adjusted according to the specific situation, so that the matching process is neither too lenient nor too strict, thus improving the accuracy and adaptability of the matching.

[0094] The number of missing requirement nodes is calculated by subtracting the number of actual requirement nodes that meet the matching criteria from the total number of necessary requirement nodes in the set of necessary requirement nodes. In the previous example, there were 5 necessary requirement nodes in the set of necessary requirement nodes, while 3 actual requirement nodes met the matching criteria. Therefore, the number of missing requirement nodes is 5 - 3 = 2, namely, the "performance period" and "performance location and method" are missing. This simple difference calculation provides a clear data foundation for calculating the missing requirement degree, making it simple, direct, and easy to understand and operate.

[0095] To determine the missingness based on the number of missing requirement nodes compared to the total number of necessary requirement nodes in the set of necessary requirement nodes, continuing with the previous example, if there are 2 missing requirement nodes and 5 necessary requirement nodes, then the missingness = 2 / 5 = 0.4, or 40%. The missingness is a quantitative indicator that clearly reflects the degree to which actual requirements are missing relative to necessary requirements. Through this indicator, we can quickly gain an intuitive understanding of the completeness of the analyzed object, facilitating comparison and evaluation between different cases.

[0096] The overall solution comprehensively examines the actual requirements from the perspective of legally mandated essential elements, ensuring that no important legal factors are overlooked. This helps in accurately grasping the key points of a case during legal analysis and case handling, avoiding legal risks caused by missing requirements. By calculating the degree of absence, the solution quantifies the absence of requirements, making the analysis results more objective and intuitive. This facilitates judgment and decision-making for legal professionals and decision-makers, providing significant reference value in assessing contract validity and case feasibility. The solution provides a systematic analysis process, making the requirement review process more standardized and regulated. Each step has clear objectives and methods, improving analysis efficiency and reducing subjectivity and arbitrariness in manual review processes.

[0097] In the second embodiment, the method for calculating the missing value includes: 1) determining the complete set of propositional elements required for the legal reasoning or business process based on relevant legal provisions, industry standards, or specific business rules. C= {c 1 ,c 2 ,⋯ ,c n } 2) Extract the set of actual constituent propositions from the given text, reasoning chain, or data. A= {a 1 ,a 2 ,⋯ ,a m } 3) Use set operations to find the set of missing essential propositions. M = C − A, that is, M = {c i ∣c i ∈C and c i ∉ A} ;4) A weight w needs to be assigned to each element proposition. i The weight reflects the importance of the element in the entire reasoning process. The weight can be determined based on expert experience, statistical data, or machine learning methods. The missing degree is D. m : For example, suppose the complete set of requirements is C={c1,c2,c3}, with weights w1=0.3, w2=0.4, and w3=0.3 respectively, and the actual set of required elements is A={c1}. Then the set of missing requirements is M={c2,c3}, and the degree of missingness is... .

[0098] The method for calculating the jump degree includes: 1) constructing a directed causal graph with initial logical information, which includes a set of condition nodes, a set of essential nodes, and a conclusion node; 2) performing a shortest directed path search from each condition node in the set of condition nodes to the conclusion node to obtain the shortest number of jumps; 3) when the shortest number of jumps is greater than a preset legal jump number threshold, marking the condition node-conclusion node pair as a jump pair; 4) dividing the number of jump pairs by the total number of condition nodes in the set of condition nodes to obtain the jump degree.

[0099] By performing a shortest directed path search and comparing it with a preset legal hop count threshold, this method can accurately identify instances where crucial steps are skipped during reasoning. In complex logical reasoning, various hidden logical jumps may exist, and this method can clearly expose these problems. For example, in fields such as legal reasoning and scientific argumentation, a small logical jump can lead to an incorrect conclusion. This method helps ensure that every step of the reasoning process is logically sound and improves the rigor of the logical chain. The preset legal hop count threshold is actually a standard set based on logical rules and professional knowledge. When the shortest hop count exceeds this threshold, it indicates that the reasoning has not followed normal logical steps. This helps to standardize the reasoning process, making the reasoning more consistent with logical rules and avoiding unreasonable jumps in reasoning. The calculation of the jump degree quantifies the jump situation in logical reasoning. By dividing the number of jump pairs by the total number of condition nodes, the jump degree is obtained, providing an objective numerical indicator for evaluating the quality of the logical chain. This indicator can intuitively reflect the severity of jump problems in logical reasoning, facilitating comparison and evaluation between different logical chains.

[0100] The method for obtaining the inversion degree includes: 1) constructing a directed causal graph with initial logical information, the directed causal graph containing multiple inference edges; 2) inputting the source node text and target node text corresponding to each inference edge into the causal direction discrimination model to output the probability of source → target; 3) when the probability is less than a preset direction threshold, marking the inference edge as an inversion edge; 4) dividing the number of inversion edges by the total number of inference edges to obtain the inversion degree.

[0101] In complex logical reasoning, causal relationships can be intricate, making manual identification of causal inversions challenging. A causal direction discrimination model can automatically and accurately determine the causal direction of each reasoning edge. Once the causal probability of an edge is found to be less than a preset direction threshold, it can be quickly marked as an inverted edge, efficiently locating errors in logical reasoning where causal relationships are reversed. By calculating the inversion degree, the severity of causal inversion problems in the entire logical chain can be clearly understood. The inversion degree is the ratio of the number of inverted edges to the total number of reasoning edges; this value directly reflects the proportion of causal inversion problems in the logical chain. This helps analysts quickly identify the approximate scope of the error, whether it is a local problem or a flaw in the overall logical structure, allowing for targeted corrections. Causality is one of the core elements of logical reasoning; causal inversion severely undermines the rationality of logic. This method rigorously checks the causal direction of each reasoning edge to ensure the correctness of causal relationships in the logical chain, thereby improving the overall quality of logical reasoning. In legal reasoning, accurate causal relationships are crucial for conviction and sentencing; this method can avoid misjudgments caused by causal inversion.

[0102] Reference Figure 4 The method of S400, which "compares the initial analysis conclusions with the preset legal knowledge graph to determine the text fragments with potential errors," that is, the method for determining the text fragments with potential errors, includes: S410: The initial analysis conclusions are split into sentences and quantized to generate a set of conclusion sentence vectors; S420: Compare the set of conclusion sentence vectors with the node vectors in the preset legal knowledge graph to identify entity conflicts, identify the support or negation relationship between entities in the conclusion and nodes in the graph, and mark the entity conflicts. S430, compare the numerical range of the initial analysis conclusions to identify whether the numerical values ​​in the conclusions fall within the official range of the corresponding conclusion entity in the legal knowledge graph, and mark numerical conflicts. S440, based on the filing date and jurisdiction of the initial analysis conclusion, performs a time-domain-geographic comparison of the legal knowledge graph to identify whether the conclusion entity is within the valid time period or region, and marks the time conflict; S450, calculate the overall error score for each conclusion sentence based on the weighted scores of entity conflict, numerical conflict and timeliness conflict; S460: When the overall error score is not lower than the preset threshold, the corresponding conclusion sentence is marked as a potential error text fragment, and the conflict type and corresponding node identifier are output for subsequent confidence analysis.

[0103] For S410, sentence segmentation can be based on punctuation marks. For example, regular expressions can be used to match common sentence-ending punctuation marks such as periods, question marks, and exclamation marks to divide the initial analysis conclusion text into individual sentences. For texts with omitted punctuation or complex sentence structures, natural language processing tools (such as NLTK and spaCy) can be used for syntactic analysis to accurately identify sentence boundaries. For vectorization, pre-trained word vector models such as Word2Vec, GloVe, or BERT based on the Transformer architecture can be used. For each segmented sentence, the word vectors of the words are combined. A common method is average pooling, which involves adding the word vectors of all words in the sentence and dividing by the number of words to obtain the sentence's vector representation. Segmenting the initial analysis conclusion into individual sentences allows subsequent conflict comparison and error analysis to be performed on specific semantic units, improving the accuracy and efficiency of processing. Converting sentences into vector form facilitates rapid information matching and comparison using vector calculation methods (such as similarity calculation), laying the foundation for subsequent entity conflict comparison and other operations.

[0104] For S420, each node text in the preset legal knowledge graph can be vectorized using the same word vector model as in step S410 to obtain node vectors. For each sentence vector in the conclusion sentence vector set, the cosine similarity formula can be used to calculate its similarity with all node vectors in the legal knowledge graph. The two parameters in the cosine similarity formula are the sentence vector and the node vector, respectively. When the similarity is higher than a set threshold, the semantics of the sentence are further analyzed to determine whether the entity in the conclusion and the graph node have a supporting or negative relationship. This can be done using rule-based methods or machine learning classifiers. If the relationship is determined to be negative, the conclusion sentence is marked as having an entity conflict with the graph node, and relevant information is recorded. By comparing with nodes in the legal knowledge graph, logical contradictions between entities in the conclusion and legal knowledge can be detected in a timely manner, which helps to identify erroneous legal reasoning and judgments. Marking entity conflicts can remind analysts to correct problematic parts, thereby improving the accuracy and reliability of the initial analysis conclusions.

[0105] For S430, regular expressions can be used to extract numerical information from the initial analysis conclusion. The node corresponding to the entity in the conclusion is then located in the legal knowledge graph. It is determined whether the node contains official numerical range information, and the extracted value is compared with the official numerical range of the corresponding node in the graph. If the value is not within the range, it is marked as a numerical conflict. In legal analysis, the accuracy of numerical values ​​is crucial. Comparing numerical ranges ensures that the values ​​in the conclusion comply with legal regulations, avoiding legal risks caused by numerical errors. Clear numerical range comparison helps to standardize the legal reasoning process, making the analysis conclusion more rigorous and reliable.

[0106] For S440, named entity recognition technology (such as a deep learning-based BiLSTM-CRF model) can be used to extract the case filing date and jurisdiction information from the initial analysis conclusions. The statute of limitations information related to the conclusion entity is searched in the legal knowledge graph, and the case filing date is compared with the valid time period. If the case filing date is not within the valid time period, it is marked as a statute of limitations conflict. Similarly, the geographical information related to the conclusion entity is searched in the legal knowledge graph, and the jurisdiction is compared with the specified geographical area. If the jurisdiction is not within the prescribed geographical area, it is marked as a statute of limitations conflict. Through time-domain-geographic comparison, it can be ensured that the applicable law in the analysis conclusions conforms to specific time and geographical requirements, avoiding legal errors caused by statute of limitations or geographical issues. Considering the timeliness and geographical factors of the law makes the legal analysis more rigorous and comprehensive, enhancing the credibility of the analysis conclusions.

[0107] For S450, specific weights are assigned to entity conflicts, numerical conflicts, and timeliness conflicts. For example, entity conflicts are weighted at 0.5, numerical conflicts at 0.3, and timeliness conflicts at 0.2. For each conclusion sentence, a score is assigned based on whether a corresponding conflict exists. A conflict is scored as 1 point, and the absence of a conflict is scored as 0 points. The scores for entity conflicts, numerical conflicts, and timeliness conflicts are multiplied by their respective weights and then summed to obtain the overall error score for each conclusion sentence. Calculating the overall error score using a weighted scoring method allows for a comprehensive consideration of the importance of different types of conflicts, providing a more complete assessment of the error severity of each conclusion sentence. The overall error score provides a quantitative indicator for subsequent screening of potentially erroneous text fragments, making the screening process more objective and efficient.

[0108] For S460, a preset threshold for the overall error score can be set based on actual needs and experience, such as 0.6. The overall error score of each conclusion sentence is compared with this preset threshold. When the overall error score is not lower than the preset threshold, the corresponding conclusion sentence is marked as a potentially erroneous text fragment, and its conflict type (entity conflict, numerical conflict, or statute of limitations conflict) and the corresponding legal knowledge graph node identifier are output. By setting a threshold to filter out potentially erroneous text fragments, analysts can focus their attention on potentially problematic parts, improving the efficiency of error analysis. The output conflict type and corresponding node identifier provide detailed information for subsequent confidence analysis, helping to further evaluate the reliability of the analysis conclusions.

[0109] The S500 method for "performing confidence analysis on text segments with potential errors and obtaining corresponding confidence analysis results" specifically includes: S510 performs semantic drift detection on text segments with potential errors and generates drift confidence; S520, performs numerical uncertainty quantification on text segments with potential errors and generates numerical confidence scores; S530, perform graph consistency backtracking on text segments with potential errors to generate graph confidence; S540 performs a weighted fusion of drift confidence, numerical confidence, and graph confidence to obtain a comprehensive confidence score; S550: When the overall confidence level is lower than the preset threshold, text segments with potential errors are marked as high-risk segments, and the corresponding confidence analysis results are output to trigger subsequent correction strategies.

[0110] Regarding S510, "Semantic drift detection is performed on text fragments with potential errors to generate drift confidence scores." For example, suppose a legal document states, "The damage caused by this infringement includes direct economic losses and indirect economic losses, where indirect economic losses refer to the reduction in the company's expected profits over the next five years." However, in relevant legal knowledge systems, indirect economic losses are usually defined as the loss of expected profits due to the infringement, not limited to a reduction in expected profits over the next five years. By comparing the text fragment with standard legal semantics, a certain degree of semantic deviation is found. A corpus-based method can be used to calculate the semantic similarity between the text fragment and the standard legal expression. If the similarity is 0.6, it can be used as the drift confidence score, i.e., the drift confidence score is 0.6. Semantic drift can lead to misunderstandings and misapplications of legal provisions. Through semantic drift detection, it is possible to promptly identify whether the text deviates semantically from the standard legal definition and interpretation. The drift confidence score can quantify the degree of this deviation, providing an important basis for subsequent assessment of error risk.

[0111] Regarding S520, "Quantifying the numerical uncertainty of potentially erroneous text fragments to generate numerical confidence scores," for example, a legal document concerning compensation amounts might state that "the infringing party must compensate the victim for losses of approximately 500,000 to 600,000 yuan." Here, the value is a range and inherently uncertain. This uncertainty can be quantified by analyzing the fluctuation range of compensation amounts in similar historical cases and the statistical variance of the data. Assuming that the uncertainty level of this range is 30% after analysis, the numerical confidence score can be set as 1 - 0.3 = 0.7. In legal analysis, the accuracy of numerical values ​​is crucial. Numerical uncertainty quantification quantifies the degree of uncertainty of numerical values ​​in text, and the numerical confidence score reflects the reliability of the value, helping to assess the risk of the numerical portion in potentially erroneous text fragments.

[0112] Regarding S530's requirement to "perform graph consistency backtracking on text fragments with potential errors to generate graph confidence scores," for example, a legal knowledge graph explicitly states that "breach of contract requires liability for breach of contract, including payment of liquidated damages and compensation for losses, with liquidated damages not exceeding 30% of the contract value." However, a potentially erroneous text fragment states that "after a breach of contract, the breaching party must pay liquidated damages equal to 50% of the contract value." Comparing this text fragment with the legal knowledge graph reveals inconsistencies. The graph confidence score can be determined based on the authority of the relevant rules in the graph and the degree of deviation between the text and the graph. If the rule is considered highly authoritative and the deviation is significant, the graph confidence score can be set to 0.2. Legal knowledge graphs are constructed based on professional legal knowledge and possess high authority and accuracy. Through graph consistency backtracking, it is possible to check whether text fragments are consistent with the rules and knowledge in the legal knowledge graph. The graph confidence score reflects the degree of conformity between the text and authoritative legal knowledge, helping to identify potential legal errors.

[0113] For S540, "Weighing and fusing the drift confidence, numerical confidence, and graph confidence to obtain the overall confidence score," assuming the weights for drift confidence, numerical confidence, and graph confidence are set to 0.3, 0.2, and 0.5 respectively, and given the previously calculated drift confidence score of 0.6, numerical confidence score of 0.7, and graph confidence score of 0.2, the overall confidence score = 0.6 × 0.3 + 0.7 × 0.2 + 0.2 × 0.5 = 0.42. Different types of confidence scores reflect the reliability of text fragments from different perspectives. Weighted fusion can comprehensively consider these factors to obtain a comprehensive and objective overall confidence score, more accurately assessing the overall reliability of the text fragment.

[0114] Regarding S550, which states that "when the overall confidence level is lower than a preset threshold, text segments with potential errors are marked as high-risk segments, and the corresponding confidence analysis results are output to trigger subsequent correction strategies," specifically, the preset threshold is set to 0.5. The previously calculated overall confidence level is 0.42, which is lower than the preset threshold. Therefore, the text segment with potential errors is marked as a high-risk segment, and confidence analysis results such as drift confidence level 0.6, numerical confidence level 0.7, graph confidence level 0.2, and overall confidence level 0.42 are output. Subsequent correction strategies can be triggered, such as re-examining the text or consulting legal experts. By setting a preset threshold, text segments with low reliability can be quickly filtered out and marked as high-risk segments, facilitating timely corrective measures. The output confidence analysis results provide detailed information for the correction process, helping analysts accurately identify problems and improve the efficiency and accuracy of corrections.

[0115] The overall solution conducts confidence analysis on potentially erroneous text fragments from multiple perspectives, including semantics, numerical values, and knowledge graph consistency. This comprehensive and in-depth assessment of text reliability and error risk avoids the limitations of single-factor evaluation. By generating various confidence levels and weighted fusion to obtain a comprehensive confidence score, the error risk of text fragments is quantified, making risk assessment more objective and accurate. This helps legal analysts and decision-makers intuitively understand the risk status of texts and make reasonable decisions. By marking high-risk fragments and triggering subsequent correction strategies, potential errors in legal texts can be identified and corrected in a timely manner, improving the quality and accuracy of legal texts and reducing legal disputes and risks caused by textual errors.

[0116] Specifically, S600 "generates an analysis scheme based on the optimized initial cited legal provisions, optimized initial logical information, and all confidence analysis results." In cases where there is a dispute over the legal interpretation, it can generate multiple possible interpretation schemes and explain the basis for each.

[0117] Based on all the confidence analysis results, the system can provide prompts to users (e.g., "This part has a confidence score of 0.7, which may require manual verification") based on the confidence analysis results (i.e., quantified confidence scores) corresponding to the uncertain parts of the results, thereby enhancing the transparency of the output.

[0118] The analysis can be multi-option, for example, for legal disputes (e.g., case law disagreements), generating 2-5 alternative interpretations, each with supporting evidence (e.g., "Option 1: No liability, based on case law X; Option 2: Partial liability, based on legal provision Y") to help users make decisions. This further ensures that the output is credible rather than a single viewpoint, and is applicable to complex legal scenarios.

[0119] Traditional models outputting a single result can be misleading; this solution introduces "uncertainty quantification + multi-perspective generation", combining KG (Knowledge Graph) and confidence model to improve robustness (Appendix: user satisfaction increased by 20%). Improvements include dynamic threshold (multiple solutions triggered when confidence <0.8) and evidence traceability (automatically extracted from KG).

[0120] Furthermore, it can record and analyze common model errors for continuous optimization of the model and knowledge graph; it can collect feedback from professional users for model fine-tuning and knowledge updates. When laws and regulations are updated, it can automatically identify the knowledge areas and model parameters that need to be updated.

[0121] Secondly, this application discloses a legal information analysis system for executing the legal information analysis method disclosed in the first aspect of this application. The system includes: The legal information analysis module is used to analyze the legal information to be analyzed based on the legal big data model and generate initial structured text. The initial structured text includes initial cited legal provisions, initial logical information, and initial analysis conclusions. The legal provision analysis and optimization module is used to analyze the initial referenced legal provisions based on the standard legal provision database, obtain the legal provision analysis results, and when the legal provision analysis results do not meet the preset legal provision requirements, call the corresponding legal provision update strategy to optimize the initial referenced legal provisions. The logic reasoning verification module is used to perform logic reasoning verification on the initial logic information and obtain the logic chain analysis results. When the logic chain analysis results do not meet the preset logic requirements, the corresponding logic update strategy is called to optimize the initial logic information. The conclusion comparison module is used to compare the initial analysis conclusions with the preset legal knowledge graph to identify text fragments with potential errors. The confidence analysis module is used to perform confidence analysis on text fragments with potential errors and obtain the corresponding confidence analysis results. The analysis scheme generation module is used to generate an analysis scheme based on the optimized initial references, optimized initial logic information, and all confidence analysis results.

[0122] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0123] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the legal information analysis methods of the foregoing embodiments of this disclosure.

[0124] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0125] like Figure 5This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 5 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0126] like Figure 5 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0127] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although Figure 5 A computer apparatus with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or included alternatively.

[0128] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the legal information analysis method of embodiments of this disclosure are performed.

[0129] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0130] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the legal information analysis methods described in the foregoing embodiments of the present disclosure are performed.

[0131] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0132] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0133] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0134] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0135] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0136] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0137] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0138] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0139] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for analyzing legal information, characterized in that, include: The legal information to be analyzed is based on the legal big data model to generate initial structured text; The initial structured text includes initial legal citations, initial logical information, and initial analysis conclusions; The initial cited legal provisions are analyzed based on the standard legal provisions database to obtain legal provisions analysis results. When the legal provisions analysis results do not meet the preset legal provisions requirements, the corresponding legal provisions update strategy is invoked to optimize the initial cited legal provisions. Logical reasoning is performed on the initial logical information to obtain the logical chain analysis results; When the logic chain analysis result does not meet the preset logic requirements, the corresponding logic update strategy is invoked to optimize the initial logic information. The initial analysis conclusions are compared with a preset legal knowledge graph to identify text fragments with potential errors; A confidence analysis is performed on the text fragments with potential errors to obtain the corresponding confidence analysis results; An analysis scheme is generated based on the optimized initial references, the optimized initial logic information, and all the confidence analysis results.

2. The legal information analysis method according to claim 1, characterized in that, The method for constructing the pre-defined legal knowledge graph includes: The constituent elements of the legal field's map are determined. These constituent elements include jurisprudential-level related elements, legal provision-level related elements, and case-level related elements. The jurisprudential-level related elements include basic concepts in the legal field, the legal provision-level related elements include legal provisions in the legal field, and the case-level related elements include related cases in the legal field and the relationships between legal subjects in the legal field. A jurisprudential knowledge architecture is constructed based on the basic concepts in the legal field. The jurisprudential knowledge architecture includes several basic concept knowledge nodes and the relationship edges between different basic concept knowledge nodes. A legal provision-level knowledge architecture is constructed based on the legal provisions of the legal field. The legal provision-level knowledge architecture includes several legal provision-type knowledge nodes and the relationship edges between different legal provision-type knowledge nodes. A case-level knowledge architecture is constructed based on the related cases in the legal field and the legal subject relationships in the legal field. The case-level knowledge architecture includes several case-type knowledge nodes and the relationship edges between different case-type knowledge nodes. A legal domain knowledge graph based on the aforementioned jurisprudential knowledge architecture, the aforementioned legal provision-level knowledge architecture, and the aforementioned case-level knowledge architecture comprises a multi-layered organizational knowledge architecture.

3. The legal information analysis method according to claim 1, characterized in that, The initial cited legal provisions are analyzed based on a standard legal provisions database to obtain legal provisions analysis results. When the legal provisions analysis results do not meet preset legal provisions requirements, a corresponding legal provisions update strategy is invoked to optimize the initial cited legal provisions, including: Semantic encoding is performed on the case description of the initial structured text to generate an initial cause-of-case vector; Events are extracted from the initial structured text to generate an initial element fingerprint; Based on the initial cause of action vector, the initial element fingerprint, and the time-geographic key composed of the filing date and jurisdiction, a two-level search is performed in the standard legal provisions database to obtain candidate legal provisions texts. The candidate legal provisions are semantically encoded to describe the case, generating candidate cause-of-case vectors; Event extraction is performed on the candidate legal texts to generate candidate element fingerprints; Obtain the semantic similarity between the initial cause-of-fact vector and the candidate cause-of-fact vector; Obtain the fingerprint similarity between the initial element fingerprint and the candidate element fingerprint; Obtain the validity of the candidate legal provision text within the time-region key; The accuracy of the legal provisions is calculated based on the semantic similarity, the fingerprint similarity, and the validity. The applicability rate of the legal provisions is calculated based on their validity status relative to the latest effective version. Pseudo-legal provisions are constructed, and log probabilities are calculated for the initial referenced legal provisions and the pseudo-legal provisions using a legal big model. The game confidence score is obtained based on the difference in log probabilities after Sigmoid mapping. When the accuracy of the legal provision, the applicability of the legal provision, and the confidence score of the game theory are all not lower than the corresponding thresholds, the candidate legal provision text is used as the optimized initial reference legal provision.

4. The legal information analysis method according to claim 3, characterized in that, The step of performing logical reasoning verification on the initial logical information to obtain the logical chain analysis result includes: The initial logical information is processed in a chain-like structure to generate a set of conditional propositions, a set of essential propositions, and a conclusion proposition. A directed causal graph is constructed based on the set of conditional propositions, the set of essential propositions, and the conclusion propositions, where nodes are propositions and edges are inference relationships between propositions; Perform missing element detection, skip step detection, and causal inversion detection on the directed causal graph to obtain the missing element degree of the proposition, the skip step degree of the skip step reasoning, and the inversion degree of the reasoning edge with the wrong direction; Calculate the logic chain integrity score based on the missing degree, the skip degree, and the inversion degree; When the integrity score of the logical chain is lower than a preset threshold, the logical chain analysis result is output. The logical chain analysis result includes a list of missing elements, a list of skip edges, and a list of reversed edges.

5. The legal information analysis method according to claim 4, characterized in that, The method for obtaining the missing value includes: The set of necessary component nodes is determined based on a preset logic template; The actual component nodes in the component proposition set are matched one by one with the component nodes in the necessary component node set using a semantic similarity threshold to obtain the number of actual component nodes that meet the matching conditions. The difference between the actual number of matching nodes and the total number of necessary nodes in the set of necessary nodes is taken as the number of missing nodes. The missingness degree is obtained by comparing the number of missing essential nodes with the total number of essential nodes in the set of essential nodes.

6. The legal information analysis method according to claim 4, characterized in that, The method for obtaining the step size includes: A directed cause-effect graph is constructed based on the initial logical information. The directed cause-effect graph contains a set of condition nodes, a set of requirement nodes, and a conclusion node. Perform a shortest directed path search from each condition node in the set of condition nodes to the conclusion node to obtain the shortest number of hops; When the shortest number of hops is greater than the preset legal number of hops threshold, the condition node-conclusion node pair is marked as a hop pair. The step degree is obtained based on the number of step pairs and the total number of condition nodes in the condition node set.

7. The legal information analysis method according to claim 4, characterized in that, The method for obtaining the inverse degree includes: A directed causal graph containing multiple inference edges is constructed using the initial logical information; Input the source node text and target node text corresponding to each inference edge into the causal direction discrimination model, and output the probability from source to target; When the probability is less than a preset direction threshold, the inference edge is marked as a reverse edge. The degree of inversion is obtained based on the number of inverted edges and the total number of inference edges.

8. The legal information analysis method according to claim 1, characterized in that, The step of comparing the initial analysis conclusions with a preset legal knowledge graph to identify text fragments with potential errors includes: The initial analysis conclusions are segmented and quantized to generate a set of conclusion sentence vectors; The set of conclusion sentence vectors is compared with the node vectors in the preset legal knowledge graph to identify the support or negation relationship between the entities in the conclusion and the graph nodes, and the entity conflict is marked. The initial analysis conclusions are compared with numerical intervals to identify whether the numerical values ​​in the conclusions fall within the official intervals of the corresponding conclusion entities in the preset legal knowledge graph, and numerical conflicts are marked. Based on the filing date and jurisdiction of the initial analysis conclusion, a time-domain-region comparison is performed on the preset legal knowledge graph to identify whether the conclusion entity is within the valid time period or region, and to mark the time conflict. Calculate the overall error score for each conclusion sentence based on the weighted scores of the entity conflict, the numerical conflict, and the timeliness conflict; When the overall error score is not lower than a preset threshold, the corresponding conclusion sentence is marked as a potential error text fragment, and the conflict type and corresponding node identifier are output for subsequent confidence analysis.

9. The legal information analysis method according to claim 1, characterized in that, The confidence analysis of the text fragments with potential errors, and the acquisition of the corresponding confidence analysis results, includes: Semantic drift detection is performed on the text fragments with potential errors to generate drift confidence scores; Numerical uncertainty quantification is performed on the text fragments with potential errors to generate numerical confidence scores; Perform graph consistency backtracking on the text fragments with potential errors to generate graph confidence scores; The drift confidence, the numerical confidence, and the graph confidence are weighted and fused to obtain a comprehensive confidence. When the overall confidence level is lower than a preset threshold, the text fragment with potential errors is marked as a high-risk fragment, and the corresponding confidence analysis result is output.

10. A legal information analysis system, characterized in that, include: The legal information analysis module is used to analyze the legal information to be analyzed based on the legal big model and generate initial structured text. The initial structured text includes initial legal citations, initial logical information, and initial analysis conclusions; The legal provision analysis and optimization module is used to analyze the initial referenced legal provisions based on the standard legal provision database, obtain the legal provision analysis results, and when the legal provision analysis results do not meet the preset legal provision requirements, call the corresponding legal provision update strategy to optimize the initial referenced legal provisions. The logical reasoning verification module is used to perform logical reasoning verification on the initial logical information and obtain the logical chain analysis results. When the logic chain analysis result does not meet the preset logic requirements, the corresponding logic update strategy is invoked to optimize the initial logic information. The conclusion comparison module is used to compare the initial analysis conclusions with a preset legal knowledge graph to identify text fragments with potential errors. The confidence analysis module is used to perform confidence analysis on the text fragments with potential errors and obtain the corresponding confidence analysis results. The analysis scheme generation module is used to generate an analysis scheme based on the optimized initial referenced legal provisions, the optimized initial logical information, and all the confidence analysis results.

11. A computer device, characterized in that, The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed, enables the at least one processor to perform the legal information analysis method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the legal information analysis method according to any one of claims 1-9.

13. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1-9.

Images