Report generation system and method based on legal knowledge graph

Abstract

The invention discloses a report generation system and method based on a legal knowledge graph, and relates to the technical field of artificial intelligence and law, and the method comprises the steps: obtaining a legal appeal of a user through a non-standardized domain corpus collection engine, carrying out the multi-dimensional intention deconstruction through a semantic slicing model, and obtaining a judicial element topology network; inputting the judicial element topology network into a law deduction engine of dynamic value perception ability, calculating a decision vector weight through a dynamic value evaluation network, and generating a judicial decision path by using a time sequence feedback mechanism; based on a judicial decision path, through an association rule mining engine of a multi-modal knowledge graph, a dynamic weighting law retrieval confidence matrix is generated in combination with semantic coupling degree analysis and logic path influence evaluation. According to the method, qualitative change of legal consultation from information retrieval to intelligent decision is realized through collaborative innovation of a legal deduction engine with dynamic value perception capability and a legal semantic constraint decoder.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and legal technology, and in particular to a report generation system and method based on a legal knowledge graph. Background Art

[0002] Although some progress has been made in the analysis of structured legal data through the combination of rule engines and general NLP models, there are still obvious limitations. It usually relies on static legal knowledge bases, making it difficult to track the revision dynamics of legal provisions in real time, resulting in the generated consulting advice possibly citing repealed or invalid laws.

[0003] In the field of artificial intelligence and legal technology, there is a lack of effective control over the logical coherence of the generated legal consulting reports. They simply list legal provisions and conclusions, fail to organize the output content according to the laws of judicial cognition, and present suggestions in a discrete manner. This not only reduces the operability of the report, but also increases the information integration costs of legal practitioners. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a report generation method based on legal knowledge graph to solve the problems of incomplete compliance detection and compliance risks in legal consulting reports.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a report generation method based on a legal knowledge graph, which includes obtaining a user's legal demands through a non-standardized domain corpus collection engine, performing multi-dimensional intent deconstruction through a semantic slicing model, and obtaining a judicial element topology network;

[0008] Input the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, calculate the decision vector weights through the dynamic value evaluation network, and use the temporal feedback mechanism to generate the judicial decision path;

[0009] Based on the judicial decision-making path, a dynamic weighted legal search confidence matrix is generated through the association rule mining engine of the multimodal knowledge graph, combined with semantic coupling analysis and logical path influence assessment;

[0010] Use the compliance situation awareness engine to verify rule path consistency and generate a compliance report with correction suggestions by quantifying risk entropy values;

[0011] The legal semantic constraint decoder is used to perform multi-dimensional verification and structural reorganization of the compliance report with revised suggestions to generate a legal consulting report.

[0012] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, wherein: obtaining the user's legal demands through a non-standardized domain corpus collection engine, performing multi-dimensional intention deconstruction through a semantic slicing model, and obtaining a judicial element topology network includes the following steps:

[0013] Input the user's legal demand data into the semantic slicing model, and parse the user's legal demand into a judicial element parsing vector through a multi-granular intent recognition and parsing method;

[0014] A weight-compensated word segmentation algorithm is used to dynamically segment and tag judicial element parsing vectors. Domain-enhanced semantic role labeling is used to map the labeling results into triples.

[0015] The judicial knowledge prior rules are used to eliminate ambiguous triples, and the triples are logically checked to obtain an iterative domain knowledge topology map.

[0016] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, wherein: inputting the judicial element topology network into the legal deduction engine with dynamic value perception capability, and calculating the decision vector weight through the dynamic value evaluation network includes the following steps:

[0017] Use the graph attention network method to extract legal node embedding data from the knowledge topology graph and obtain the initial legal decision through the multi-layer perceptron;

[0018] Use the temporal difference strategy distillation collaborative algorithm to optimize initial legal decisions and build a legal deduction engine with dynamic value perception capabilities;

[0019] The legal node embedding data is input into the legal deduction engine with dynamic value perception capability, and the decision vector weight is calculated using a heterogeneous attention gating mechanism.

[0020] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, generating a judicial decision path using a temporal feedback mechanism includes the following steps:

[0021] Use the state memory-contradiction perception collaborative network to record the state changes of legal elements of decision vector weights;

[0022] Through temporal causal convolution, we can mine the logical contradictions in the state changes of potential legal elements and generate correction signals.

[0023] The decision vector weights and correction signals are combined using policy entropy weighted fusion and probability deduction tree search, and the Monte Carlo tree search method is used to generate the judicial decision path.

[0024] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, wherein: based on the judicial decision path, through the association rule mining engine of the multimodal knowledge graph, and using the semantic coupling degree for analysis, generating a dynamic weighted legal retrieval confidence matrix includes the following steps:

[0025] The judicial decision path is converted into a time-aware legal feature vector through a spatiotemporal modal collaborative reasoning framework. The legal feature vector is calculated using the Apriori algorithm and a cross-modal graph neural network to obtain an association rule set.

[0026] The association rule set is analyzed through the legal semantic coupling analysis model, and the cross-modal semantic similarity of the element pairs in the association rule set is calculated;

[0027] The cross-modal semantic similarity is integrated through the gated attention fusion mechanism to generate a dynamic weighted legal retrieval confidence matrix.

[0028] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, wherein: using the compliance situation awareness engine to verify the consistency of the rule path, and generating a compliance report with correction suggestions by quantifying the risk entropy value, the following steps are included:

[0029] A bidirectional graph matching algorithm is used to compare the dynamic weighted legal retrieval confidence matrix with the judicial decision path, and to calculate the clause coverage and conflict coefficient indicators.

[0030] Construct a multi-scale risk assessment model based on clause coverage and conflict coefficient indicators and calculate risk entropy;

[0031] The risk entropy value is quantified through a multi-scale risk entropy fusion algorithm, and combined with gradient backpropagation and Monte Carlo simulation to generate a compliance report with correction suggestions.

[0032] As a preferred solution of the report generation method based on the legal knowledge graph described in the present invention, wherein: the compliance report of the revised suggestion is multi-dimensionally verified and restructured by the legal semantic constraint decoder, and the generation of the legal consulting report includes the following steps:

[0033] The semantic vector sequence is converted into legal attribute labels through the legal semantic distillation encoder;

[0034] Based on the legal database, the validity of the legal provisions with legal attribute labels is queried, and the legal attribute labels of the validity of the legal provisions are verified in multiple dimensions through the timeliness verification engine combined with the legal revision map method. The legal attribute labels of the validity of the legal provisions are structurally reorganized using the dynamic priority sorting algorithm method to obtain a legal consulting report.

[0035] In a second aspect, the present invention provides a report generation system based on a legal knowledge graph, including a topological network module, which obtains user legal demands through a non-standardized domain corpus collection engine, performs multi-dimensional intent deconstruction through a semantic slicing model, and obtains a judicial element topological network;

[0036] The semantic slicing module inputs the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, calculates the decision vector weights through the dynamic value evaluation network, and generates the judicial decision path using the temporal feedback mechanism;

[0037] The deep strategy module, based on the judicial decision-making path, generates a dynamic weighted legal search confidence matrix through the association rule mining engine of the multimodal knowledge graph, combined with semantic coupling analysis and logical path influence assessment;

[0038] The compliance reporting module uses the compliance situation awareness engine to verify the consistency of rule paths and generates compliance reports with correction suggestions by quantifying risk entropy values.

[0039] The consulting report module uses a legal semantic constraint decoder to perform multi-dimensional verification and structural reorganization of the revised compliance report to generate a legal consulting report.

[0040] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the report generation method based on the legal knowledge graph as described in the first aspect of the present invention.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of a report generation method based on a legal knowledge graph as described in the first aspect of the present invention.

[0042] The beneficial effects of the present invention are: through the collaborative innovation of the legal deduction engine with dynamic value perception capabilities and the legal semantic constraint decoder, a qualitative change from information retrieval to intelligent decision-making in legal consultation is achieved. In terms of the reasoning engine, a graph attention network is used to extract legal node features, and the decision vector is dynamically optimized in combination with the temporal difference strategy distillation collaborative algorithm and the heterogeneous attention gating mechanism, so that legal reasoning has the ability of continuous evolution. Tests show that the reasoning accuracy is significantly improved, and the state changes of legal elements are accurately captured through LSTM temporal feedback, which greatly improves the efficiency of generating decision paths for complex cases. In terms of semantic decoding, the legal semantic distillation encoder and the legal revision map are innovatively integrated to achieve real-time verification of the timeliness of the provisions, and the response speed to legal revisions is far faster than manual verification. Through the structured reorganization of the dynamic priority algorithm, the report suggestion ranking is more in line with judicial practice, effectively solving the long-standing problems of static knowledge base lag and discretized suggestion in the field of legal consultation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 Flowchart of the method for generating legal consulting reports based on legal knowledge graph.

[0045] Figure 2 System diagram for the legal consulting report generation system based on the legal knowledge graph.

[0046] Figure 3 This is a diagram of the judicial decision-making path architecture.

[0047] Figure 4 Flowchart for dynamically weighted legal search confidence matrix. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0051] Reference Figures 1 to 4 , which is the first embodiment of the present invention, provides a report generation method based on a legal knowledge graph, comprising the following steps:

[0052] S1. Obtain users' legal demands through a non-standardized domain corpus collection engine, perform multi-dimensional intention deconstruction through a semantic slicing model, and obtain a judicial element topology network.

[0053] S1.1. Input the user's legal claim data into the semantic slicing model, and parse the user's legal claim into a judicial element parsing vector through the multi-granularity intent recognition and parsing method.

[0054] Furthermore, the user's legal claims are encoded through a language model pre-trained in the legal field, generating a word vector sequence of legal semantic features, identifying legal terms at the lexical level, capturing the causal relationship of the claims at the sentence level, and analyzing the overall litigation intent at the paragraph level. Multi-granularity features are integrated through a bidirectional LSTM network to obtain analytical vectors of three types of judicial elements: entities, behaviors, and relationships.

[0055] S1.2. Use the weight-compensated word segmentation algorithm to dynamically segment and tag the judicial element parsing vector, and map the annotation results into triples through domain-enhanced semantic role annotation.

[0056] Furthermore, a dynamic window mechanism is used to adapt to legal terms of different lengths, and a domain-enhanced semantic role labeling module is used to construct a legal-specific predicate-argument framework, mapping the word segmentation results into a triple structure of (subject, legal act, and object). A graph neural network trained on legal cases is used to improve labeling accuracy by capturing typical argument patterns in case law.

[0057] S1.3. Use judicial knowledge prior rules to eliminate ambiguous triples, perform logical consistency verification on the triples, and obtain an iterative domain knowledge topology map.

[0058] Furthermore, the initial rule base is automatically constructed through case-based statistical learning, and then manually verified and supplemented by legal experts. In the ambiguity elimination stage, a constraint satisfaction-based reasoning algorithm is used to optimize the labeling results by minimizing the logical conflict scores between triples. The generated domain knowledge topology map is stored in a dynamic graph structure, with nodes representing legal elements and edge weights reflecting the strength of the logical association between elements.

[0059] S2. Input the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, and calculate the decision vector weight through the dynamic value evaluation network.

[0060] S2.1. Use the graph attention network method to extract the legal node embedding data in the knowledge topology graph and obtain the initial legal decision through the multi-layer perceptron.

[0061] Furthermore, the topological relationship between nodes is established through the legal concept hierarchy. The first layer of attention of the network focuses on the type correlation of legal elements and calculates the feature propagation weights between similar legal nodes. The second layer of attention analyzes the logical dependencies between cross-type legal elements. After the node features are subjected to the graph convolution operation, a multi-layer perceptron consisting of three fully connected layers is obtained. Each hidden layer adopts an activation function specific to the legal field to obtain the initial legal decision vector in dimensions such as the probability of winning prediction and the responsibility division recommendation.

[0062] S2.2. Use the temporal difference strategy distillation collaborative algorithm to optimize the initial legal decision and build a legal deduction engine with dynamic value perception capabilities.

[0063] Furthermore, the temporal difference strategy distillation collaborative algorithm framework adopts a dual network structure, which includes two parallel legal reasoning models: the behavioral network and the target network. The behavioral network processes the input judicial element topology network in real time and generates a preliminary decision-making plan. The target network updates parameters based on the historical optimal decision. Knowledge transfer is achieved between the two networks through periodic parameter synchronization. At the same time, a priority experience replay mechanism is introduced, focusing on retaining typical judicial decision-making cases for model training. KL divergence is used as a metric for strategy improvement during the optimization process to ensure the progressive optimization of legal reasoning logic.

[0064] S2.3. Input the legal node embedding data into the legal deduction engine with dynamic value perception capability, and use the heterogeneous attention gating mechanism to calculate the decision vector weight, which is expressed as:

[0065]

[0066] Among them, G is the decision vector weight, V is the legal value matrix, K is the legal key matrix, Q is the legal query vector, and d is the dimension adjustment factor.

[0067] Furthermore, the heterogeneous attention gating mechanism constructs multiple attention subspaces operating in parallel, each focusing on different types of legal logic relationship analysis. The query vector is derived from the legal feature encoding of the current case, while the key and value matrices are composed of precedent features from the legal knowledge base. The computation process first encodes the query vector's legal domain-specific position, then calculates the correlation strength of each legal element using scaled dot-product attention. The final output decision vector weights are filtered through a legal logic constraint layer to eliminate anomalous weight assignments that violate fundamental legal principles.

[0068] S3. Use the temporal feedback mechanism to generate judicial decision-making paths.

[0069] S3.1. Use the state memory-contradiction perception collaborative network to record the state changes of legal elements of decision vector weights.

[0070] Furthermore, the State Memory-Contradiction Perception Collaborative Network utilizes a gated recurrent unit design specifically optimized for the temporal nature of legal decision-making processes. The input layer receives a sequence of decision vector weights from the dynamic value assessment network. A forget gate controls the degree of retention of historical legal element states, while an input gate adjusts the integration strength of new decision features. The hidden state output at each time step contains the evolutionary trajectory of case elements, with a particular focus on the state transition patterns of key legal elements. Training utilizes time series data from legal cases, focusing on learning the correspondence between verdicts and changes in element states.

[0071] S3.2, through the temporal causal convolution, the logical contradictions in the state changes of potential legal elements are mined and correction signals are generated.

[0072] Furthermore, temporal causal convolution mining potential constructs a multi-scale feature extraction architecture, which includes parallel one-dimensional convolution branches to process legal status changes in different time spans respectively. The network uses dilated convolution technology to expand the receptive field and identifies logical contradiction patterns from short-term to long-term through a hierarchical feature pyramid. The output layer is connected to the contradiction classifier to convert the detected abnormal pattern into a directional correction signal vector. The signal strength reflects the severity of the logical contradiction. The network training uses a manually annotated legal logic contradiction dataset to improve the recognition ability of implicit contradictions through adversarial training.

[0073] S3.3. Use policy entropy weighted fusion and probability deduction tree search to combine the decision vector weight and the correction signal, and use the Monte Carlo tree search method to generate the judicial decision path.

[0074] Furthermore, the combination ratio of the correction signal with the original decision vector is dynamically adjusted according to the credibility of the correction signal. The Monte Carlo tree search constructs a legal-specific search strategy, expands each legal decision node into a set of possible subsequent legal states, and evaluates the expected legal effects of each path through simulation and deduction. The search process introduces a legal constraint pruning strategy to eliminate decision branches that violate basic legal principles in advance. The final generated judicial decision path contains a complete legal reasoning chain and the expected effect evaluation of key nodes.

[0075] S4. Based on the judicial decision-making path, the association rule mining engine of the multimodal knowledge graph is used, and the semantic coupling degree is used for analysis to generate a dynamic weighted legal retrieval confidence matrix.

[0076] S4.1. The judicial decision-making path is converted into a time-aware legal feature vector through a spatiotemporal modal collaborative reasoning framework. The legal feature vector is calculated using the Apriori algorithm and a cross-modal graph neural network to obtain an association rule set.

[0077] Furthermore, the spatiotemporal modal collaborative reasoning framework adopts a bidirectional GRU architecture, which scans the judicial decision-making path forward and backward through the time axis, and outputs a legal feature vector that integrates historical and future contextual information at each time step. The Apriori algorithm has been adapted and improved in the legal field, setting the minimum support as a dynamic threshold to automatically discover high-frequency co-occurring legal element combinations. The cross-modal graph neural network constructs a heterogeneous node connection framework, and simultaneously processes three types of modal data: legal text nodes, case nodes, and factual element nodes. Cross-modal association rules are learned through the graph attention mechanism, and the contrast loss function is used during network training to enhance the ability to capture potential associations between different modalities.

[0078] S4.2. Analyze the association rule set using the legal semantic coupling analysis model and calculate the cross-modal semantic similarity of the element pairs in the association rule set.

[0079] Furthermore, the legal semantic coupling analysis model constructs a dual-tower structure, which performs deep semantic encoding on legal provisions and case facts respectively. It adopts an adversarial training strategy and encourages the two encoders to learn modality-independent common legal feature representations through the discriminator network. The similarity calculation stage introduces a distance measurement function specific to the legal field, focusing on professional dimensions such as requirement matching and responsibility relevance. It optimizes through triplet loss to ensure that the distance between similar case-legal pairs in the embedding space is smaller than that of unrelated pairs.

[0080] S4.3. The cross-modal semantic similarity is integrated through the gated attention fusion mechanism to generate a dynamic weighted legal retrieval confidence matrix.

[0081] Furthermore, the gated attention fusion mechanism designs a learnable gating weight vector to dynamically adjust the contribution ratio of different modal features. The mechanism includes two levels: feature-level gating and rule-level gating. The former controls the confidence calculation method of a single association rule, and the latter determines the integration strategy of cross-modal features. The generated dynamic weighted matrix adopts a three-dimensional tensor structure, which records the confidence scores of three types of associations: law-case, law-fact, and case-fact, and performs rationality calibration through the legal logic constraint layer. The legal retrieval confidence matrix update module tracks changes in judicial practice in real time to maintain the timeliness of the confidence score.

[0082] S5. Use the compliance situation awareness engine to verify the consistency of rule paths and generate a compliance report with correction suggestions by quantifying the risk entropy value.

[0083] S5.1. Use a bidirectional graph matching algorithm to compare the dynamic weighted legal search confidence matrix with the judicial decision path, and calculate the clause coverage and conflict coefficient indicators.

[0084] Furthermore, the bidirectional graph matching algorithm adopts an improved Hungarian algorithm framework to establish an optimal correspondence between the judicial decision-making path and the legal retrieval confidence matrix, construct a bipartite graph structure, and use the legal requirements in the decision-making path as the left node and the legal rules in the confidence matrix as the right node. The edge weight calculation comprehensively considers the two dimensions of semantic similarity and logical compatibility, and solves the maximum weight matching through dynamic programming. Two core indicators are generated during the verification process. The clause coverage reflects the proportion of key legal requirements covered by the decision-making path, and the conflict coefficient quantifies the degree of deviation between the path and the current legal system.

[0085] S5.2. Construct a multi-scale risk assessment model based on clause coverage and conflict coefficient indicators, and calculate the risk entropy value, which is expressed as:

[0086] Among them, H is the risk entropy value, U i is the risk quantification value of the i-th dimension, ω i is the contribution weight of the i-th dimension, and i is the dimension index.

[0087] Furthermore, the multi-scale risk assessment model adopts a hierarchical architecture design, which includes three evaluation dimensions: micro, meso and macro. The micro dimension focuses on the applicability accuracy of individual legal provisions, the meso dimension analyzes the system coordination of the combination of legal requirements, and the macro dimension evaluates the expected social effects of the overall decision-making path. The weight distribution strategy of each dimension is learned through a case library annotated by legal experts, and the attention mechanism is used to dynamically adjust the assessment focus of different case types. The risk entropy value calculation introduces temperature coefficient adjustment to ensure the comparability of the results in different legal fields.

[0088] S5.3. Quantify the risk entropy value through a multi-scale risk entropy fusion algorithm, combine gradient backpropagation with Monte Carlo simulation, and generate a compliance report with correction suggestions.

[0089] Furthermore, the multi-scale risk entropy fusion algorithm constructs a feature pyramid structure, captures risk patterns of different granularities through void space pyramid pooling, inputs the risk entropy value into a deep network composed of residual connection blocks, and the network output layer is connected to a text generation module adapted to the legal field. The gradient backpropagation process adopts a curriculum learning strategy to gradually improve the modeling ability of complex legal logic. The Monte Carlo simulation stage constructs a legal state transition model, and generates a differentiated set of correction plans through random sampling, which are integrated into a structured compliance report containing risk level labels and specific improvement suggestions.

[0090] S6. Use the legal semantic constraint decoder to perform multi-dimensional verification and structural reorganization of the compliance report with revised suggestions to generate a legal consulting report.

[0091] S6.1. Convert the semantic vector sequence into legal attribute labels through the legal semantic distillation encoder.

[0092] Furthermore, the general legal corpus is fine-tuned by incorporating case documents from specific jurisdictions, extracting semantic features through a multi-level Transformer architecture, and connecting the legal attribute classifier at the output layer. This classifier adopts a hierarchical labeling system design, first identifying the basic legal categories, and then refining them into specific legal relationships and liability types. Adversarial training technology is introduced into the encoding process to ensure that the generated attribute labels are not affected by differences in the surface expression of the text, but reflect the deep-level legal characterization.

[0093] S6.2. Based on the legal database, the validity of the legal provisions with legal attribute labels is queried. The legal attribute labels of the validity of the legal provisions are verified in multiple dimensions through the timeliness verification engine combined with the legal revision map method. The legal attribute labels of the validity of the legal provisions are structurally reorganized using the dynamic priority sorting algorithm method to obtain a legal consulting report.

[0094] Furthermore, the timeliness verification engine constructs a legal timeliness knowledge graph, with nodes representing legal provisions and edges recording timeliness relationships such as amendments and repeals. A graph traversal algorithm automatically detects the validity status of cited legal provisions. A multi-dimensional verification system implements a three-level review: a syntactical layer checks the compliance of the citation format, a semantic layer verifies the contextual fit of the provisions, and a practical layer evaluates the adoption rate in judicial practice. A dynamic priority sorting algorithm employs a multi-objective optimization framework, simultaneously considering the three dimensions of legal effectiveness, consistency with the adjudication's essence, and urgency of implementation. Pareto frontier analysis is used to determine the optimal sequence of recommendations, ultimately generating a structured consulting report containing legal basis, risk warnings, and implementation guidance.

[0095] This embodiment also provides a report generation system based on a legal knowledge graph, including:

[0096] The topological network module uses a non-standardized domain corpus collection engine to obtain users' legal demands, deconstructs multi-dimensional intent through a semantic slicing model, and obtains a judicial element topological network.

[0097] The semantic slicing module inputs the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, calculates the decision vector weights through the dynamic value evaluation network, and generates the judicial decision path using the temporal feedback mechanism;

[0098] The deep strategy module, based on the judicial decision-making path, generates a dynamic weighted legal search confidence matrix through the association rule mining engine of the multimodal knowledge graph, combined with semantic coupling analysis and logical path influence assessment;

[0099] The compliance reporting module uses the compliance situation awareness engine to verify the consistency of rule paths and generates compliance reports with correction suggestions by quantifying risk entropy values.

[0100] The consulting report module uses a legal semantic constraint decoder to perform multi-dimensional verification and structural reorganization of the compliance report with revised suggestions to generate a legal consulting report.

[0101] Furthermore, the non-standardized domain corpus collection engine collects legal consultation texts, voices, and documents submitted by users in real time through a multi-source data interface, uses an adaptive cleaning algorithm to eliminate noise data, and obtains a judicial element topology network.

[0102] Semantic Slicing System: A legal-specific NLP model based on a hierarchical attention mechanism decomposes user requests into entity-relationship-attribute triplets and constructs a weighted judicial element topology network.

[0103] Deep Strategy Optimization Inference Engine: This hybrid architecture integrates graph neural networks and reinforcement learning. It uses a dynamic value assessment network to calculate the decision weights of various legal elements and generate a judicial decision path with time series markings.

[0104] Multimodal knowledge graph system: This system integrates a three-dimensional knowledge graph of legal provisions, judicial interpretations, and precedents, and uses a cross-modal attention mechanism to generate a confidence matrix that reflects the strength of legal associations.

[0105] Compliance Situation Awareness Engine: Detects deviations between decision paths and existing laws based on a bidirectional graph matching algorithm and outputs compliance reports with graded warnings through a risk entropy quantification model.

[0106] Legal Semantic Constraint Decoder: Using Legal-BERT and a dynamic priority sorting algorithm, compliance recommendations are reorganized into a legal advisory report that complies with judicial document standards, consisting of three parts: legal basis, risk warnings, and operational guidelines.

[0107] This embodiment also provides a computer device, which is suitable for a report generation method based on a legal knowledge graph, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a report generation method based on a legal knowledge graph as proposed in the above embodiment.

[0108] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0109] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a report generation method based on a legal knowledge graph as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0110] In summary, the present invention has achieved a qualitative change in legal consultation from information retrieval to intelligent decision-making through the collaborative innovation of a legal deduction engine with dynamic value perception capabilities and a legal semantic constraint decoder. In terms of the reasoning engine, a graph attention network is used to extract legal node features, and the decision vector is dynamically optimized in combination with a temporal difference strategy distillation collaborative algorithm and a heterogeneous attention gating mechanism, so that legal reasoning has the ability of continuous evolution. Tests show that the reasoning accuracy is significantly improved, and the state changes of legal elements are accurately captured through LSTM temporal feedback, which greatly improves the efficiency of generating decision paths for complex cases. In terms of semantic decoding, the legal semantic distillation encoder and the legal revision map are innovatively integrated to achieve real-time verification of the timeliness of the provisions, and the response speed to legal revisions far exceeds manual verification. Through the structured reorganization of the dynamic priority algorithm, the report recommendation ranking is more in line with judicial practice, effectively solving the long-standing problems of static knowledge base lag and discretized recommendation in the field of legal consultation.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A report generation method based on a legal knowledge graph, characterized by: include, The user's legal demands are acquired through a non-standardized domain corpus collection engine, and multi-dimensional intent deconstruction is performed through a semantic slicing model to obtain a judicial element topology network. Input the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, calculate the decision vector weights through the dynamic value evaluation network, and use the temporal feedback mechanism to generate the judicial decision path; Based on the judicial decision-making path, a dynamic weighted legal search confidence matrix is generated through the association rule mining engine of the multimodal knowledge graph, combined with semantic coupling analysis and logical path influence assessment; Use the compliance situation awareness engine to verify rule path consistency and generate a compliance report with correction suggestions by quantifying risk entropy values; The legal semantic constraint decoder is used to perform multi-dimensional verification and structural reorganization of the compliance report with revised suggestions to generate a legal consulting report.

2. The report generation method based on the legal knowledge graph according to claim 1 is characterized in that: The user's legal demands are acquired through the non-standardized domain corpus collection engine, and multi-dimensional intention deconstruction is performed through the semantic slicing model to obtain the judicial element topology network, which includes the following steps: Input the user's legal demand data into the semantic slicing model, and parse the user's legal demand into a judicial element parsing vector through a multi-granular intent recognition and parsing method; A weight-compensated word segmentation algorithm is used to dynamically segment and tag judicial element parsing vectors. Domain-enhanced semantic role labeling is used to map the labeling results into triples. The judicial knowledge prior rules are used to eliminate ambiguous triples, and the triples are logically checked to obtain an iterative domain knowledge topology map.

3. The report generation method based on the legal knowledge graph according to claim 2 is characterized in that: Inputting the judicial element topology network into the legal deduction engine with dynamic value perception capability and calculating the decision vector weight through the dynamic value evaluation network includes the following steps: Use the graph attention network method to extract legal node embedding data from the knowledge topology graph and obtain the initial legal decision through the multi-layer perceptron; Use the temporal difference strategy distillation collaborative algorithm to optimize initial legal decisions and build a legal deduction engine with dynamic value perception capabilities; The legal node embedding data is input into the legal deduction engine with dynamic value perception capability, and the decision vector weight is calculated using a heterogeneous attention gating mechanism.

4. The report generation method based on the legal knowledge graph according to claim 3 is characterized in that: The generation of judicial decision paths using the temporal feedback mechanism includes the following steps: Use the state memory-contradiction perception collaborative network to record the state changes of legal elements of decision vector weights; Through temporal causal convolution, we can mine the logical contradictions in the state changes of potential legal elements and generate correction signals. The decision vector weights and correction signals are combined using policy entropy weighted fusion and probability deduction tree search, and the Monte Carlo tree search method is used to generate the judicial decision path.

5. The report generation method based on the legal knowledge graph according to claim 4 is characterized in that: Based on the judicial decision path, through the association rule mining engine of the multimodal knowledge graph and using semantic coupling analysis, the dynamic weighted legal retrieval confidence matrix is generated, which includes the following steps: The judicial decision path is converted into a time-aware legal feature vector through a spatiotemporal modal collaborative reasoning framework. The legal feature vector is calculated using the Apriori algorithm and a cross-modal graph neural network to obtain an association rule set. The association rule set is analyzed through the legal semantic coupling analysis model, and the cross-modal semantic similarity of the element pairs in the association rule set is calculated; The cross-modal semantic similarity is integrated through the gated attention fusion mechanism to generate a dynamic weighted legal retrieval confidence matrix.

6. The report generation method based on the legal knowledge graph according to claim 5 is characterized in that: Use the compliance situation awareness engine to verify the consistency of rule paths and generate compliance reports with correction suggestions by quantifying risk entropy values. The following steps are included: A bidirectional graph matching algorithm is used to compare the dynamic weighted legal retrieval confidence matrix with the judicial decision path, and to calculate the clause coverage and conflict coefficient indicators. Construct a multi-scale risk assessment model based on clause coverage and conflict coefficient indicators, and calculate the risk entropy value; The risk entropy value is quantified through a multi-scale risk entropy fusion algorithm, and combined with gradient backpropagation and Monte Carlo simulation to generate a compliance report with correction suggestions.

7. The report generation method based on the legal knowledge graph according to claim 6 is characterized in that: The legal semantic constraint decoder is used to perform multi-dimensional verification and structural reorganization of the compliance report of the revised proposal. The generation of the legal consulting report includes the following steps: The semantic vector sequence is converted into legal attribute labels through the legal semantic distillation encoder; Based on the legal database, the validity of the legal provisions with legal attribute labels is queried, and the legal attribute labels of the validity of the legal provisions are verified in multiple dimensions through the timeliness verification engine combined with the legal revision map method. The legal attribute labels of the validity of the legal provisions are structurally reorganized using the dynamic priority sorting algorithm method to obtain a legal consulting report.

8. A report generation system based on a legal knowledge graph, based on the report generation method based on a legal knowledge graph according to any one of claims 1 to 7, characterized in that: include, The topological network module uses a non-standardized domain corpus collection engine to obtain users' legal demands, deconstructs multi-dimensional intent through a semantic slicing model, and obtains a judicial element topological network. The semantic slicing module inputs the judicial element topology network into the legal deduction engine with dynamic value perception capabilities, calculates the decision vector weights through the dynamic value evaluation network, and generates the judicial decision path using the temporal feedback mechanism; The deep strategy module, based on the judicial decision-making path, generates a dynamic weighted legal search confidence matrix through the association rule mining engine of the multimodal knowledge graph, combined with semantic coupling analysis and logical path influence assessment; The compliance reporting module uses the compliance situation awareness engine to verify the consistency of rule paths and generates compliance reports with correction suggestions by quantifying risk entropy values. The consulting report module uses a legal semantic constraint decoder to perform multi-dimensional verification and structural reorganization of the compliance report with revised suggestions to generate a legal consulting report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a report generation method based on a legal knowledge graph as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a report generation method based on a legal knowledge graph as described in any one of claims 1 to 7 are implemented.

Images