Local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph

Abstract

The application discloses a local legislation compliance intelligent detection system and method based on deep semantic analysis and a multi-modal legal knowledge graph, relates to the technical field of computer technology and legal cross technology, and comprises the following steps: constructing a legal knowledge graph; designing a multi-dimensional rule sub-library with a dynamic weight adjustment mechanism; constructing a legal provision deep semantic analysis model based on a Deepseek-based large language model, realizing intelligent generation of a national field legal regulation triple, and improving the field adaptability of triple generation; performing multi-dimensional conflict detection based on the knowledge graph, a conflict detection rule library and a semantic analysis result; and automatically generating a multi-dimensional legal conflict report. The application integrates multi-source legal data by constructing a legal knowledge graph and combining a multi-dimensional rule sub-library, performs deep semantic analysis on legal provisions by using a special model based on Deepseek, and judges explicit and implicit conflicts of contradiction points and contexts.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of computer technology and law, and particularly relates to a local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph. BACKGROUND

[0002] In the field of local legislation compliance intelligent detection, current technical means have many limitations. Existing legal text analysis methods mostly use simple keyword search or basic rule matching technology. This way can only superficially and shallowly analyze the text, and cannot deeply understand the complex logical relationship between legal provisions and the connotation of legal principles.

[0003] In the conflict detection of local regulations, especially single regulations of autonomous regions and superior laws, traditional methods are difficult to meet actual needs. On the one hand, manual review method is extremely inefficient, which needs to consume a lot of manpower and time cost. Reviewers need to compare a large number of regulations word by word and sentence by sentence, which not only has high work intensity, but also is easy to miss. At the same time, the subjectivity of manual review is strong, and different reviewers may draw different conclusions on the same regulation due to differences in knowledge reserve, understanding ability and judgment standard, resulting in inconsistency and inaccuracy of the review results.

[0004] On the other hand, existing automatic detection technology lacks systematic modeling of legislative authority, legal reservation principle and the boundary of the right to vary. For example, in judging the legislative authority of regulations of autonomous regions, it is difficult to accurately identify which clauses belong to the power of overstepping and which belong to the legal local detailed provisions. For the legal reservation principle, it is difficult to accurately distinguish between absolute reservation matters and relative reservation matters, and then to judge whether the local regulations conflict with the superior laws on these key matters. In the assessment of the right to vary of autonomous regions, there is no scientific and reasonable quantitative standard to define the scope of "not violating the basic principles of law", so that the compliance judgment of the varied clauses lacks reliable basis.

[0005] In addition, the existing technology performs poorly in processing large-scale regulation texts. With the increasing number of local legislation and the increasing complexity of regulation content, traditional detection methods are difficult to adapt to new needs in terms of detection speed and accuracy. Moreover, the existing technology lacks standardization, intelligence and pertinence in generating detection reports and providing modification suggestions, and cannot provide efficient and practical support for legislative organs.

[0006] In summary, the prior art has obvious deficiencies in intelligent detection of local legislation compliance, mainly manifested in lack of systematic modeling, low detection efficiency and accuracy, weak large-scale text processing capability, and lack of practicality of reports and recommendations. These deficiencies seriously restrict the development of local legislation compliance review work, and there is an urgent need for a more advanced, efficient and accurate intelligent detection system and method to fill this technical gap. SUMMARY

[0007] To solve the above technical problems, the present application provides an intelligent detection system and method for local legislation compliance based on deep semantic analysis and multi-modal legal knowledge graph, which realizes the automatic and accurate identification and early warning of conflicts between local laws and regulations, especially single regulations in autonomous regions and superior laws, improves the detection efficiency and accuracy, and provides scientific and practical review support for the legislative authority.

[0008] To achieve the above purpose, the present application provides an intelligent detection system for local legislation compliance based on deep semantic analysis and multi-modal legal knowledge graph, comprising:

[0009] A data acquisition module for receiving regulation text input by a local legislation unit and performing format preprocessing;

[0010] A legal knowledge graph construction module for extracting entities and relationships from multi-source legal texts and constructing a structured knowledge network;

[0011] A semantic analysis module for calling a large language model to perform deep semantic analysis on target regulation provisions and generate structured semantic blocks;

[0012] A conflict detection module for performing authority boundary detection and clause conflict analysis based on the legal knowledge graph and multi-dimensional rule sub-base;

[0013] A report generation module for generating a multi-dimensional legal conflict report based on the conflict detection results;

[0014] A visual display module for displaying conflict tracing paths and legal basis networks based on graphical components.

[0015] To achieve the above purpose, the present application also provides an intelligent detection method for local legislation compliance based on deep semantic analysis and multi-modal legal knowledge graph, comprising:

[0016] Constructing a legal knowledge graph to convert multi-source legal texts into a structured knowledge network containing entities and relationships;

[0017] Designing a multi-dimensional rule sub-base, which includes a hierarchical comparison rule sub-base, a semantic comparison rule sub-base and a special rule sub-base for ethnic regions;

[0018] The large language model is used for deep semantic analysis of the input target regulation provisions, and structured semantic blocks in the form of subject-action-condition triplets are extracted;

[0019] Based on the legal knowledge graph and the multi-dimensional rule sub-library, the authority boundary detection and clause conflict analysis are performed on the target regulation provisions.

[0020] According to the conflict detection result, a legal conflict report is generated, which contains conflict trace information and legal basis.

[0021] Optionally, the process of constructing the legal knowledge graph comprises:

[0022] The multi-source legal texts include constitutions, laws, administrative regulations, local regulations and autonomous regulations;

[0023] The legislative authority boundary, legal reservation items and threshold entities of the right to vary are extracted from the legal texts;

[0024] The entities are semantically mapped to unify synonymous and different-shaped terms;

[0025] A subject-action-condition triplet relationship model is constructed to describe the logical relationship between legal provisions;

[0026] The structured data and unstructured text are integrated using a four-tuple framework to convert into computable nodes;

[0027] The knowledge graph is stored in RDF or JSON-LD format, and the logical consistency is verified based on the description logic rules.

[0028] Optionally, the design of the multi-dimensional rule sub-library comprises:

[0029] The hierarchical comparison rule sub-library sets the legal effect level parameters and constructs the judgment matrix to dynamically adjust the weight factor;

[0030] The semantic comparison rule sub-library configures keywords, expression patterns and conflict type indicators;

[0031] The special rule sub-library of ethnic regions stores the entity basis, procedure conditions and record standards for the exercise of the right to vary;

[0032] Based on new regulations or judicial documents, the rules are continuously learned and automatically supplemented and the weights are adjusted.

[0033] Optionally, the process of deep semantic analysis comprises:

[0034] The target regulation provisions are input into the large language model, and the text is analyzed based on the preset prompt word template;

[0035] The subject-action-condition triplets are output;

[0036] Identify the conditional relationship, causal relationship and hierarchical relationship between clauses;

[0037] Align the analysis result with the entity of the legal knowledge graph to form a structured semantic block.

[0038] Optionally, the process of executing the authority out-of-bound detection and clause conflict analysis includes:

[0039] Extract the authority information of the target regulation clause, and compare the authority boundary in the legal knowledge graph;

[0040] Extract the subject-action-condition triple element in the target regulation clause;

[0041] Search the superior law association node in the legal knowledge graph;

[0042] Calculate the semantic deviation degree, and trigger the early warning when the deviation degree exceeds the preset threshold.

[0043] Optionally, the process of generating the legal conflict report includes:

[0044] Construct a clause evolution path graph based on time series;

[0045] Generate a multi-level legal basis citation network through the legal knowledge graph;

[0046] Quantify the conflict risk level from the authority, content and procedure dimensions;

[0047] Generate a structured report containing a conflict summary, basis and revision suggestions.

[0048] Optionally, the process of quantifying the conflict risk level includes:

[0049] Determine the authority dimension risk level according to the authority out-of-bound detection result;

[0050] Determine the content dimension risk level according to the clause conflict analysis result;

[0051] Determine the procedure dimension risk level according to the compliance of the alternative power record standard.

[0052] Technical effects of the present application:

[0053] (1) Legal hierarchy relationship management efficiency improvement: The scheme builds a legal knowledge graph, based on entity recognition and standardized annotation technology of legal texts, automatically extracts core elements such as regulation name, clause number, legislative organ, and effective time, and constructs a legal effectiveness transmission network. This design can systematically present the topological structure of the hierarchical relationship in the legal system, support legislative review personnel to quickly locate the associated superior law of the target clause, and significantly improve the efficiency of combing the vertical relationship of the legal system. Through the visual interactive interface of the knowledge graph, it assists the review personnel to intuitively trace the evolution of the regulation revision, and reduces the time cost of cross-regulation comparison.

[0054] (2) Multi-dimensional legal conflict detection: Based on large language model to realize the automatic identification and attribution analysis of legal conflict. The system uses three mechanisms of hierarchical conflict detection, semantic deviation calculation, and procedure compliance verification to automatically identify multiple conflict types. And through the associated reasoning ability of the knowledge graph, it realizes the multi-level basis matching of conflict clauses and related constitutional provisions, laws, and administrative regulations, and provides verifiable conflict judgment evidence chain for the review personnel.

[0055] (3) Conflict detection accuracy improvement: By building a structured legal comparison rule library, dynamically configuring key semantic recognition elements, including core attention keywords, typical expression templates, conflict type indicators, and their corresponding examples. Accurately guide the large model to perform deep semantic matching and judgment in the process of clause comparison, effectively reduce the false positive and false negative rates, and ensure the high accuracy and explainability of conflict type identification.

[0056] (4) Comprehensive and scientific modification suggestions: After detecting legal conflicts, the system calls large-scale language models, combines context semantics and legal principles, and automatically generates clause-level correction schemes, alternative path recommendations, and risk control strategies. The generated suggestions not only accurately target conflict types, but also provide example expression templates and procedural operation processes, ensuring that the modification opinions not only conform to the spirit of superior law, but also take into account the needs of national regional autonomy, achieving the organic unity of scientificity and practicality.

[0057] (5) Standardized government report automatic generation throughout the process: Based on the configurable template-driven engine and report framework, the conflict detection results, visual charts, and modification suggestions are automatically filled into the pre-defined government document format, supporting multiple output formats such as PDF and Word; through parameterized control and modular components, significantly reducing manual typesetting and proofreading work, ensuring report structure consistency and format specification, improving generation efficiency and compliance. BRIEF DESCRIPTION OF DRAWINGS

[0058] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The illustrations, together with their description, serve to explain the application without imposing undue limitation on the application. In the drawings:

[0059] Figure 1 This is a flowchart illustrating the intelligent detection method for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph, according to an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram illustrating the composition of the legal and regulatory conflict detection rule base according to an embodiment of the present invention;

[0061] Figure 3 This is a schematic diagram illustrating the construction of the legal and regulatory conflict detection rule base according to an embodiment of the present invention;

[0062] Figure 4 This is a schematic diagram of the overall collision detection structure according to an embodiment of the present invention;

[0063] Figure 5 This is a detailed implementation diagram of the legal and regulatory conflict detection rule base according to an embodiment of the present invention;

[0064] Figure 6 This is a schematic diagram of the conflict detection module of the legal and regulatory conflict detection rule base in an embodiment of the present invention;

[0065] Figure 7 This is a schematic diagram of the structure of the intelligent detection system for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph, according to an embodiment of the present invention. Detailed Implementation

[0066] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0067] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0068] like Figure 7 As shown, this embodiment provides a local legislative compliance intelligent detection system based on deep semantic analysis and multimodal legal knowledge graph, including:

[0069] The data acquisition module is used to receive the legal texts input by local legislative bodies and perform formatting preprocessing.

[0070] The legal knowledge graph construction module is used to extract entities and relationships from multi-source legal texts and build a structured knowledge network.

[0071] The semantic analysis module is configured to invoke a large language model to perform deep semantic analysis on the target legal provisions, and generate structured semantic blocks;

[0072] The conflict detection module is configured to perform authority boundary detection and clause conflict analysis based on the legal knowledge graph and the multi-dimensional rule sub-library;

[0073] The report generation module is configured to generate a multi-dimensional legal conflict report according to the conflict detection results;

[0074] The visualization display module is configured to display the conflict traceability path and the legal basis network based on graphical components.

[0075] As shown in Figures 1-6 the embodiment also provides an intelligent detection method for local legislation compliance based on deep semantic analysis and multi-modal legal knowledge graph, which includes:

[0076] constructing a legal knowledge graph containing legislative authority boundaries and discretionary power threshold attributes;

[0077] designing a multi-dimensional rule sub-library with a dynamic weight adjustment mechanism;

[0078] constructing a deep semantic analysis model for legal provisions based on a large language model based on Deepseek, realizing intelligent generation of legal regulation triples in the ethnic field, and improving the field adaptability of triple generation;

[0079] performing multi-dimensional conflict detection based on the knowledge graph, the conflict detection rule library, and the semantic analysis results;

[0080] automatically generating a multi-dimensional legal conflict report, covering conflict analysis, targeted correction suggestions, and compliance evaluation.

[0081] Further, the legal knowledge graph containing the legislative authority boundaries and the discretionary power threshold attributes is constructed, which includes:

[0082] Through deep analysis of multi-source legal texts such as the Constitution, laws, administrative regulations, local regulations, and autonomous regulations, special legal entities with associated logic such as legislative authority boundaries, legal reservation items, and discretionary power thresholds are extracted, and the core nodes in the graph are constructed accordingly;

[0083] During the entity construction process, the system performs semantic recognition and standardization processing on text elements such as regulation name, clause number, legislative organ, effective time, and revision record, to ensure the uniqueness and consistency of the entities;

[0084] In terms of relationship modeling, the graph supports the construction of the effectiveness transmission relationship between superior and inferior laws, as well as the establishment of potential conflict relationships between parallel legal norms and horizontal correlation relationships between discretionary power application clauses, thereby forming a multi-dimensional legal norm network structure;

[0085] For the provisions related to the content of regional ethnic autonomy, the system further conducts attribute labeling to clearly indicate its legislative authority boundary, whether it belongs to legal reservation items, and the threshold of applicable flexibility, so as to provide attribute support for subsequent conflict detection;

[0086] In addition, the system also integrates structured legal databases and unstructured legal texts (such as judicial documents, legislative interpretations, etc.), and through a unified semantic embedding model, it constructs a legal knowledge representation with context weight adjustment capability, realizing the unified expression and intelligent calling of heterogeneous legal data.

[0087] Further, a multi-dimensional rule sub-library with dynamic weight adjustment mechanism is designed, including:

[0088] The system designs and constructs a multi-dimensional rule sub-library system with dynamic weight adjustment mechanism to realize accurate compliance comparison of local legislative texts in different dimensions. The rule system includes three sub-systems: hierarchical comparison rule sub-library, semantic comparison rule sub-library, and special rule sub-library for ethnic regions, and supports dynamic updating based on continuous learning mechanism;

[0089] The hierarchical comparison rule sub-library is used to clarify the effectiveness level between laws, administrative regulations and local regulations. By setting the hierarchical parameters between normative documents, a judgment matrix of "lower laws cannot conflict with higher laws" is established, providing a basis for the institutional constraint relationship between laws and regulations.

[0090] The semantic comparison rule sub-library configures a set of key semantic recognition elements, including key legal keywords, typical expression patterns, conflict type indicators and their semantic examples, to assist large language models in identifying possible expression deviations, logical conflicts and specification overlaps at the semantic level of provisions.

[0091] The special rule sub-library for ethnic regions includes rules related to the exercise of flexibility in the special flexibility space of the legislation practice in ethnic autonomous regions, including statutory entities, procedural conditions, and record review standards, to support the system's multi-dimensional evaluation of the legality, procedural completeness, and application boundaries of flexible provisions;

[0092] Each of the above rule sub-libraries supports dynamic updating mechanism. The system can learn rules based on the accumulation of legislative interpretation texts, typical judicial cases and record review opinions, realize automatic supplement of content and intelligent adjustment of rule weight factors, and thus improve the adaptability and forward-looking of the rule system to legal evolution and policy changes.

[0093] Further, a deep semantic analysis model of legal provisions is constructed using a large language model based on Deepseek, realizing the intellectualization of the generation of triples of laws and regulations in the ethnic field and improving the field adaptability of triple generation, including:

[0094] The system uses a large language model based on Deepseek to perform deep semantic analysis of legal provisions, generates triples that reflect the internal structure of legal norms, and constructs a logical relationship network between legal provisions accordingly.

[0095] This semantic analysis process takes legal texts as input, automatically extracts the three core elements of legal subjects, behaviors, and applicable conditions in the provisions through the semantic modeling capabilities of large models, forms triples of "subject-behavior-condition", and is used to express the basic semantic structure of legal provisions;

[0096] To improve the normativity and accuracy of semantic analysis, the system first standardizes legal terms, unifies terms with different forms and expressions to standard legal concepts defined in the graph, ensuring the consistency and comparability of semantics between different provisions;

[0097] On this basis, the system further identifies the logical structural relationships between provisions, including the condition relationship, causal relationship, and hierarchical relationship between clauses or adjacent clauses, and structures them in the form of graph edges to build a logical support network suitable for conflict detection and comparison analysis;

[0098] In addition, the system embeds abstract legal concepts such as constitutional principles and principles of regional ethnic autonomy into the semantic space, converts these abstract rules into computable vector representations through training or introducing predefined legal principle vector models, and participates in the subsequent provision semantic similarity calculation and rule matching process, thereby realizing the intelligent expression and comparison of the deep structure of legal norms.

[0099] Further, based on the knowledge graph, conflict detection rule base, and semantic analysis results, multi-dimensional conflict detection is performed, including:

[0100] First, the system conducts authority boundary detection, analyzes the information of the institution that formulates the local regulations, combines the legal authority boundary data recorded in the knowledge graph, and judges whether the institution has the corresponding legislative authority, identifying the normative provisions that may exceed the authorized authority or violate the principle of legislative division of labor;

[0101] Second, the system performs clause conflict analysis, extracts the core legal elements in the target clause, including triples of legal subjects, behaviors, and conditions, searches for relevant upper-level law clauses in the knowledge graph, and calculates the element matching degree and semantic deviation degree based on the set rule model;

[0102] When the semantic deviation of the target clause from the superior law is detected and exceeds the preset threshold, the system automatically triggers the conflict warning mechanism, marking the specific conflict points and reference laws, thereby assisting the legislative authority in accurately identifying possible systemic conflicts and compliance risks.

[0103] Further, an automated multi-dimensional legal conflict report is generated, including:

[0104] First, in terms of conflict traceability visualization, the system constructs an evolution path diagram of the conflicting clause based on the time series information and version evolution records of the clause, visually presenting the changing trajectory of the target clause from the initial formulation to the current version in terms of compliance, clearly identifying the node where the conflict first appeared and its subsequent modification, expansion, or deviation trend, which helps trace the source of risk evolution;

[0105] Second, in terms of legal basis query matching, the system constructs a multi-level reference and attachment relationship network between the target clause and the constitution, laws, administrative regulations, and judicial interpretations it depends on by combining graph query and semantic embedding matching technology, showing the legal basis and reference chain of the clause in the entire legal system;

[0106] In terms of risk level assessment, the system conducts quantitative analysis of each clause with compliance risks based on multiple detection dimensions such as authority boundary crossing, clause conflict, and legislative procedure, outputs a multi-dimensional risk level matrix, and clearly marks different risk levels and their corresponding urgent correction priorities;

[0107] The system structures and integrates the above analysis results, generates a standardized local legal regulations compliance conflict comparison report according to government document format requirements, and the report content includes key elements such as problem description, reference basis, evaluation conclusion, and revision suggestions, and supports output in formats such as PDF and Word, meeting the actual needs of archiving, review and record, and cross-system integration.

[0108] Further, automatically generate targeted clause modification suggestions, including:

[0109] Generate clause-level revision suggestions: based on the specific conflict types marked in the detection report, and combined with the language model's learning ability for similar compliance texts, automatically generate suggested revised clauses and point out the reference expression methods of the superior law;

[0110] Recommendation of alternative paths: for clauses with legislative variation space for national regional autonomy, according to the entity conditions and procedure paths stored in the variation right rule library, provide suggestions for setting variation basis and improving procedure flow according to law;

[0111] Output risk control strategy: according to the risk assessment matrix results, the modification priority of high-risk clauses is proposed, and the necessary reporting and review suggestions are proposed to assist the legislative authority to carry out systematic revision;

[0112] Generate compliance revision reference template: automatically summarize typical revision schemes for similar conflict problems in similar laws and regulations to form a structured template, which is convenient for local legislative units to refer to;

[0113] Generate final detection report: embed the above revision suggestions in the standardized detection report in the "problem-reason-suggestion" three-part structure to ensure that the report is understandable and operable, facilitating legislative review and subsequent adjustment.

[0114] The embodiment of the application provides a local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph based on deep semantic analysis and multi-modal legal knowledge graph, aiming to solve the problems of heavy artificial work burden, low conflict recognition efficiency, and lack of systematicness of modification suggestions in the existing local legislation review process. The method combines the natural language understanding ability of the large model and the structured expression advantage of the domain knowledge graph, constructs a multi-dimensional rule comparison mechanism, realizes intelligent identification of the compliance of local regulations and superior laws in the aspects of keyword meaning, system essence, expression method, etc. Through knowledge graph construction and heterogeneous specification mapping, a structured legal relationship network is formed; combined with the deep semantic understanding of the large language model, the key elements in the regulations are extracted, and conflict detection is performed through matching the rule library; finally, the system automatically generates a standardized report containing conflict provisions, identification reasons and modification suggestions. The method can be widely applied to national autonomous local legislation review, local government compliance self-checking, and various legal supervision scenes such as state organ review and record-keeping, and provides an innovative solution path for realizing internal coordination of the regulation system, standardization and intelligentization of compliance review.

[0115] Next, the key technologies required to implement the local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph provided by the embodiment will be introduced:

[0116] One of the core technologies for implementing the local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph provided by the embodiment is to construct a legal knowledge graph, aiming to unify and integrate legal texts, structured regulation information and historical revision records, and construct a knowledge network with deep semantic and visual expression, providing a solid foundation for subsequent rule library construction and intelligent comparison reasoning.

[0117] The legal knowledge graph refers to a knowledge network with semantic properties and visual characteristics, which abstracts legal provisions, regulation structure, legislative authority, revision history and other information as nodes and edges based on a graph model, as shown in Figure 1 .

[0118] Firstly, the central law, local regulations, autonomous regulations and administrative rules and regulations are constructed into a full-domain data fusion channel. Through the heterogeneous data collaborative processing mechanism, the three-dimensional analysis of structured data (XML / JSON), semi-structured documents (Word / PDF) and unstructured explanatory text is realized. This mechanism innovatively integrates chapter number topology analysis, clause semantic unit segmentation and cross-regulation reference network modeling, combined with spatiotemporal double-chain metadata reconstruction technology, to construct a domain constraint network of legal effectiveness hierarchy horizontally, and embed a time evolution dimension of historical revision track vertically, forming a "domain-hierarchy-time-space" four-tuple legal effectiveness representation framework, which fundamentally solves the semantic heterogeneity and spatiotemporal conflict problems of multi-source regulation data.

[0119] In the semantic network construction phase, the system breaks through the limitations of traditional rule matching, establishes a complex legal entity recognition system, covering "legislative authority boundary", "special provisions of autonomous regulations", "requirements for the exercise of discretion", etc. Fine-grained entity types, and based on the logic of legal behavior, a seven-tuple relationship model (subject, object, behavior, condition, authority, consequence, time limit) is constructed, and the dynamic role path of "authorization-limitation", "priority application-revision replacement" and other relationships is explicitly defined. For the complex associations such as "cross-effect conflict" and "implicit authority coverage" implied in legal texts, a graph structure reasoning mechanism is introduced, and the legal logic is deeply deconstructed through the self-consistency verification of the semantic network.

[0120] Further, the system innovatively designs a cross-modal knowledge fusion channel to convert visual information such as implementation flowcharts and regulation hierarchy diagrams into computable nodes with legal semantics. Through visual semantic analysis technology, a mapping rule library of graphical symbols and text clauses is established, and based on the cross-modal attention mechanism, the correlation strength between flowchart nodes, revision history timelines and text semantics is quantified, finally realizing the deep coupling of information in a unified vector space. This process not only preserves the visual features of legal knowledge, but also generates associated edge attributes with explanatory and traceable properties through semantic projection technology.

[0121] The storage and conflict detection of the knowledge graph use a "structure storage-logical reasoning" dual-engine architecture to solidify legal knowledge in the form of "node-edge-attribute" triplets, and based on description logic rules, explicitly mark the effectiveness transmission path of superior and inferior laws, the conflict mode of parallel regulations and the application boundary of discretion clauses.

[0122] Another core work of implementing the local legislation compliance intelligent detection system and method based on deep semantic analysis and multi-modal legal knowledge graph provided by the embodiment is to design a legal regulation conflict comparison rule base, which aims to help the large model more accurately perform explicit and implicit conflict judgment on the conflict points and context, and improve the understanding ability of the model on the legal regulation conflict. The construction process thereof relies on the collaborative work of human and AI, and ensures the comprehensiveness and accuracy of the rules.

[0123] The design of the rule base follows the principle of "hierarchical classification and collaborative iteration", and is composed of three types of sub-bases, i.e., hierarchical comparison, semantic comparison and special rules for ethnic regions, as well as a dynamic updating and quality control mechanism. In the hierarchical comparison sub-base, the present application first marks the legal documents at each level, such as the Constitution, laws, administrative regulations, local regulations and autonomous regulations, with the effectiveness, and guides the model to determine whether the lower law conflicts with the upper law by presetting the effectiveness level parameters and the determination matrix; at the same time, a weight factor is introduced, so that the rules can be automatically adjusted according to newly enacted legislation interpretation or judicial adjudication, so as to maintain the accuracy and robustness of the determination matrix in different legal environments.

[0124] In the construction process of the semantic comparison sub-base, the present application combines the manual annotation of legal experts with the deep learning ability of AI models to systematically extract the core semantic elements, typical expression modes and conflict type indicators in legal texts, and form semantic matching templates, as shown in Figure 2 The legal experts generate initial semantic rules by fine annotation on high-frequency keywords such as "rights", "obligations", "prohibitions", "should" and "shall not" and their context cases; the AI model performs batch analysis on massive legal texts on the basis of the initial semantic rules, mines potential synonymous expressions and implicit conflict indication modes, and feeds the candidate rules back to the experts for review. Such iterative optimization enables the semantic comparison sub-base to maintain the rigor of the law while continuously expanding the coverage of emerging legal language phenomena, thereby significantly improving the recognition and classification accuracy of the model on complex semantic conflicts

[0125] In view of the special needs of the autonomous regulations in ethnic regions, the present application specially sets up a special rule sub-base for ethnic regions, which is used to manage the entity basis, procedural conditions and record review standards of the exercise of the right of flexibility. The system clearly defines the scope of application and the boundary of the right by attribute annotation on the clauses related to the exercise of the right of flexibility in the autonomous regulations, and further refines the exercise conditions and process nodes in combination with typical cases. In the subsequent legislation dynamics, the system continuously supplements and fine-tunes the sub-base based on the latest cases and expert feedback, so as to ensure that the special rules meet the legal authority and take into account the flexible space of the autonomous regions in actual application.

[0126] In order to ensure the long-term stable operation and precision improvement of the rule base, the present application constructs a dynamic updating and quality control mechanism, as shown in Figure 3The system can obtain new legislative documents, judicial interpretations and judicial documents from authoritative channels such as the Standing Committee of the National People's Congress and the Supreme People's Court at any time through manual transmission, identify potential new rules through text extraction technology, and adjust the weight distribution within the sub-library according to the evaluation results. At the same time, legal experts regularly review and correct the rules updated by AI to ensure that the rules are consistent with the legal spirit and are operable.

[0127] Finally, when the legal regulation clauses to be tested are input into the system, the three types of rule sub-libraries of hierarchy, semantics and special ethnic regions are effective in artificial and AI collaboration. The large model can complete conflict identification and classification within seconds by traversing the decision matrix, matching semantic templates and checking special rules, and output structured conflict analysis results and targeted revision suggestions. The comparison rule library not only provides accurate decision-making basis for multi-dimensional conflict detection, but also lays a solid foundation for subsequent report generation, visualization traceability and risk assessment.

[0128] Then, the system first calls the Deepseek-based large language model for deep semantic analysis, and combines the knowledge graph, conflict detection rule library and analysis results to perform conflict detection in multiple dimensions, and finally generates an intelligent legal conflict report, covering conflict analysis, targeted revision suggestions and compliance assessment, such as Figure 5 as shown.

[0129] Specifically, as shown in Figure 6 , the system first processes the input legal clauses in a unified format, disassembles information such as section numbers, paragraphs, legal subjects, action verbs, conditional modifiers, etc. into structured data units, and constructs a corresponding temporary analysis buffer. Subsequently, through the pre-designed Prompt template in the legal field, the instruction is "Please deeply analyze the core legal elements of the following clauses, including subject, obligation, prohibition, condition and scope of application", and the legal regulation comparison rule library is input to guide the judgment, and then the structured clauses are input into the large-scale pre-training language model based on Deepseek. The model returns not only the natural language analysis of each legal specification, but also outputs the extracted triples, obligation intensity labels and the implicit logical relationship between clauses in JSON or XML format.

[0130] After completing semantic analysis, the system aligns the structured elements output by the model with the entity nodes and attributes in the legal knowledge graph through a mapping engine: first, according to the entity standardization rule base, the elements of “subject”, “action”, “condition” and the like are mapped to the corresponding nodes; then, the hierarchical and conflict links between clauses are automatically completed using the “efficacy transmission”, “parallel conflict” and “flexible application” relationship edges in the graph. Based on this fusion graph, the system dispatches a multi-dimensional conflict detection engine, which first detects the semantic deviation of the newly analyzed nodes from the graph nodes along the predefined dimensions of “authority boundary”, “scope of application”, “behavior obligation”, “procedure condition” and the like; then, in combination with the “prohibition-allowance”, “mandatory-exception” and “inconsistent conditions” and other typical patterns in the rule base, pattern matching and conflict level calculation are performed, and the conflict type (substantive conflict, expression conflict), urgency and impact range are output.

[0131] After the detection is completed, the system automatically summarizes and structures the conflict results. In order to further realize the legal conflict detection and reporting automation function proposed in the present application, the system as a whole adopts a modular architecture design, which is divided into six core sub-modules according to function: data acquisition module, legal knowledge graph construction module, semantic analysis module, conflict detection module, report generation module and visualization display module. The sub-modules realize the collaborative transmission of structured data streams through a unified data exchange interface, supporting the whole process of intelligent detection of regulatory compliance.

[0132] Among them, the data acquisition module is responsible for receiving the input of regulatory texts from local legislative units (supporting formats such as Word, PDF, XML), and calling OCR and formatting preprocessing engines for paragraph segmentation, coding standardization and structured tag generation;

[0133] The legal knowledge graph construction module extracts core semantic units such as legislative authority, legal reservation items and flexible right conditions in regulations based on pre-trained entity recognition models and rule templates, and constructs a graph of entities and relationships. The graph data structure uses a hybrid representation based on RDF and JSON-LD, and is stored in a graph database (Neo4j or GraphDB) to support semantic queries and graph evolution updates;

[0134] Detailed modules such as Figure 4 The semantic analysis module takes legal texts as input, calls a large language model based on Deepseek fine-tuning, performs deep semantic understanding tasks, and outputs structured semantic blocks including legal triples (subject-action-condition), logical relationship annotations, obligation intensity and the like, with the data format unified as a JSON object stream format;

[0135] The conflict detection module jointly calls the graph matching engine and the rule comparison rule sub-library to analyze the legality of the target clause and compare the content consistency. The matching process is based on graph traversal, semantic deviation calculation and rule matrix evaluation mechanism. The system can dynamically call the weight adjustment strategy to update the conflict judgment logic;

[0136] The report generation module is responsible for automatically generating a standardized conflict report based on the detection results, including conflict clause number, associated superior law basis, semantic deviation analysis, risk assessment level, revision suggestions and other information. This module supports customizing document templates and can export PDF / Word formats according to government document specifications, suitable for archiving and reviewing;

[0137] The visualization display module is based on Web graphical components (D3.js or ECharts) to build a graph visualization interface and conflict path graph interaction panel, supporting time dimension evolution view and law clause dependency graph display, realizing conflict "visibility, traceability and explainability".

[0138] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A local legislative compliance intelligent detection system based on deep semantic analysis and multimodal legal knowledge graph, characterized in that: include: The data acquisition module is used to receive the legal texts input by local legislative bodies and perform formatting preprocessing. The legal knowledge graph construction module is used to extract entities and relationships from multi-source legal texts and build a structured knowledge network. The semantic parsing module is used to call a large language model to perform deep semantic parsing on the target legal provisions and generate structured semantic blocks; The conflict detection module is used for execution permission overstepping detection and clause conflict analysis based on legal knowledge graph and multi-dimensional rule sub-library. The report generation module is used to generate multi-dimensional legal conflict reports based on conflict detection results. The visualization module is used to display the conflict tracing path and legal basis network based on graphical components; The design of the multidimensional rule sub-library includes: A hierarchical comparison rule sub-library is established, which sets legal validity level parameters, constructs a judgment matrix, and dynamically adjusts weighting factors. A semantic comparison rule sub-library, configured with keywords, expression patterns, and conflict type indicator words; The sub-database of special rules for ethnic regions stores the substantive basis, procedural conditions, and filing standards for exercising the right of modification; Based on new regulations or court judgments, it continuously learns and automatically supplements rules and adjusts weights; The process of execution permission out-of-bounds detection and clause conflict analysis includes: Extract information about the enacting authority of the target legal provisions and compare it with the authority boundaries in the legal knowledge graph; Extract the subject-behavior-condition triplet elements from the target legal provisions; Retrieve nodes related to higher-level laws in the legal knowledge graph; Calculate semantic deviation and trigger an alert when the deviation exceeds a preset threshold.

2. The method of the intelligent detection system for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph as described in claim 1, characterized in that, include: Construct a legal knowledge graph to transform multi-source legal texts into a structured knowledge network containing entities and relationships; Design a multi-dimensional rule sub-library, which includes a hierarchical comparison rule sub-library, a semantic comparison rule sub-library, and a special rule sub-library for ethnic regions; The large language model is used to perform deep semantic parsing on the input target legal provisions to extract structured semantic blocks in the form of subject-behavior-condition triples; Based on the legal knowledge graph and the multidimensional rule sub-base, the system performs over-boundary detection of execution authority for target legal provisions and analysis of clause conflicts. A legal conflict report is generated based on the conflict detection results. The legal conflict report includes conflict tracing information and legal basis.

3. The intelligent detection method for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph as described in claim 2, characterized in that, The process of constructing the legal knowledge graph includes: The multi-source legal texts include the Constitution, laws, administrative regulations, local regulations, and ethnic autonomous regulations; Extracting entities from legal texts regarding the boundaries of legislative authority, matters reserved by law, and thresholds for the right of modification; Semantic mapping of entities to unify synonymous and heterogeneous terms; Construct a subject-behavior-condition triple relationship model to describe the logical relationships between legal provisions; The quadruple framework is used to integrate structured data and unstructured text, transforming them into computable nodes; Store the knowledge graph in RDF or JSON-LD format and verify logical consistency based on the rules describing the logic.

4. The intelligent detection method for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph as described in claim 2, characterized in that, The deep semantic parsing process includes: Input the target legal provisions into the large language model, and parse the text based on the preset prompt word template; Output the subject-behavior-condition triple; Identify the conditional, causal, and hierarchical relationships between clauses; The parsing results are aligned with the entities in the legal knowledge graph to form structured semantic blocks.

5. The intelligent detection method for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph as described in claim 2, characterized in that, The process of generating a legal conflict report includes: Construct a clause evolution path diagram based on time series; Generate a multi-level legal basis citation network using a legal knowledge graph; Quantify the level of conflict risk from the perspectives of permissions, content, and procedures; Generate a structured report that includes a conflict summary, supporting evidence, and revision recommendations.

6. The intelligent detection method for local legislative compliance based on deep semantic analysis and multimodal legal knowledge graph as described in claim 5, characterized in that, The process of quantifying the conflict risk level includes: Determine the risk level of the permission dimension based on the permission over-boundary detection results; Determine the risk level of each content dimension based on the results of the clause conflict analysis; The risk level of the procedure is determined based on compliance with the standard for filing the right of modification.

Images