Case reasoning teaching platform for economic law based on knowledge graph

Abstract

The application discloses a knowledge graph-based economic law case reasoning teaching platform, and relates to the technical field of digital teaching, wherein by integrating regulation clauses, case facts and academic paper data, regulation entities, case entities and academic entities are extracted and cross-dimension reference relations are constructed. Entity feature vectors and semantic relation weights are embedded into graph nodes and edges to form a structured legal knowledge network. Based on the case facts input by a user, an interpretable reasoning report containing regulation application paths is generated through semantic similarity retrieval. The regulation revision state is tracked in real time, and the validity markers of clauses and the associated case reference states in the graph are dynamically updated to ensure the legal timeliness of teaching resources. Combined with the entity nodes currently browsed by the user and the timeliness marker data, a personalized recommendation list is generated, and the knowledge graph edge weights are adjusted in real time through feedback operation data; the structured expression and dynamic evolution of legal knowledge are realized.

Description

Technical Field

[0001] This invention relates to the field of digital teaching technology, specifically to a knowledge graph-based teaching platform for economic law case reasoning. Background Technology

[0002] With the deepening application of digital technology in legal education, economic law case teaching platforms are gradually evolving from traditional text retrieval to intelligent reasoning services. Current mainstream teaching systems mostly adopt a technical architecture combining keyword matching and database querying, achieving basic retrieval functions through structured storage of legal provisions and related cases. However, existing technological systems have significant shortcomings in addressing the unique dynamics, interconnectedness, and adaptability to teaching scenarios specific to the field of economic law: economic law teaching requires simultaneous involvement of three types of data sources: legal provisions, judicial cases, and academic theories. Existing systems typically employ independent database storage structures, making it difficult to effectively model the applicability relationships between regulations and cases, and the supporting connections between academic theories and legal practice. Furthermore, the frequency of legal revisions in the economic law field is high (an average annual revision rate exceeding 15%), and existing platforms largely rely on manual annotation of timeliness status, failing to track the revision status in real time. This means that legal provisions cited in case teaching may be repealed or invalid, posing a risk of misleading instruction. In addition, existing systems lack intelligent analysis capabilities for user learning behavior, failing to provide personalized case recommendations based on user learning progress and knowledge gaps, thus failing to meet the needs of learners at different levels. In terms of case-based reasoning, existing systems can only provide simple case retrieval functions and cannot achieve in-depth reasoning based on legal logic, making it difficult to help learners build a complete framework for legal application thinking. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a knowledge graph-based teaching platform for economic law case reasoning.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows:

[0005] A knowledge graph-based teaching platform for economic law case-based reasoning includes:

[0006] A multi-source data acquisition module is used to acquire raw text data, which includes legal provisions, case fact descriptions, and paper abstracts.

[0007] The semantic parsing module is used to perform structured parsing on the original text data, extract legal entities, case entities, and academic entities, and generate reference relationship data between entities;

[0008] The knowledge graph construction module is used to construct an economic law knowledge graph based on the legal entities, case entities, academic entities and the reference relationships between entities; the nodes in the knowledge graph store the entity feature vectors corresponding to the legal entities, case entities, and academic entities, and the edges store the semantic relationships and weights of preset types;

[0009] The case reasoning module is used to receive case facts input by the user, retrieve similar case entities based on the knowledge graph, and generate reasoning result data containing the applicable legal path;

[0010] The timeliness update module is used to monitor the revision status data of legal provisions, parse the effective time, repeal time and substitution relationship data in the revised legal text, generate timeliness mark data and update the validity mark of legal entities and the reference status of related cases in the knowledge graph.

[0011] The dynamic recommendation and feedback engine generates a personalized recommendation list based on the entity nodes currently viewed by the user and time-sensitive data. It also dynamically adjusts the edge weights of the knowledge graph through feedback operation data and triggers timeliness verification simultaneously.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0013] 1. Through the knowledge graph construction and time-sensitive update module, scattered data is transformed into a dynamically evolving semantic network, supporting adaptive adjustments to changes in the legal environment;

[0014] 2. By employing semantic parsing and graph reasoning techniques, legal elements are extracted from case facts to achieve multi-dimensional calculation of case similarity, significantly improving the matching accuracy of similar causes of action;

[0015] 3. Through a dynamic recommendation and feedback engine, user operation data is injected into the knowledge graph for weight adjustment, enabling the platform to have a self-learning ability that becomes more accurate with use, thus forming a closed-loop teaching system. Attached Figure Description

[0016] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0017] Figure 1 This is a diagram showing the system module composition of the present invention;

[0018] Figure 2 This is a flowchart illustrating the generation of a personalized recommendation list according to the present invention;

[0019] Figure 3 This is a flowchart of the feedback closed-loop linkage module of the present invention;

[0020] Figure 4 This is a flowchart illustrating the workflow of the time-update module of the present invention.

[0021] Figure 5 This is a flowchart of the case reasoning module of the present invention. Detailed Implementation

[0022] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0023] Application Overview

[0024] In traditional economic law case teaching platforms, the three data sources—legal provisions, judicial cases, and academic theories—are stored in independent databases. This results in a lack of effective modeling of the applicable relationships and supporting connections between entities, making it impossible to track the status of legal revisions and update the expiration dates of related cases in real time. When users initiate case reasoning requests based on case fact descriptions, the system can only perform keyword matching based on independently stored case databases. It cannot identify the connection paths between cases and the latest legal provisions or academic theories, nor can it verify the expiration status of cited provisions. For example, in scenarios where users input facts related to cases after the revision of the Anti-Monopoly Law, similar cases retrieved by the system may cite outdated legal provisions that have been superseded, and the associated academic papers may not contain theoretical analysis of the revised provisions. Because the citation relationship between case entities and legal entities is not dynamically updated, the system cannot automatically identify the impact of expired provisions on the case's reference status, resulting in the reasoning results obtained by users containing outdated legal basis. If these problems are not addressed, the case reasoning results output by the system will face the risk of expired legal provisions for a long time, reducing the credibility and professionalism of the teaching scenario. Users cannot build cross-data source connections through a single platform, resulting in a lack of academic theoretical support for the analysis of legal application paths. Meanwhile, the lack of a timely verification mechanism will lead to an accumulation of discrepancies between the recommended results and the real legal environment, affecting the effectiveness of legal thinking training in case-based teaching.

[0025] Faced with the aforementioned problems, this application first addresses the two major issues of isolated data sources and disconnect from timeliness by reconstructing the technical approach. Traditional solutions rely on independent databases to store regulations, cases, and academic data, resulting in a lack of cross-source association paths. This application considers establishing a unified knowledge graph architecture to break down data silos through entity relationship mapping. However, simply building a static graph cannot solve the timeliness risks caused by dynamic revisions of regulations; a dynamic update mechanism for timeliness markers needs to be introduced. Further analysis reveals that if only historical association paths are relied upon during case reasoning, the impact of regulatory revisions on the reference value of existing cases may be overlooked. Therefore, timeliness verification needs to be linked with the recommendation system to form a closed-loop feedback. By integrating multi-source analysis, dynamic graph updates, and feedback-driven weight adjustments, the application of law ultimately achieves timeliness synchronization and cross-domain association of legal application paths.

[0026] In this regard, such as Figure 1 As shown, this application proposes a knowledge graph-based teaching platform for economic law case reasoning, including:

[0027] The multi-source data acquisition module is used to acquire raw text data, including legal provisions, case fact descriptions, and paper abstracts. The multi-source data acquisition module is a component that integrates three heterogeneous data sources: legal provisions, case fact descriptions, and paper abstracts. Specifically, it can be implemented by using web crawlers combined with API interface calls, which solves the problem of difficulty in correlation modeling caused by isolated data sources in traditional systems.

[0028] The semantic parsing module is used to perform structured parsing of the original text data, extract legal entities, case entities, and academic entities, and generate reference relationship data between entities. The semantic parsing module refers to the processing unit that performs entity recognition and relationship extraction on unstructured text. Specifically, it can be implemented by combining a deep learning-based named entity recognition model with a rule engine to solve the problem of missing entity associations caused by the semantic complexity of legal texts.

[0029] The knowledge graph construction module is used to construct an economic law knowledge graph based on legal entities, case entities, academic entities, and the reference relationships between entities. Nodes in the knowledge graph store entity feature vectors corresponding to legal entities, case entities, and academic entities, while edges store semantic relationships and weights of preset types. The knowledge graph construction module refers to a graph database structure that stores entity feature vectors and semantic relationships. Specifically, it can be implemented using the Neo4j graph database combined with vectorization representation algorithms to achieve visualized modeling and efficient retrieval of dynamic relationships in legal knowledge.

[0030] The case reasoning module receives case facts input by the user, retrieves similar case entities based on the knowledge graph, and generates reasoning result data containing the applicable legal path. The case reasoning module is an intelligent analysis component based on graph neural network similarity matching. Specifically, it can be implemented by combining graph attention network with cosine similarity algorithm to solve the problem that traditional keyword matching cannot generate the applicable legal path.

[0031] The timeliness update module is used to monitor the revision status data of legal provisions, parse the effective date, repeal date and substitution relationship data in the revised legal text, generate timeliness mark data and update the validity mark of legal entities and the reference status of related cases in the knowledge graph. The timeliness update module refers to the data processing unit that automatically tracks the revision status of regulations. Specifically, it can be implemented by using natural language processing technology to parse the timestamps and clause change types in the revised text, eliminating the risk of misleading teaching caused by the lag of manual annotation.

[0032] The dynamic recommendation and feedback engine is used to generate a personalized recommendation list based on the entity nodes currently viewed by the user and the time-sensitive data. The aforementioned entity nodes currently viewed by the user refer to the nodes whose entity feature vectors correspond to legal entities, case entities, and academic entities stored during the construction of the economic law knowledge graph.

[0033] Furthermore, the knowledge graph edge weights are dynamically adjusted based on feedback operation data, and timeliness verification is triggered synchronously. The dynamic recommendation and feedback engine refers to an adaptive system that adjusts the knowledge graph weights according to user behavior. Specifically, it can be implemented by combining collaborative filtering algorithms with a real-time weight update mechanism to solve the problem that static recommendations cannot adapt to the dynamic evolution of legal knowledge.

[0034] This application constructs a legal knowledge graph system with timeliness awareness capabilities. Through multi-source data fusion, dynamic relationship modeling, and feedback closed-loop mechanism, it realizes full-dimensional correlation analysis and real-time updates of laws, cases, and theories in legal teaching scenarios, effectively solving the three major technical bottlenecks of traditional systems: data silos, timeliness delays, and rigid recommendations.

[0035] The working process and principle of this application are as follows:

[0036] The multi-source data acquisition module acquires raw text data, including legal provisions, case fact descriptions, and paper abstracts. The semantic parsing module performs structured parsing of the raw text data, extracting legal entities, case entities, and academic entities, and generating citation relationship data between entities. The knowledge graph construction module constructs an economic law knowledge graph based on the extracted entity and citation relationship data. Nodes in the graph store entity feature vectors, and edges store semantic relationships and weights. The case reasoning module receives case facts input by the user, retrieves similar case entities based on the knowledge graph, and generates reasoning result data containing the applicable legal path. The timeliness update module monitors the revision status data of legal provisions, parses time and substitution relationship data in the revised text, generates timeliness marker data, and updates the validity markers of legal entities and the reference status of related cases in the graph. The dynamic recommendation and feedback engine generates a personalized recommendation list based on the entity nodes currently viewed by the user and the timeliness marker data, dynamically adjusts the graph edge weights through feedback operation data, and synchronously triggers timeliness verification.

[0037] The modules work together to achieve the expected functions:

[0038] The multi-source data acquisition module provides basic data for knowledge graphs;

[0039] The semantic parsing module transforms unstructured text into structured data, providing input for graph construction;

[0040] The knowledge graph construction module integrates entity and relationship data into a unified graph structure, supporting subsequent case reasoning and dynamic updates;

[0041] The case reasoning module performs similarity matching based on the graph to generate reasoning results that include the path of legal application;

[0042] The timeliness update module monitors the status of regulatory revisions in real time to ensure the timeliness of legal knowledge in the map;

[0043] The dynamic recommendation and feedback engine generates personalized recommendations based on user behavior and timely data, and adjusts the graph structure through feedback data to form a closed-loop optimization.

[0044] This paper employs a knowledge graph structure to achieve the associative representation of multi-source data, overcoming the limitations of traditional independent databases. A timely update mechanism is introduced to address the timeliness risks caused by dynamic revisions of regulations. A dynamic recommendation and feedback engine is designed to achieve personalized recommendations and adaptive optimization of the knowledge graph structure. The selection of these features helps improve the system's relevance, timeliness, and adaptability.

[0045] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0046] The multi-source data acquisition module uses web crawling technology to obtain raw text data from official legal databases, judicial case databases, and academic paper databases. The semantic parsing module employs natural language processing technology to extract information such as the responsible party and constituent elements from legal provisions using rule matching methods; to extract elements such as the subject and behavior from case fact descriptions using named entity recognition and relation extraction algorithms; and to generate supporting relationships between theoretical viewpoints and legal provisions from paper abstracts using keyword extraction and co-occurrence analysis methods.

[0047] The knowledge graph construction module uses a graph database to store the extracted entity and relation data. Nodes store entity IDs, types, and feature vectors, while edges store relation types and weights. The case reasoning module transforms user-input case facts into feature vectors, calculates the similarity between these vectors and case entities in the graph using cosine similarity, selects cases with similarity scores above a threshold, and extracts their associated applicable legal paths.

[0048] The timeliness update module periodically checks for regulatory revision announcements, parses the revised content, and generates timeliness marker data. For detected revised regulations, it traverses the case entities in the graph that reference the regulations, updating the case's reference status based on the match between the revised content and the case facts. The dynamic recommendation and feedback engine expands the path along high-weight edges in the graph based on the entity nodes currently viewed by the user, generating a recommendation list. The system records the user's operation data on the recommendation list, calculates the incremental values ​​of edge weights, and updates the graph structure.

[0049] Through the above-described solutions, this application achieves an intelligent upgrade of the economic law case teaching platform. The introduction of a knowledge graph resolves the correlation issues among the three data sources: regulations, cases, and academic theories, supporting cross-source legal application path analysis. The timely update mechanism enables real-time tracking of regulatory revision status and dynamic updates of the reference status of related cases, reducing the risk of citing outdated legal provisions in teaching. The dynamic recommendation and feedback engine provides personalized learning paths and continuously optimizes the knowledge graph structure through user feedback, improving the system's adaptability. These improvements effectively enhance the accuracy and timeliness of case reasoning results, providing more reliable and professional intelligent support for economic law case teaching.

[0050] In some of the solutions mentioned above in this application, the dynamic recommendation service relies on the edge weights of the knowledge graph to generate a personalized recommendation list, but does not consider the impact of time status on path expansion, which may result in the recommendation results containing invalid nodes; at the same time, it lacks a mechanism to predict user feedback risks and cannot dynamically adjust the recommendation strategy.

[0051] like Figure 2The diagram shows a flowchart for generating a personalized recommendation list. This application further proposes to expand the path along the edges in the knowledge graph with weights higher than a preset threshold based on the entity nodes currently viewed by the user; prioritize recommending hub nodes that simultaneously connect legal entities, case entities, and academic entities, wherein the determination condition for a hub node is that the number of associated edges exceeds a preset number and the timeliness status is valid; pre-calculate the negative feedback risk probability for entities in the recommendation list, and if the probability exceeds a preset risk threshold, replace it with a backup entity.

[0052] During path expansion, only edges with weights higher than a preset threshold are selected to ensure the reliability of the recommended path. Hub nodes must have more than a preset number of associated edges and be in a valid time status to ensure that the recommended content covers multiple types of entities and is timely. The probability of negative feedback risk is dynamically assessed by linearly weighting the historical negative feedback rate, time-lapse coefficient, associated weight decay value, and the user's recent behavior score. When the probability of risk exceeds a preset threshold, a backup entity is used to replace the original recommended item to avoid negative user feedback.

[0053] Specifically, the path expansion process filters edges in the knowledge graph with weights higher than a preset threshold based on the entity nodes currently viewed by the user. For example, the edge weight must be greater than 0.8, generating a set of candidate paths. During the hub node determination process, nodes in the candidate paths are traversed, the number of associated edges is counted, and the timeliness status is verified. For example, the number of associated edges must exceed 5, and the timeliness status must be marked as valid. Nodes meeting these conditions are added to the recommendation list. For entities in the recommendation list, a risk probability is calculated using a linear formula. For example, the timeliness decay coefficient in the formula is set to 0.3, and the historical negative feedback rate weight is 0.5. If the calculation result exceeds 0.7, replacement logic is triggered. Entities with similarity higher than a preset value from the backup entity library are called, such as case entities with a cosine similarity greater than 0.85, and the recommendation list is updated. This process improves the accuracy of recommendation results and user satisfaction by dynamically filtering high-weight paths, verifying node timeliness, and predicting feedback risks, while avoiding misleading teaching due to recommended invalid content or high-risk entities.

[0054] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0055] Based on the entity node currently viewed by the user, the path is expanded along the edges in the knowledge graph with a weight higher than 0.7. The path expansion uses a breadth-first search algorithm with a search depth of 3 layers.

[0056] Hub nodes that connect regulatory entities, case entities, and academic entities are given priority. A hub node is defined as having more than 10 associated edges and a valid expiration date. Hub nodes are sorted in descending order of the number of associated edges.

[0057] The probability of negative feedback risk is pre-calculated for entities in the recommendation list. This probability is calculated using a weighted average of four indicators: historical negative feedback rate, time-related decay coefficient, association weight decay value, and recent user behavior score. If the probability exceeds 0.6, the entity is replaced with a backup entity. Backup entities are randomly selected from entities of the same type.

[0058] Through the above technical solutions, this application realizes a personalized recommendation function based on knowledge graphs. Through path expansion algorithms, the system can discover laws, cases, and academic resources related to the user's current interests. Prioritizing hub nodes improves the relevance and timeliness of recommended content. A negative feedback risk prediction mechanism reduces the possibility of inappropriate content recommendations, enhancing the user experience. Overall, this solution improves the relevance and effectiveness of case-based teaching in economic law.

[0059] In some of the solutions mentioned above in this application, the probability of negative feedback risk needs to be pre-calculated when generating a personalized recommendation list in order to optimize the recommendation results. However, the existing methods lack the coordinated consideration of multi-dimensional dynamic factors in probability calculation, which makes it impossible to accurately predict the potential negative feedback risk of users.

[0060] This application further proposes a formula for calculating the probability of negative feedback risk as a linear weighted calculation of historical negative feedback rate, time-effect decay coefficient, correlation weight decay value, and user recent behavior score.

[0061] Negative feedback risk probability P risk Based on historical negative feedback rate R hist , aging attenuation coefficient C time Association weight decay value W rel and user recent behavior score B recent The linear weighted calculation enables dynamic risk assessment; when the risk probability exceeds a preset threshold, a backup entity is used to replace the original recommended item to avoid negative user feedback.

[0062] The formula for calculating the probability of negative feedback risk is: P risk =αR hist +βC time +γW rel +δB recent α, β, γ, δ are the historical negative feedback rates R hist , aging attenuation coefficient C time Association weight decay value W rel and user recent behavior score B recent The preset weighting coefficients.

[0063] Historical negative feedback rate R hist The calculation formula is:

[0064]

[0065] Aging decay coefficient C time The calculation formula is:

[0066]

[0067] Correlation weight decay value Among them W edge W represents the edge weight of the user's current entity. max This represents the maximum weight of edges of the same type in the graph;

[0068] User Recent Behavior Score

[0069] Historical negative feedback rate R hist The historical negative feedback rate for similar entities (such as cases related to "contract breach") aims to capture the inherent risks of the entities themselves (e.g., the high negative feedback tendency of outdated legal provisions); the time-depletion coefficient C time To mitigate the diminishing impact of the timeliness of regulations on the reference value of cases, the aim is to dynamically reduce the recommendation priority of related older regulatory cases; the correlation weight decay value W rel This represents the degree of weight decay of entity-related edges in a knowledge graph, used to suppress the recommendation of weakly related paths (e.g., the association between outdated academic theories and new judicial interpretations); User recent behavior score B recent It provides feedback on the sensitivity of recommendations to users' short-term operational behavior, enabling rapid responses to changes in users' immediate preferences (such as recent focus on studying the "anti-monopoly" field).

[0070] Negative feedback risk probability P risk By dynamically quantifying the potential risks of recommended content, this approach addresses the negative feedback issues in legal education platforms caused by outdated knowledge, user preference biases, and weakened logical connections. Through multi-dimensional weighting (α, β, γ, δ are configurable coefficients), it achieves accurate prediction of risk probabilities, thereby optimizing the recommendation strategy.

[0071] Specifically, the historical negative feedback rate reflects users' long-term preference biases towards specific types of entities, the timeliness decay coefficient addresses the issue of entity association failure caused by regulatory revisions, the association weight decay value captures changes in the dynamic strength of associations between entities in the knowledge graph, and the user's recent behavior score quantifies the user's current interest tendencies. By linearly weighting four types of data—for example, setting the weight of the historical negative feedback rate at 0.3, the timeliness decay coefficient at 0.4, the association weight decay value at 0.2, and the user's recent behavior score at 0.1—the system can comprehensively assess the potential negative feedback risk of recommended entities. When the probability calculation result exceeds a threshold, it is automatically replaced with a backup entity that is timely and has a high degree of user behavior matching, avoiding the recommendation of outdated or insufficiently relevant content. This calculation model improves the accuracy and timeliness of recommendation results by integrating static historical data and dynamic timeliness data.

[0072] The weighting coefficients α, β, γ, and δ can be optimized and adjusted using machine learning methods to adapt to the needs of different application scenarios. The calculated negative feedback risk probability P... risk Used for subsequent adjustments to the recommendation strategy.

[0073] Through the above technical solution, this application can comprehensively consider multiple factors and quantitatively assess the negative feedback risk of recommended entities. This allows for a more accurate prediction of user acceptance of recommended content, effectively reducing the incidence of invalid recommendations and negative experiences, and improving the overall performance and user satisfaction of the recommendation system.

[0074] Example explanation:

[0075] The scenario is as follows: A user is studying the topic of "Company Equity Disputes," and the system recommends the case of "Dispute over Shareholders' Right to Know in Company A" (Case No. C-2023-058). The user needs to calculate the probability of negative feedback risk in this case. The relevant parameters are as follows:

[0076] Historical negative feedback rate R hist :

[0077] Total exposure count for similar entities (shareholder right to know disputes): 120 times;

[0078] Number of times users marked it as "irrelevant": 18;

[0079] Calculation yields This category of cases has a historical negative feedback rate of 15%.

[0080] Aging decay coefficient C time :

[0081] Relevant regulations: Article 33 of the Company Law (last revised on March 15, 2023);

[0082] Current time: 2024-01-01;

[0083] Preset decay period: 12 months (standard value in the field of economic law);

[0084] Calculation yields Because the regulations were revised nine months ago, the case's reference value has decreased by 75%. (Relevance weight decay value W) rel :

[0085] Current edge weight: The relevance weight W between Case C-2023-058 and Article 33 of the Company Law. edge =0.62;

[0086] Maximum weight of edges of the same type: W max =0.91 (the highest correlation between similar cases and legal provisions);

[0087] Calculation yields The case's relevance to the legal provisions is 31.8% lower than the best value among similar cases. User's recent behavior score is B. recent :

[0088] The number of times the user skipped "shareholder dispute" cases within 24 hours: 4;

[0089] System preset frequency threshold: 3 times;

[0090] Calculation yields User resistance has reached its threshold.

[0091] Weighting coefficients design: α = 0.4, β = 0.3, γ = 0.2, δ = 0.1.

[0092] Calculate: P risk =0.4×0.15+0.3×0.25+0.2×0.318+0.1×1=0.288;

[0093] If the decision threshold is set to 0.25, risk control must be initiated if the threshold is exceeded.

[0094] High-risk handling:

[0095] Case C-2023-058 was removed from the recommended list;

[0096] The backup case, "Dispute over Shareholder Dividend Rights of Company B" (P), was activated. risk =0.12) as a substitute.

[0097] Knowledge graph update:

[0098] Reduce the weight of associated edges;

[0099] Timeliness verification:

[0100] The statute of limitations decay factor C was recalculated in 32 cases that triggered Article 33 of the Company Law. time .

[0101] In some of the solutions mentioned above in this application, during the process of generating a recommendation list based on the entity nodes and time-sensitive marker data currently viewed by the user, user feedback data, including user feedback on personalized recommendation list operations, fails to be applied to the knowledge graph weight adjustment in real time. This results in a lag between the recommended content and user needs. At the same time, the edge weights that are not triggered lack an automatic weight reduction mechanism, which may retain outdated associations and affect the accuracy of recommendations. Furthermore, changes in time-sensitive status caused by regulatory revisions cannot trigger association verification.

[0102] This application further proposes a dynamic recommendation and feedback engine that includes a feedback closed-loop linkage module, such as... Figure 3The diagram shows the workflow of the feedback closed-loop linkage module, which performs the following data processing steps:

[0103] Receive secondary operation data from users on the recommendation list, parse the secondary operation data and generate real-time weight adjustment instructions;

[0104] The incremental values ​​of the knowledge graph edge weights are calculated based on the real-time weight adjustment instructions and the time-sensitive tag data, and the weights of the associated edges and the ranking of the hub nodes are updated.

[0105] The system synchronizes user feedback data, including user actions on personalized recommendation lists, to the timeliness update module, triggering monitoring of revisions to relevant regulations and verification of the timeliness of academic papers.

[0106] The feedback closed-loop linkage module includes:

[0107] The behavior log analysis unit parses secondary operation data to obtain user behavior time-series data and operation type data; operation types include explicit feedback and implicit feedback.

[0108] The weight decay unit reduces the weight of edge weights that have not been triggered within a preset period according to a preset decay rate.

[0109] The behavior log analysis unit uses an event classification model to distinguish between explicit and implicit feedback. Explicit feedback includes user-initiated ratings or marking actions, while implicit feedback includes dwell time and click frequency. The weight decay unit sets the edge weight decay period to 30 days. If the edge weight is not triggered within the period, it is linearly reduced at a decay rate of 0.5% per day. The calculation of incremental values ​​is based on dynamic adjustment of the operation type coefficient. The explicit feedback coefficient is configured to be 0.8, and the implicit feedback coefficient is configured to be 0.3. The behavior frequency weight is adjusted according to the legal validity status in the timeliness marking data. The weight multiplier for valid status is 1.0, and the multiplier for partially invalid status is 0.6.

[0110] Specifically, when a user favorites an academic paper node in the recommendation list, the behavior log analysis unit identifies this action as explicit feedback, parses the action timestamp and node ID to generate an action record; the weight decay unit detects that an edge between a related regulation and a case has not been accessed by the user within 30 days, and reduces the original weight from 0.75 to 0.63; the feedback closed-loop linkage module calculates the incremental value ΔW = 0.8 × 1.0 × 1.2 = 0.96 based on the explicit feedback coefficient of 0.8, the current regulation validity multiplier of 1.0, and the single operation frequency weight of 1.2, and updates the weight of the corresponding edge from 1.20 to 2.16; the synchronous trigger timeliness update module verifies the revision status of the related regulation. If it detects that the regulation revision has caused the case reference status to be downgraded, it automatically reduces the hub ranking of the relevant academic paper node and updates the risk warning mark in the user interface through the dynamic recommendation engine.

[0111] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0112] The dynamic recommendation and feedback engine includes a feedback closed-loop linkage module, which performs the following data processing steps:

[0113] First, the system receives data on secondary user actions related to the recommendation list. This includes actions such as clicking, saving, sharing, or commenting on recommended legal provisions, case studies, or academic papers. The system then analyzes this secondary action data and generates real-time weight adjustment instructions.

[0114] Secondly, the incremental values ​​of the knowledge graph edge weights are calculated based on real-time weight adjustment instructions and timely tagging data. For example, for a recommended item clicked by a user, the corresponding knowledge graph edge weight increases by 0.1; for a recommended item saved by a user, the corresponding edge weight increases by 0.2. Considering the timeliness factor, the weight increment for recently revised regulations will be increased accordingly. The system then updates the weights of associated edges and the ranking of hub nodes.

[0115] Secondly, user feedback data is synchronized to the timeliness update module, triggering the monitoring of revisions to relevant regulations and the verification of the timeliness of academic papers. For example, when a user raises a question about a regulation, the system will automatically check the latest revision status of that regulation.

[0116] The feedback loop module also includes a behavior log analysis unit and a weight decay unit. The behavior log analysis unit parses secondary operation data to obtain user behavior time-series data and operation type data. Operation types include explicit feedback (such as ratings and comments) and implicit feedback (such as browsing duration and click frequency). The weight decay unit reduces the weight of edges that have not been triggered within a preset period according to a preset decay rate. For example, for knowledge graph edges that have not been interacted with by users for 30 days, their weight decreases by 1% per day.

[0117] Through the above technical solutions, this application achieves dynamic updating and personalized recommendations for knowledge graphs. The system can adjust the structure of the knowledge graph based on real-time user feedback, improving the accuracy and timeliness of recommendations. Simultaneously, through a feedback loop mechanism, the system can promptly capture regulatory revisions and updates to academic viewpoints, ensuring the timeliness and accuracy of teaching content. This dynamic adjustment mechanism effectively solves the problem of outdated teaching content caused by frequent regulatory revisions in economic law teaching, improving the quality and effectiveness of case-based teaching.

[0118] In some of the solutions described above in this application, when the dynamic recommendation and feedback engine adjusts the edge weights of the knowledge graph based on user feedback, the incremental calculation of the edge weights does not take into account the cumulative impact of differences in user operation types and behavior frequency on timeliness. This results in insufficient matching between the weight adjustment and the dynamic changes in timeliness, affecting the effectiveness of the recommendation results.

[0119] This application further proposes the following formula for calculating the incremental value:

[0120] ΔW = Operation type coefficient × Time-delay coefficient × Behavior frequency weight, where the operation type coefficient is configured according to the operation type, and the behavior frequency weight is configured according to the time-delay marker data;

[0121] The formula for calculating the weight of an associated edge is W. new =W old +ΔW, where W new W represents the updated weight of the associated edges. old The weights of the original associated edges;

[0122] The formula for calculating the importance of a hub node is: Node Importance = ∑(Associated Edge Weight × Time-Based Decay Coefficient).

[0123] The operation type coefficient is pre-configured with different values ​​to differentiate the contribution of explicit and implicit feedback to weight adjustments, with explicit feedback corresponding to a higher coefficient value. The timeliness decay coefficient is dynamically calculated based on the interval between the effective date of the regulatory revision and the current time; the longer the interval, the smaller the coefficient. The behavior frequency weight is exponentially weighted based on the number of times a user operates on the same entity within a preset period. The edge weight update process adds ΔW to the original weight, allowing high-frequency operations to generate a higher cumulative increment within the validity period. The node importance calculation introduces a timeliness coefficient to ensure that the edge weights associated with the revised regulations are downweighted in the importance assessment.

[0124] Specifically, when a user provides explicit feedback to a regulatory entity in the recommendation list, the action type coefficient is set to 0.8, while implicit feedback corresponds to a coefficient of 0.3. The timeliness decay coefficient is calculated based on the difference between the effective date of the regulation and the current time, at a decay rate of 5% per month. If the regulation revision took effect 6 months ago, the coefficient is 0.7. The behavior frequency weight is accumulated based on the number of times the user clicked on the node in the past 7 days, with a decay factor of 0.9. For example, 3 clicks correspond to a weight of 1 + 0.9 + 0.81 = 2.71. The edge weight increment ΔW is obtained by multiplying the three factors. For example, explicit feedback corresponds to ΔW = 0.8 × 0.7 × 2.71 ≈ 1.52, and this increment is added to the original edge weight. When calculating node importance, the updated weights of the associated edges are multiplied by the timeliness coefficients and then summed. For example, if a node is associated with three edges with weights of 2.5, 3.0, and 1.8, and corresponding timeliness coefficients of 0.7, 1.0, and 0.5, then its importance is (2.5 × 0.7) + (3.0 × 1.0) + (1.8 × 0.5) = 4.45. This mechanism integrates user behavior intensity, operational intent, and regulatory timeliness status during weight adjustments, ensuring that hub nodes in the recommendation list always reflect the latest and most valid knowledge relationships.

[0125] In practice, the operation type coefficient can be set as follows:

[0126] Explicit likes: 1.5; explicit saves: 1.2; implicit views: 0.8.

[0127] The time-related decay coefficient can be set as follows: 1.0 within 1 year, 0.8 for 1 to 3 years, and 0.6 for more than 3 years.

[0128] The frequency weight of behavior can be set as follows: 0.5 for a single occurrence, 1.0 for 2-5 occurrences, and 1.5 for more than 5 occurrences.

[0129] For example, a user explicitly favorited a case study that occurred 2 years ago, and the user recently viewed the case study 3 times.

[0130] The increment value ΔW = 1.2 × 0.8 × 1.0 = 0.96.

[0131] Assuming the original weight W old If the weight is 5, then the updated weight W new =5 + 0.96 = 5.96.

[0132] For a hub node, assuming it is associated with 3 edges with weights of 5.96, 4.5, and 3.2, and timeliness coefficients of 1.0, 0.8, and 0.6, the importance of this node is:

[0133] 5.96×1.0+4.5×0.8+3.2×0.6=10.68.

[0134] Through the above technical solution, this application achieves dynamic adjustment of edge weights and node importance in a knowledge graph. The calculation of edge weights comprehensively considers user operation type, timeliness, and behavior frequency, enabling the knowledge graph to more accurately reflect the strength of relationships between entities. The calculation of hub node importance further integrates multi-dimensional information, which helps to identify and recommend key knowledge points. This dynamic adjustment mechanism improves the timeliness and personalization of the knowledge graph, thereby enhancing the accuracy of case reasoning and the relevance of teaching recommendations.

[0135] In some of the solutions described above in this application, the knowledge graph-based economic law case reasoning teaching platform needs to perform structured parsing of legal provisions, case fact descriptions, and thesis abstracts. However, existing technologies have the following problems: they lack efficient and accurate methods for extracting the responsible parties, constituent elements, and legal consequences from the textual data of legal provisions; they struggle to accurately extract the subject, behavior, and outcome data from the textual data of case fact descriptions; and they cannot effectively generate supporting relationship data between theoretical viewpoints and legal provisions from the textual data of thesis abstracts, resulting in insufficient accuracy in knowledge graph construction.

[0136] This application further proposes specific methods for structured parsing of raw text data, including:

[0137] Specifically, for the text data of legal provisions, data on the responsible parties, constituent elements, and legal consequences are extracted using preset regular expressions. The regular expressions are pre-designed based on the sentence structure characteristics of the legal text, for example, triggering matching with keywords such as "shall" and "shall not," and the matched text fragments are stored according to preset fields.

[0138] For textual data describing case facts, an event extraction model is used to extract subject, behavior, and outcome data. The event extraction model is trained using a bidirectional long short-term memory network. The input layer receives the segmented case text, and the output layer generates a label sequence of subject-behavior-outcome.

[0139] For the text data of the paper abstract, keyword co-occurrence analysis is used to generate supporting relationship data between theoretical viewpoints and legal provisions. Keyword co-occurrence analysis is based on word frequency statistics and association rule mining, extracting sentence fragments that simultaneously contain legal provision numbers and academic terms as supporting evidence.

[0140] Specifically, in the process of parsing legal provisions, regular expressions accurately locate the responsible parties (such as "legal persons" and "natural persons"), constituent elements (such as "subjective intent" and "gross negligence"), and legal consequences (such as "compensation amount" and "administrative penalty") through pattern matching, avoiding semantic fragmentation caused by traditional word segmentation techniques. In case fact parsing, bidirectional long short-term memory networks accurately identify the subjects (such as plaintiffs and defendants), behaviors (such as breach of contract and unfair competition), and results (such as judgments and compensation amounts) in cases by capturing the dependencies between text sequences, solving the problem of poor adaptability of traditional rule-based methods to complex sentence structures. In paper abstract parsing, keyword co-occurrence analysis automatically generates the supporting relationship between theoretical viewpoints and legal provisions by statistically analyzing the co-occurrence frequency of legal provision numbers (such as Article 52 of the Contract Law) and academic concepts (such as "defects in contract validity"), overcoming the inefficiency of manual annotation. The collaboration of these three technologies achieves deep structuring of multi-source data, providing a high-precision data foundation for knowledge graph construction.

[0141] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0142] For the textual data of legal provisions, data on the responsible parties, constituent elements, and legal consequences are extracted using preset regular expressions. For example, the regular expression "(\w+)should(\w+), otherwise(\w+)" is used to match legal provisions and extract the responsible parties, behavioral obligations, and illegal consequences.

[0143] For textual data describing the facts of a case study, an event extraction model is used to extract the subject, behavior, and outcome data. The event extraction model is trained using a BiLSTM network. Specifically, the case text is first segmented and tagged with parts of speech, then input into the BiLSTM network. The network output layer uses CRF for sequence labeling to identify the subject, behavior, and outcome elements of the event.

[0144] For the text data of the paper abstract, keyword co-occurrence analysis is used to generate supporting relationship data between theoretical viewpoints and legal provisions. Specific steps include: extracting keywords from the abstract; calculating the co-occurrence frequency of keywords and legal provisions; calculating the association strength based on the co-occurrence frequency; and generating a supporting relationship diagram between keywords and legal provisions.

[0145] Through the aforementioned technical solution, this application achieves structured parsing of different types of text data, extracting core elements and their relationships from legal provisions, case facts, and academic viewpoints. This multi-dimensional text parsing method improves the accuracy and completeness of knowledge graph construction, laying the foundation for subsequent case reasoning and knowledge recommendation. Simultaneously, the use of automated parsing technology reduces the workload of manual annotation, improving the efficiency of knowledge graph construction.

[0146] In some of the solutions mentioned above in this application, the dimensional division of entity feature vectors fails to accurately reflect the differences between different types of entities in teaching scenarios, resulting in insufficient accuracy of knowledge graph node representation, which affects the matching efficiency of the case reasoning module and the personalized adaptation capability of the recommendation engine.

[0147] This application further proposes that the entity feature vector includes the following dimensions: the feature vector of a legal entity consists of the frequency of citation of clauses, the timeliness weight, and the number of related academic papers; the feature vector of a case entity consists of the range of disputed amount, the court level coefficient, and the legal provision matching degree; and the feature vector of a user entity consists of research direction tags, knowledge blind spot tags, and the frequency of historical operations.

[0148] Specifically, the frequency of clause citations is dynamically calculated by counting the number of times a legal entity is cited by a case entity in a statistical knowledge graph; the statute of limitations weight is generated based on the effective status marking of the legal entity in the statute of limitations update module, and the weight decreases by a preset ratio when the legal entity is marked as partially invalid; the number of associated academic papers is determined by the keyword co-occurrence analysis results of the paper abstracts in the semantic parsing module. The disputed amount range uses a logarithmic function to map the disputed amount involved in the case to a discrete range value; the court level coefficient sets gradient weights based on the court level corresponding to the case entity; the legal provision matching degree is calculated by the textual similarity between the case fact description and the legal provisions. Research direction tags are generated based on high-frequency entity types in the user's historical operation data; knowledge blind spot tags are identified by the distribution of entity types that the user has not triggered operations on; historical operation frequency records the frequency of the user's access to different entity nodes.

[0149] Specifically, the frequency of citations of legal provisions reflects their practical application value in case reasoning, while the timeliness weight ensures the effectiveness of the legal provisions in the knowledge graph, and the number of associated academic papers reflects the strength of theoretical support. The disputed amount range and the court level coefficient of a case entity together constitute a quantitative indicator of case influence, and the legal provision matching degree accurately characterizes the strength of the association between the case and the legal provisions. The research direction tags of user entities drive the targeted retrieval of the recommendation engine, the knowledge blind spot tags are used to identify teaching blind spots, and the frequency of historical operations establishes a user preference model. Through the collaborative calculation of multi-dimensional feature vectors, the knowledge graph construction module can accurately distinguish the inherent attribute differences of entity types, the case reasoning module automatically weights the core dimensions when performing similarity matching, and the dynamic recommendation engine achieves accurate personalized recommendations based on user feature vectors. For example, when a user's research direction tag shows that they are interested in the field of antitrust law, the recommendation engine prioritizes screening case entities with disputed amount ranges higher than a set threshold and matches legal entities with timeliness weights greater than a preset value.

[0150] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0151] Entity feature vectors include the following dimensions:

[0152] The feature vector of a regulatory entity consists of the frequency of clause citations, the timeliness weight, and the number of associated academic papers. The frequency of clause citations is obtained by counting the number of case entities in the knowledge graph that cite the clause. The timeliness weight is calculated based on historical data of regulatory revisions, with the latest revision having a timeliness weight of 1, decreasing by 0.1 for each subsequent revision. The number of associated academic papers is obtained by retrieving the number of academic entities in the knowledge graph that have citation relationships with the regulatory entity.

[0153] The feature vector of a case entity consists of the disputed amount range, the court level coefficient, and the legal provision matching degree. The disputed amount range is divided into 5 levels based on the specific monetary values ​​involved in the case facts. The court level coefficient starts at 1.0 for the Supreme People's Court, decreasing by 0.1 for each subsequent level. The legal provision matching degree is obtained by calculating the semantic similarity between the legal provisions cited in the case and the facts of the case.

[0154] The feature vector of a user entity consists of research direction tags, knowledge blind spot tags, and historical operation frequency. Research direction tags are generated based on clustering of legal and case entities recently viewed by the user. Knowledge blind spot tags are generated by analyzing high-frequency keywords that the user has not viewed. Historical operation frequency records the number of times the user interacts with various entities, such as clicking and saving.

[0155] Through the above technical solutions, this application achieves multi-dimensional feature representations of legal entities, case entities, and user entities. The feature vectors of legal entities reflect the application frequency, timeliness, and academic attention of the provisions. The feature vectors of case entities reflect the economic impact, judicial authority, and accuracy of legal application of the case. The feature vectors of user entities characterize learners' research interests, knowledge gaps, and usage habits. This multi-dimensional feature representation provides rich semantic information for subsequent case reasoning and personalized recommendations, helping to improve the retrieval accuracy and recommendation relevance of the knowledge graph.

[0156] In some of the solutions described above in this application, the validity update module updates the validity markers of legal entities in the knowledge graph by monitoring the status data of legal revisions. However, the update of the reference status of related case entities after the revision of regulations lacks an automated matching mechanism, which may cause case entities to continue to cite expired regulations, resulting in the risk of misleading teaching.

[0157] like Figure 4 The diagram shown is a workflow diagram of the timeliness update module; this application further proposes that the execution logic of the timeliness update module includes:

[0158] Obtain historical data on regulatory revisions and extract timestamp fields and clause change type data from the revised texts;

[0159] Generate effective status markers for regulatory entities and calculate reference effectiveness attenuation coefficients for related cases based on clause change type data;

[0160] When revision status data of a regulation text is detected, the case entities that reference that regulation are traversed in the knowledge graph;

[0161] Based on the degree of match between the revised content and the facts of the case, the reference status of the case entity is updated to "valid reference" or "partially invalid reference".

[0162] The timestamp field is parsed using the ISO 8601 standard format to extract the effective date and repeal date fields, generating the legal entity's validity period marker. Clause change type data is divided into three categories: repeal, partial amendment, and addition / supplementation. The reference validity attenuation coefficient for related cases is set with decreasing weights according to the change type: 0 for repeal, 0.5 for partial amendment, and 0.2 for addition / supplementation. When traversing related case entities, the affected case set is filtered out using the legal provision matching degree field in the case entity feature vector. The matching degree calculation uses a text similarity algorithm, comparing the semantic similarity between the legal application portion of the case's factual description and the content of the legal amendment. When the similarity exceeds a threshold, the case's reference status is marked as partially invalid.

[0163] Specifically, when Article 102 of an economic law is detected to have been repealed, the timeliness update module extracts the repeal timestamp from the law's revision record and updates the law's entity validity to an invalid state. It then traverses case entities in the knowledge graph that reference the law, extracting specific application scenario text related to Article 102 from the case's factual description. The semantic similarity between the case text and the repealed clause is calculated using a BERT model. When the similarity exceeds 0.85, the reference state of the case entity is marked as partially invalid, and the weight of its associated edges in the knowledge graph is reduced to 30% of the original value. For partially revised legal provisions, such as the addition of a third paragraph to Article 15 of an anti-monopoly law implementation rule, the system extracts the effective date of the new clause, calculates the reference validity decay coefficient of the associated case as 0.2, multiplies the legal provision matching degree field value in the case entity's feature vector by 0.8, and adds an application hint for the new clause to the reasoning result data. Through this mechanism, the dynamic association and update of the revised content of the law and the case entity status are achieved, avoiding the use of invalid legal basis in teaching.

[0164] As a preferred embodiment, the solution of the present application is specifically implemented as follows: the execution logic of the aging update module is realized through the following data processing steps: after acquiring regulatory revision history data, natural language processing technology is adopted to extract the timestamp field in the revised text, wherein the extraction rule of the timestamp field is based on the ISO 8601 standard, matching the format of "effective from YYYY MM DD" in the text; for the parsing of clause change type data, a classification model is used to identify three types of labels of addition, modification and repeal in the revised content, wherein the classification model is obtained by fine-tuning based on the BERT pre-trained model; generating an effective status mark of the regulatory entity according to the effective timestamp, wherein the effective status mark includes three discrete statuses of "effective", "to be effective" and "repealed"; the reference effectiveness attenuation coefficient of the associated case is calculated by a linear attenuation function, wherein the attenuation base is the original matching degree, and the attenuation factor is the weight value corresponding to the clause change type; when a regulatory revision is detected, traverse the case entities that cite the regulation in the knowledge graph, wherein the traversal process uses the adjacency node index of the graph database to accelerate the query; in the reference state update logic of the case entity, the text similarity algorithm is adopted for the matching degree calculation, wherein when the similarity between the case fact description and the revised clause is lower than 0.6, it is marked as "partially invalid reference", and when the similarity is higher than 0.8, it is marked as "valid reference".

[0165] Through the above technical solutions, the present application solves the problem of update lag caused by the existing platforms relying on manual labeling of aging status, which is specifically reflected in: automatically extracting the timestamp and change type in the regulatory revised text, dynamically generating the regulatory effectiveness mark and the case reference status, so as to avoid teaching misleading caused by the repeal or revision of regulations. Further, by calculating the reference effectiveness attenuation coefficient of associated cases, the aging correlation strength between cases and regulations is quantified, so that the teaching platform can automatically screen valid cases and improve the consistency between cases and current regulations.

[0166] In some of the above solutions of the present application, when the aging update module performs regulatory revision monitoring and case reference status update, if the reference status of the case entity is marked as a partially invalid reference, the prior art lacks a mechanism for automatically triggering full-text scanning of associated academic papers and regulatory revision monitoring, resulting in a lag risk in the timeliness collaborative update of academic theory and legal practice.

[0167] The present application further proposes a feedback linkage unit in the dynamic recommendation and feedback engine, which automatically triggers full-text scanning of associated academic papers and regulatory revision monitoring when it is detected that the reference status of a case entity is marked as a partially invalid reference.

[0168] The feedback linkage unit comprises an academic paper full-text scanning module and a regulatory revision monitoring module. The academic paper full-text scanning module extracts regulatory clauses cited in the paper using a pre-set keyword matching algorithm and verifies their relevance with regulatory entities in the knowledge graph. If a revised regulatory clause is detected, a timeliness marker is generated. The regulatory revision monitoring module periodically crawls revision announcements from official databases, parses substitution relationship data in the revised text, and updates the validity markers of relevant regulations in the knowledge graph. The feedback linkage unit and the timeliness update module communicate bidirectionally via a data interface. When the case reference status changes, it triggers timeliness verification of the associated academic paper and simultaneously sends the verification result back to the knowledge graph construction module to update the relationships between entities.

[0169] Specifically, when a case entity is marked as partially invalidated due to regulatory revisions, the feedback linkage unit first extracts a list of academic papers associated with that case entity and uses a full-text scanning module to compare each legal clause cited in the papers. If it is found that a clause cited in the paper has been repealed or replaced, the paper's expiration status is marked as needing updating, and this mark is synchronized to the academic entity node in the knowledge graph. Simultaneously, the regulatory revision monitoring module continuously monitors the revision dynamics of related regulations. For example, when the substitution relationship data for a certain regulatory clause is updated, the weight coefficient of the corresponding edge in the knowledge graph is automatically adjusted. During this process, the scanning module uses a keyword matching threshold set at 85% similarity to ensure that some differences in expression do not affect the identification of core clauses. Through this mechanism, the system can maintain consistency between legal practice and academic theory in real time, avoiding recommendation errors caused by time lags.

[0170] As a preferred embodiment, the solution of this application is implemented as follows: When the reference status of a case entity is marked as a partially invalid reference, the feedback linkage unit immediately initiates a full-text scanning process of the associated academic paper. Specifically, by calling a natural language processing interface, semantic analysis is performed on the full-text content of the associated paper to detect whether there are paragraphs referencing revised regulations, and the detection results are written into the timeliness verification queue. At the same time, the regulation revision monitoring thread is activated, sending a real-time monitoring request to the preset regulation database subscription interface, with the monitoring period set to 24 hours. If new revision data of the associated regulations is captured within the monitoring period, the parsing function of the timeliness update module is automatically called to generate timeliness-marked data to update the corresponding node attributes in the knowledge graph.

[0171] Through the aforementioned technical solution, this application effectively addresses the issue of delayed synchronization of related academic resources when case reference status becomes invalid. By establishing an automated triggering mechanism, it achieves real-time linkage between verifying the validity of academic paper citations and tracking regulatory revisions, avoiding a decrease in the credibility of recommended content due to delays caused by manual maintenance. This technical approach enables the relationship between academic theory and legal practice in the knowledge graph to dynamically adapt to changes in legal provisions, ensuring the timeliness and accuracy of theoretical support during teaching reasoning.

[0172] In some of the solutions mentioned above in this application, a method for case entity matching based on knowledge graphs was proposed. However, in this process, the existing methods lack a mechanism to verify the timeliness of the matching results and do not generate a structured reasoning report, which may lead to the use of outdated legal provisions or the neglect of theoretical support during the teaching process.

[0173] like Figure 5 The diagram shows the workflow of the case reasoning module. This application further proposes a method for retrieving similar case entities based on a knowledge graph, generating reasoning result data containing the applicable legal path. Specifically, this includes: converting the case facts input by the user into feature vectors and performing cosine similarity matching with case entities in the knowledge graph; extracting the associated applicable legal path and academic viewpoint data for case entities with a matching degree higher than a preset threshold; and generating a reasoning report containing similar case comparisons, timeliness risk warnings, and recommended learning paths.

[0174] Specifically, when user-input case facts are transformed into feature vectors, the same vectorization rules as those used for case entities in the knowledge graph are adopted to ensure consistency of the vector space; during cosine similarity matching, a dynamic threshold is set, and the matching accuracy is dynamically adjusted based on the timeliness markers of case entities in the knowledge graph; when extracting the applicable path of regulations, reference status data in the timeliness update module is called synchronously to identify the validity markers of related regulations; when generating the reasoning report, case comparison analysis, regulatory revision risk labeling, and recommended learning paths are integrated to form a structured output document.

[0175] Specifically, the case facts input by the user are transformed into feature vectors composed of dimensions such as the disputed amount range and the degree of legal provision matching through a semantic parsing module. In the knowledge graph, the vector distance with all case entities is calculated using a cosine similarity algorithm, and a set of cases with similarity higher than a preset threshold is selected. For matched cases, the associated legal and academic entities are extracted along the edge weights of the knowledge graph, forming a complete legal application path chain. Simultaneously, the effective status markers of associated regulations are obtained through a timeliness update module; if a repeal or replacement relationship is detected, the timeliness risk level is marked in the reasoning report. The recommended learning path is based on the entity nodes currently viewed by the user in a dynamic recommendation engine, prioritizing the display of learning sequences that simultaneously contain valid regulations, typical cases, and authoritative papers. Thus, this method improves the accuracy and educational value of case reasoning results through multi-dimensional data fusion and a timeliness linkage mechanism.

[0176] As a preferred embodiment, the solution of this application is implemented as follows: The case facts input by the user are preprocessed and transformed into a 512-dimensional feature vector. The textual data of the case facts is used to extract semantic features in three dimensions: subject, behavior, and result, through a BiLSTM event extraction model. This feature vector is input into a knowledge graph matching engine, and cosine similarity is calculated with the case entity feature vectors stored in the graph, with a matching threshold set to 0.85. For successfully matched case entities, their associated three-layer legal application paths are extracted. The primary path is the directly cited legal provisions, the secondary path is judicial interpretations, and the tertiary path is the theoretical support in associated academic papers. The system automatically generates a structured reasoning report, which includes a similar case comparison matrix, legal provision timeliness verification results, and a supplementary learning material index. Timeliness verification is achieved by calling the effective status marker of the legal entity.

[0177] Through the aforementioned technical solution, this application effectively addresses the risks of invalid legal citations and missing cross-source data connections in economic law case teaching. Specifically, at the implementation level, an automated feature vector matching mechanism achieves precise mapping between case facts and knowledge graphs, avoiding subjective biases from manual retrieval; multi-layered extraction of the applicable legal path ensures three-dimensional verification of legal provisions, judicial interpretations, and academic theories; and real-time access to time-sensitive data automatically marks the citation risks of repealed clauses in the reasoning report, eliminating teaching errors caused by outdated regulations. This solution elevates the accuracy of the legal basis in the case reasoning process to the latest state of judicial practice, while establishing a complete cross-data source verification chain.

[0178] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A knowledge graph-based case-based reasoning teaching platform for economic law, characterized by: include: A multi-source data acquisition module is used to acquire raw text data, which includes legal provisions, case fact descriptions, and paper abstracts. The semantic parsing module is used to perform structured parsing on the original text data, extract legal entities, case entities, and academic entities, and generate reference relationship data between entities; The knowledge graph construction module is used to construct an economic law knowledge graph based on the legal entities, case entities, academic entities and the reference relationships between entities; the nodes in the knowledge graph store the entity feature vectors corresponding to the legal entities, case entities, and academic entities, and the edges store the semantic relationships and weights of preset types; The case reasoning module is used to receive case facts input by the user, retrieve similar case entities based on the knowledge graph, and generate reasoning result data containing the applicable legal path; The timeliness update module is used to monitor the revision status data of legal provisions, parse the effective time, repeal time and substitution relationship data in the revised legal text, generate timeliness mark data and update the validity mark of legal entities and the reference status of related cases in the knowledge graph. The dynamic recommendation and feedback engine is used to generate personalized recommendation lists based on the entity nodes currently viewed by the user and the time-sensitive data. The dynamic recommendation and feedback engine includes a feedback closed-loop linkage module that performs the following data processing steps: Receive secondary operation data from users on the recommendation list, parse the secondary operation data and generate real-time weight adjustment instructions; The incremental values ​​of the knowledge graph edge weights are calculated based on the real-time weight adjustment instructions and the time-sensitive tag data, and the weights of the associated edges and the ranking of the hub nodes are updated. The system synchronizes user feedback data, including user actions on personalized recommendation lists, to the timeliness update module, triggering monitoring of revisions to relevant regulations and verification of the timeliness of academic papers.

2. The knowledge graph-based economic law case reasoning teaching platform according to claim 1, characterized in that: The process of generating a personalized recommendation list based on the entity nodes and time-sensitive data currently being viewed by the user includes: Based on the entity node currently being viewed by the user, the path is extended along the edges in the knowledge graph whose weights are higher than a preset threshold. Hub nodes that connect legal entities, case entities, and academic entities are given priority. The criteria for determining a hub node are that the number of associated edges exceeds a preset number and the validity period is valid. The probability of negative feedback risk is pre-calculated for entities in the recommendation list. If the probability exceeds the preset risk threshold, the entity is replaced with a backup entity.

3. The knowledge graph-based economic law case reasoning teaching platform according to claim 2, characterized in that: The formula for calculating the probability of negative feedback risk is: P risk =αR hist +βC time +γW rel +δB recent , where R hist For the historical negative feedback rate, C time W is the time-degradation coefficient. rel B is the correlation weight decay value. recent The score represents the user's recent behavior score; and α, β, γ, and δ represent the historical negative feedback rate, the time-related decay coefficient, the correlation weight decay value, and the preset weight coefficient of the user's recent behavior score, respectively. The historical negative feedback rate R hist The calculation formula is: The aging attenuation coefficient C time The calculation formula is: The associated weight decay value Among them W edge W represents the edge weight of the user's current entity. max This represents the maximum weight of edges of the same type in the graph; The user's recent behavior score 4. The knowledge graph-based economic law case reasoning teaching platform according to claim 3, characterized in that: The feedback closed-loop linkage module includes: The behavior log analysis unit is used to parse secondary operation data and obtain user behavior time sequence data and operation type data; the operation type includes explicit feedback and implicit feedback. The weight decay unit is used to reduce the weight of edge weights that have not been triggered within a preset period according to a preset decay rate.

5. The knowledge graph-based economic law case reasoning teaching platform according to claim 4, characterized in that: The formula for calculating the incremental value is: ΔW = Operation type coefficient × Time-decrease coefficient × Behavior frequency weight, wherein the operation type coefficient is configured according to the operation type, and the behavior frequency weight is configured according to the time-decrease marker data; The formula for calculating the weight of the associated edge is: W new =W old +ΔW, where W new W represents the updated weight of the associated edges. old The weights of the original associated edges; The formula for calculating the importance of the hub node is as follows: Node importance = ∑(associated edge weight × time decay coefficient).

6. The knowledge graph-based economic law case reasoning teaching platform according to claim 1, characterized in that: The structured parsing of the original text data includes: The text data of legal provisions is extracted using preset regular expressions to identify the responsible parties, constituent elements, and legal consequences. For textual data describing case facts, subject, behavior, and result data are extracted using an event extraction model trained with a BiLSTM network. From the text data of the paper abstract, keyword co-occurrence analysis is used to generate supporting relationship data between theoretical viewpoints and legal provisions.

7. The knowledge graph-based economic law case reasoning teaching platform according to claim 1, characterized in that: The entity feature vector includes the following dimensions: The feature vector of a regulatory entity consists of the frequency of citation of clauses, the weight of timeliness, and the number of related academic papers. The feature vector of a case entity consists of the range of disputed amount, the coefficient of court level of appeal, and the degree of matching of legal provisions. The feature vector of a user entity consists of research direction labels, knowledge blind spot labels, and historical operation frequency.

8. The knowledge graph-based economic law case reasoning teaching platform according to claim 1, characterized in that: The execution logic of the timeliness update module includes: Obtain historical data on regulatory revisions, and extract timestamp fields and clause change type data from the revised texts; The effective status marker of the regulatory entity is generated based on the timestamp field, and the reference effectiveness decay coefficient of the related cases is calculated based on the clause change type data. When revision status data of a regulation text is detected, the case entities that reference that regulation are traversed in the knowledge graph; Based on the degree of match between the revised content and the facts of the case, the reference status of the case entity is updated to "valid reference" or "partially invalid reference".

9. The knowledge graph-based economic law case reasoning teaching platform according to claim 8, characterized in that: The dynamic recommendation and feedback engine also includes a feedback linkage unit, which automatically triggers a full-text scan of associated academic papers and monitoring of regulatory revisions when the reference status of a case entity is detected as "partially invalid reference".

10. The knowledge graph-based economic law case reasoning teaching platform according to claim 1, characterized in that: The step of retrieving similar case entities based on the knowledge graph and generating reasoning result data containing the applicable legal paths includes: The user-input case facts are transformed into feature vectors and matched with case entities in the knowledge graph using cosine similarity. For case entities with a matching degree higher than a preset threshold, extract their associated legal application paths and academic viewpoint data; Generate a reasoning report that includes comparisons of similar cases, timeliness risk warnings, and recommended learning paths.

Images