Malicious litigation intelligent supervision method, system, device and storage medium

Abstract

The application provides a malicious lawsuit intelligent supervision method and system based on a dynamic legal knowledge graph, equipment and a storage medium, and belongs to the technical field of legal artificial intelligence. The method obtains dynamic judicial data, identifies the data type and sets a decay coefficient, calculates the time limit weight by using an exponential decay model, and generates a dynamic data vector. A knowledge graph is constructed based on the dynamic data vector, the relationship weight between entities is updated in real time, and the behavior abnormality intensity value is calculated by using an information entropy model. The matching degree of the judicial data and a legal case library is calculated in parallel as a compliance index value. When the compliance index is lower than a first threshold value or the behavior abnormality intensity value is higher than a second threshold value, a supervision process is started. According to the supervision result, the rule library is dynamically updated in combination with a legal compliance loss function, intelligent and dynamic supervision of malicious lawsuits is realized, and the judicial efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of legal artificial intelligence technology, specifically to a method for intelligent supervision of malicious litigation based on dynamic legal knowledge graphs, a system for intelligent supervision of malicious litigation based on dynamic legal knowledge graphs, computer equipment, and computer-readable storage media. Background Technology

[0002] With the continuous development of judicial practice, malicious litigation has become increasingly complex and covert, causing serious waste of judicial resources and damaging litigation order and the legitimate rights and interests of the parties involved. To effectively supervise and prevent malicious litigation, the legal technology field has developed various solutions based on knowledge graphs and intelligent analysis.

[0003] Currently, legal knowledge graph technology has been widely applied in fields such as legal consultation, case retrieval, and litigation analysis. For example, CN119514657A discloses a legal consultation report generation system based on legal knowledge graphs. This system collects data from legal databases, user behavior records, and interdisciplinary data sources to construct a legal knowledge graph, providing users with personalized legal consultation services.

[0004] Regarding litigation strategies and risk prediction, CN120047003A introduces a method for generating litigation strategies and predicting the success rate based on multimodal data fusion and intelligent reasoning. This method analyzes case information and evidence through natural language processing, OCR technology and image recognition, and constructs a dynamic knowledge graph by combining historical judgment databases to achieve early warning of the risk of losing the case and recommendation of litigation requests.

[0005] In addition, CN119046476A proposes a legal element analysis method and system based on a large language model and knowledge graph. This system extracts information through a large language model, generates a knowledge graph containing entity information and relational information, and performs legal element analysis on this basis.

[0006] However, existing technologies still have the following shortcomings:

[0007] First, existing legal knowledge graphs are mostly built on static rule bases and lack dynamic evidence weight decay mechanisms. Legal data has a significant time-sensitive characteristic; different types of judicial data differ in their timeliness, and outdated data may unduly influence decision-making, reducing the accuracy and timeliness of oversight.

[0008] Secondly, there is a lag in updating legal rules. Existing systems largely rely on pre-set static rule bases for judgment, making it impossible to respond promptly to new malicious litigation patterns and the ever-changing legal environment. When new malicious litigation behaviors emerge, the system lacks adaptive learning capabilities and cannot adjust its monitoring strategies in real time, resulting in a significant reduction in monitoring effectiveness.

[0009] Furthermore, existing technologies lack effective mechanisms for detecting abnormal behavior. Malicious litigation often manifests as a sudden change in the behavior patterns of the litigants, but existing systems mostly focus on static feature matching and lack the ability to dynamically monitor behavioral changes, making it difficult to detect hidden malicious litigation behavior in a timely manner.

[0010] Therefore, there is an urgent need for an intelligent supervision method for malicious litigation that can dynamically handle the timeliness of legal data, update the rule base in real time, and effectively detect abnormal behavior, so as to improve the ability to identify and prevent malicious litigation behavior. Summary of the Invention

[0011] In view of this, in order to solve the problem of rule lag and the technical problem of existing solutions relying on static rule bases, the purpose of this invention is to propose a malicious litigation intelligent supervision method based on dynamic legal knowledge graph, which effectively solves the problem of quantifying the timeliness of judicial evidence and improves the accuracy of malicious litigation identification.

[0012] To achieve the above objectives, in a first aspect, the present invention provides a method for intelligent monitoring of malicious litigation based on a dynamic legal knowledge graph, comprising the following steps:

[0013] The system acquires dynamic judicial data and identifies the type of judicial data. Based on the data type, it sets a corresponding attenuation coefficient and automatically calculates the timeliness weight of the judicial data using an exponential attenuation model. The timeliness weight is then marked on the judicial data to generate a dynamic data vector. The timeliness weight attenuates as the data timeliness interval increases.

[0014] A knowledge graph is constructed based on the dynamic data vectors, and the weight values ​​of the relationships between entities in the graph are updated in real time. The intensity value of behavioral anomalies is calculated using the information entropy model. An early warning is triggered when the intensity value of behavioral anomalies exceeds a preset threshold.

[0015] The system calculates the matching degree between the case characteristics of the judicial data and the legal case database in parallel as a compliance indicator value, and simultaneously receives the abnormal behavior intensity value; when the compliance indicator value is lower than the first threshold or the abnormal behavior intensity value is higher than the second threshold, the supervision process is initiated.

[0016] Based on the monitoring results, the rule base is dynamically updated using a legal compliance loss function, with the update direction constrained by the judicial interpretation vector.

[0017] Preferably, the generation of the dynamic data vector further includes the following steps:

[0018] The system analyzes the input judicial data and classifies it into time-sensitive, time-stable, and long-term valid types based on its content and format.

[0019] A basic attenuation coefficient is set for each type of data, and dynamically adjusted based on the credibility of the data source and the degree of structure.

[0020] Calculate the time interval Δt based on the data generation time and the current time, and apply the exponential function w=k·e −d·△t The timeliness weight w is automatically calculated, so that the timeliness weight w decreases exponentially with the increase of Δt; where k is the normalization coefficient and d is the dynamically adjusted decay coefficient.

[0021] The statute of limitations weight w is used as metadata to mark the original judicial data, and a dynamic data vector with the statute of limitations weight is output.

[0022] Preferably, the calculation of the abnormal behavior intensity value further includes:

[0023] Dynamic data vectors carrying time-sensitive weights are mapped to knowledge graphs, where nodes represent judicial entities and edges represent relationships between entities;

[0024] The change in edge weights is dynamically calculated using a neural network model, so that the relation weights are adjusted in real time as the data vector changes.

[0025] Based on the probability distribution of edge weight changes, the intensity value of behavioral anomalies is calculated using the information entropy model.

[0026] When the intensity of abnormal behavior exceeds a preset threshold, the malicious litigation monitoring process is triggered.

[0027] Preferably, the compliance indicator value is obtained by comparing the similarity between the dynamic data vector and the feature vector of the legal case database, specifically including:

[0028] Extract the n-dimensional feature vector from dynamic data;

[0029] The n-dimensional feature vector is matched with the feature vectors in the legal case database to obtain the number of matched features;

[0030] The matching degree is obtained by calculating the proportion of the number of matched features to the total feature dimension n.

[0031] The matching degree will be used as a compliance indicator value.

[0032] Preferably, the dynamic updating of the rule base further includes the following steps:

[0033] Obtain the output results of the malicious litigation supervision process, and extract the adopted supervision decisions and related data features;

[0034] Based on the deviation between the predicted vector and the legal standard calculated using the judicial interpretation vector, a legal compliance loss value is generated.

[0035] The rule vector is modified along the negative gradient direction of legal compliance loss, and the update amount is constrained by the learning rate and the judicial compliance weight coefficient.

[0036] The newly added data features are input into the knowledge graph nodes to complete the closed-loop update.

[0037] Preferably, the method further includes generating a two-dimensional heatmap by integrating legal rules and feature contribution algorithms, wherein the horizontal axis maps the subject association hierarchy, the vertical axis maps the compliance index value, and a continuous heatfield representing the risk value is generated in the coordinate system with color depth; the feature contribution algorithm uses SHAP value calculation, and the risk value is generated by weighted aggregation of the contribution of each feature, with the weight dynamically adjusted by the partial derivative of the legal rules.

[0038] Preferably, the feature contribution algorithm further includes:

[0039] Monte Carlo sampling is performed on the feature set of the input judicial data to generate a combination of feature subsets;

[0040] Each feature subset is input into the pre-trained risk prediction model, the output prediction value is compared with the full feature prediction value, and the feature marginal contribution is obtained.

[0041] The SHAP contribution of each feature is output by weighting all marginal contribution values ​​according to the size of the feature subset.

[0042] The partial derivative of the judicial interpretation vector with respect to the features is used as a weight to dynamically adjust the contribution of SHAP.

[0043] Secondly, the present invention also provides an intelligent monitoring system for malicious litigation based on a dynamic legal knowledge graph, comprising:

[0044] The evidence statute of limitations processing module is used to acquire dynamic judicial data, identify the type of judicial data, set a corresponding decay coefficient according to the data type, automatically calculate the statute of limitations weight of the judicial data through an exponential decay model, mark the statute of limitations weight on the judicial data, and generate a dynamic data vector; wherein, the statute of limitations weight decays as the data statute of limitations interval increases;

[0045] The entropy control early warning module constructs a knowledge graph based on the dynamic data vector, updates the weight values ​​of the relationships between entities in the graph in real time, and calculates the intensity value of behavioral anomalies through the information entropy model; when the intensity value of behavioral anomalies exceeds a preset threshold, an early warning is triggered.

[0046] The dual-path trigger engine calculates in parallel the matching degree between the case features of the judicial data and the legal case database as a compliance indicator value, while simultaneously receiving the abnormal behavior intensity value; when the compliance indicator value is lower than the first threshold or the abnormal behavior intensity value is higher than the second threshold, the supervision process is initiated.

[0047] The rule self-optimization module dynamically updates the rule base based on the supervision results and in conjunction with the legal compliance loss function, wherein the update direction is constrained by the judicial interpretation vector.

[0048] In another aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, which, when executed by the processor, performs the intelligent supervision method for malicious litigation based on a dynamic legal knowledge graph as described above.

[0049] In another aspect, the present invention provides a computer-readable storage medium storing computer program instructions that, when executed, implement the intelligent supervision method for malicious litigation based on a dynamic legal knowledge graph as described in any of the preceding claims.

[0050] Compared with existing technologies, the intelligent supervision method for malicious litigation based on dynamic legal knowledge graphs in this invention effectively solves the problem of delayed judgment of judicial evidence timeliness through a dynamic evidence timeliness quantification mechanism, eliminates interference from outdated evidence, and improves the accuracy of heterogeneous data processing. Utilizing entropy-controlled knowledge graph early warning, it breaks through the bottleneck of hidden association identification, accurately capturing malicious patterns such as shell company capital increases / bulk litigation, significantly reducing the false alarm rate compared to traditional single-threshold rules. A dual-path hybrid trigger engine compensates for the deficiencies of a single judgment path, and case-type differentiated thresholds improve adaptability to complex scenarios. Based on a judicial constraint self-optimization system, the rule base update response time is compressed from days to minutes, ensuring that newly added features comply with judicial principles. Combined with an interpretable risk decision-making system (two-dimensional heatmap), it solves the AI ​​black box problem, providing prosecutors with decision support for one-click retrieval of related evidence through equity penetration risk visualization and legal clause tracing chains.

[0051] Specifically, the present invention has the following beneficial effects:

[0052] (1) This invention sets attenuation coefficients according to data type (time-sensitive / stable / long-term), and dynamically adjusts the coefficients based on the credibility of the data source and the degree of structuring, and automatically calculates the time-related weights through an exponential function, thereby solving the problem of the lag in manual judgment of evidence timeliness and eliminating interference from outdated evidence; and performs differentiated processing on different data types to improve the accuracy of timeliness quantification.

[0053] (2) The entropy control knowledge graph early warning mechanism of the present invention constructs a graph based on time-weighted weighting, updates edge weights in real time through neural networks, and quantifies the intensity of behavioral mutations through information entropy, thereby breaking through the bottleneck of hidden association identification. For example, the entropy model captures abnormal patterns such as capital increase of shell companies / batch litigation; and filters normal business behavior fluctuations by replacing a single threshold (such as monitoring only the number of lawsuits) with probability distribution entropy values, thereby reducing the false alarm rate.

[0054] (3) This invention uses a dual-path hybrid trigger engine to calculate legal compliance and behavioral abnormality in parallel, dynamically adjusts the case type of the threshold, and uses dual indicators to trigger supervision in a coordinated manner to make up for the defects of a single judgment path. Legal compliance (preventing missed judgment) is combined with behavioral abnormality (preventing misjudgment), which greatly reduces the false trigger rate.

[0055] (4) The judicial constraint self-optimization system of the present invention eliminates the rule update delay by driving the update through the legal compliance loss function; and avoids the rule base from deviating from judicial principles by constraining the gradient direction through the judicial interpretation vector.

[0056] (5) The heat map of the present invention maps the subject association hierarchy to the horizontal axis and the compliance index to the vertical axis, and uses color depth to dynamically represent the risk value, transforming complex legal rule conflicts and subject association networks into interactive visual evidence, which significantly improves the interpretability and judicial compliance of malicious litigation risk decisions.

[0057] This invention constructs the first intelligent supervision system for malicious litigation in the judicial field with timeliness awareness, adaptive evolution, and legal compliance protection through a full-chain innovation of dynamic evidence decay → entropy control graph early warning → dual-path hybrid triggering → judicial constraint optimization → explainable decision-making. It provides the procuratorial organs with a closed-loop solution covering risk identification, decision-making, and source tracing. Detailed Implementation

[0058] The technical solution of the present invention will now be clearly and completely described with reference to specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] Example 1

[0060] This embodiment provides an intelligent monitoring method for malicious litigation based on a dynamic legal knowledge graph. This method processes judicial data in real time, constructs a dynamic knowledge graph, and utilizes intelligent algorithms to monitor and issue early warnings for malicious litigation behavior. The specific implementation steps are as follows:

[0061] First, dynamic judicial data is acquired and its type is identified. A corresponding attenuation coefficient is set based on the data type, and the time-sensitivity weight of the judicial data is automatically calculated using an exponential decay model. This time-sensitivity weight is then assigned to the judicial data, generating a dynamic data vector. The time-sensitivity weight decays as the data's time-sensitivity interval increases.

[0062] During the generation of dynamic data vectors, the system parses the input judicial data and categorizes it into three types based on its content and format: time-sensitive, time-stable, and long-term valid. Time-sensitive data includes the latest judicial interpretations and temporary regulations; time-stable data includes routine precedents and court rulings; and long-term valid data includes legal provisions and basic principles. A basic attenuation coefficient is assigned to each data type, with a higher attenuation coefficient for time-sensitive data, a medium attenuation coefficient for time-stable data, and a lower attenuation coefficient for long-term valid data. The coefficient is dynamically adjusted based on the credibility of the data source and the degree of data structure. For example, data released by the Supreme People's Court has higher credibility than data from local courts, and highly structured data has higher reliability than unstructured data; these factors affect the final attenuation coefficient.

[0063] Calculate the time interval Δt based on the data generation time and the current time, and apply the exponential function w=k·e -d·△t The timeliness weight w is automatically calculated, causing it to decrease exponentially with increasing Δt. Here, k is a normalization coefficient used to ensure the weight value remains within a reasonable range, and d is a dynamically adjusted decay coefficient. The timeliness weight w is used as metadata to mark the original judicial data, outputting a dynamic data vector with the timeliness weight label.

[0064] Next, a knowledge graph is constructed based on dynamic data vectors, and the weight values ​​of the relationships between entities in the graph are updated in real time. The intensity value of behavioral anomalies is calculated using an information entropy model. An alert is triggered when the intensity value of behavioral anomalies exceeds a preset threshold.

[0065] In calculating the intensity of abnormal behavior, a dynamic data vector carrying time-sensitive weights is mapped to a knowledge graph. Nodes represent judicial entities, such as parties, courts, and cases, while edges represent relationships between entities, such as litigation relationships and agency relationships. A neural network model dynamically calculates changes in edge weights, allowing relationship weights to adjust in real-time as the data vector changes. Based on the probability distribution of edge weight changes, an information entropy model is used to calculate the intensity of abnormal behavior. When the intensity of abnormal behavior exceeds a preset threshold, a malicious litigation monitoring process is triggered.

[0066] The calculation of the change in edge weights includes:

[0067] The historical edge weights and the changes in the evidence vector are concatenated to form the input vector;

[0068] The hidden state is calculated using a trainable weight matrix;

[0069] The ReLU function is used to generate the changes in edge weights;

[0070] The change in edge weight is added to the current edge weight, and the weight is updated.

[0071] The calculation of the behavioral anomaly intensity value using the information entropy model specifically includes:

[0072] Extract edges from the knowledge graph whose weight changes exceed a threshold within a preset period;

[0073] The probability density of the weight changes of the filtering edges is estimated to generate a discrete probability distribution;

[0074] The intensity value of behavioral anomalies is output by summing the product of the probability distribution and the logarithmic function.

[0075] It also includes adjusting the entropy value based on the timeliness weight of dynamic data to generate the final anomaly intensity index.

[0076] Simultaneously, the matching degree between case characteristics of judicial data and the legal case database is calculated in parallel as a compliance indicator value, while also receiving a behavioral anomaly intensity value. When the compliance indicator value falls below the first threshold or the behavioral anomaly intensity value exceeds the second threshold, the monitoring process is initiated.

[0077] The compliance indicator value is obtained by comparing the similarity between the dynamic data vector and the feature vector of the legal case database. Specifically, this includes: extracting the n-dimensional feature vector of the dynamic data, which includes features of the litigation subject, litigation claims, and evidence; performing similarity matching between the n-dimensional feature vector and the feature vector of the legal case database to obtain the number of matched features; calculating the proportion of the number of matched features to the total feature dimension n to obtain the matching degree; and using the matching degree as the compliance indicator value.

[0078] For example, by comparing dynamic evidence with the Supreme People's Procuratorate's case database, the matching degree Qlaw is calculated as (number of matching features) / total number of features. Assuming that only 3 out of 10 features match, then Qlaw = 0.3. When the compliance indicator value is less than 0.5 and the anomaly intensity value is greater than 0.8, the supervision process is initiated (e.g., when "invalid patent still being litigated" behavior is detected).

[0079] Based on the monitoring results, the rule base is dynamically updated using a legal compliance loss function, with the update direction constrained by the judicial interpretation vector. The dynamic update of the rule base further includes the following steps: obtaining the output results of the malicious litigation monitoring process, extracting the adopted monitoring decisions and associated data features; calculating the deviation between the prediction vector and the legal standard based on the judicial interpretation vector, generating a legal compliance loss value; correcting the rule vector along the negative gradient direction of the legal compliance loss, with the update amount constrained by the learning rate and the judicial compliance weight coefficient; and inputting the newly added data features into the knowledge graph nodes to complete the closed-loop update.

[0080] The output includes warnings with legal basis (such as those marked "Violation of Case 2 of the Supreme People's Procuratorate"), injecting legal rule derivatives into the interpretable model to make AI decisions conform to judicial reasoning logic.

[0081] Furthermore, this method integrates legal rules and feature contribution algorithms to generate a two-dimensional heatmap. The horizontal axis maps the entity association hierarchy (equity penetration hierarchy), and the vertical axis maps compliance indicator values. A continuous heatmap representing risk values ​​by color depth is generated within the coordinate system. The feature contribution algorithm uses SHAP values ​​for calculation, and the risk value is generated by weighted aggregation of the contributions of each feature, with the weights dynamically adjusted by the partial derivatives of legal rules.

[0082] In the implementation of the feature contribution algorithm, Monte Carlo sampling is performed on the feature set of the input judicial data to generate a feature subset combination; each feature subset is input into the pre-trained risk prediction model, the output prediction value is compared with the full feature prediction value to obtain the feature marginal contribution; all marginal contribution values ​​are weighted and averaged according to the size of the feature subset to output the SHAP contribution of each feature; the partial derivative value of the judicial interpretation vector with respect to the feature is used as the weight to dynamically adjust the SHAP contribution.

[0083] Through the above methods, the system can monitor judicial data flow in real time, identify potential malicious litigation behavior, and provide intelligent supervision and early warning mechanisms, effectively improving judicial efficiency and fairness.

[0084] Case Study: A Professional Formulaic Expression of a Procuratorate's Case

[0085] Business Background: A technology company initiated 53 malicious lawsuits in 8 provinces through multiple shell companies, taking advantage of the data silos in the traditional judicial system (isolated cross-domain data and delayed rule updates) to covertly transfer the relevance of the litigation subjects and create blind spots in judicial supervision.

[0086] The business problem solved by this invention:

[0087] Data timeliness out of control: Traditional systems cannot dynamically reduce the weight of outdated evidence (such as a patent invalidation decision from 2 years ago still being misused). This solution achieves automatic decay of electronic data through an exponential decay model (d=0.03, Δt=2 years→w=0.94).

[0088] Hidden correlation identification is lacking: It is difficult for humans to detect the abnormal pattern of "sudden increase in registered capital within a day + batch of cross-provincial lawsuits". This solution uses knowledge mutation entropy to quantify behavioral abnormalities (sudden increase in lawsuits → abnormal intensity value = 0.96 > 0.8).

[0089] Lagging rule updates: Adding new malicious litigation features (such as shell company control chains) requires manual input for more than 48 hours. This solution uses a legal compliance loss function to dynamically update the rule base, achieving a self-optimization closed loop within ≤25 minutes.

[0090] Solution result:

[0091] Accurately identify the actual controller's connections (equity penetration through 3 layers → heat map risk value R=0.91);

[0092] The system generates cross-provincial prosecutorial recommendations and transfers leads to public security authorities. The rule base has also been dynamically updated with features such as "sudden increase in registered capital + batch litigation".

[0093] When newly added features (such as "sudden increase in registered capital + batch litigation") violate legal principles, the judicial constraints will be forcibly corrected and updated.

[0094] Example 2

[0095] This embodiment provides an intelligent monitoring system for malicious litigation based on a dynamic legal knowledge graph, including an evidence statute of limitations processing module, an entropy control early warning module, a dual-path triggering engine, and a rule self-optimization module. The modules are connected through a data bus to form a complete monitoring link.

[0096] The system's overall architecture adopts a layered design, with the bottom layer being the data acquisition and processing layer, the middle layer being the knowledge graph construction and analysis layer, and the top layer being the supervision, decision-making, and optimization layer. Data interaction between each layer is achieved through standardized interfaces, ensuring the system's stability and scalability.

[0097] The evidence statute of limitations processing module is the system's fundamental data processing unit. This module includes a data acquisition unit, a type identification unit, a decay coefficient configuration unit, and a statute of limitations weight calculation unit. The data acquisition unit connects to various judicial databases via API interfaces to acquire judicial data in real time, including but not limited to judgments, rulings, indictments, and answers. The type identification unit uses deep learning algorithms to classify the acquired judicial data, identifying it as belonging to different types such as judgments, rulings, litigation documents, or evidence. The decay coefficient configuration unit sets corresponding decay coefficients according to different types of judicial data; for example, a decay coefficient of 0.95 is set for judgments, 0.90 for indictments, and 0.85 for ordinary evidence. The statute of limitations weight calculation unit automatically calculates the statute of limitations weight of the judicial data using an exponential decay model. This module marks the calculated statute of limitations weight in the judicial data, ultimately generating a dynamic data vector containing statute of limitations information.

[0098] In a preferred embodiment, the evidence statute of limitations processing module further includes a data cleaning unit for preprocessing the original judicial data, including removing redundant information, correcting format errors, and supplementing missing fields, thereby improving the accuracy of subsequent processing. The data cleaning unit employs a combination of rule-based filtering and machine learning to identify and process abnormal data, ensuring the quality of data entering the system.

[0099] The entropy control and early warning module is the core analysis unit of the system. This module includes a knowledge graph construction unit, a relationship weight update unit, and an information entropy calculation unit. The knowledge graph construction unit receives dynamic data vectors from the evidence statute of limitations processing module and constructs a dynamic legal knowledge graph using technologies such as entity recognition, relationship extraction, and ontology mapping. This knowledge graph includes various entity types such as litigation subjects, case facts, legal clauses, and judicial interpretations, as well as various relationship types such as litigation relationships, agency relationships, and evidence support relationships. The relationship weight update unit updates the weight values ​​of relationships between entities in the graph in real time based on newly added data, with an update frequency of once per hour. The information entropy calculation unit calculates the behavioral anomaly intensity value of behavioral mutations using an information entropy model, with the formula A = -Σ(Pi·logPi), where A is the behavioral anomaly intensity value, and Pi is the probability of behavior i occurring. For example, when the behavioral anomaly intensity value exceeds a preset threshold of 8.5, the system automatically triggers an early warning, generates an early warning report, and pushes it to relevant regulatory personnel.

[0100] In a preferred embodiment, the entropy control early warning module also integrates a multi-dimensional anomaly detection algorithm, which not only calculates information entropy, but also combines time-series pattern analysis, social network analysis and other technologies to comprehensively evaluate the degree of anomaly from the time dimension, relationship dimension and behavior pattern dimension, thereby improving the accuracy of early warning and reducing the false alarm rate.

[0101] The dual-trigger engine is the core of the system's decision-making process. This engine comprises a case feature extraction unit, a compliance calculation unit, and a supervision process initiation unit. The case feature extraction unit extracts multi-dimensional case feature vectors from judicial data, including litigation frequency, geographical distribution of litigation, type of litigation claims, and type of evidence. The compliance calculation unit calculates the matching degree between the case features in the judicial data and the legal case database in parallel, generating a compliance index value between 0 and 1, with lower values ​​indicating poorer compliance. This unit also receives behavioral anomaly intensity values ​​from the entropy control early warning module. Based on the dual-trigger mechanism, the supervision process initiation unit automatically initiates the supervision process, including in-depth case review, party behavior analysis, and litigation risk assessment, for example, when the compliance index value falls below a first threshold of 0.4 or the behavioral anomaly intensity value exceeds a second threshold of 7.5.

[0102] In a preferred embodiment, the dual-path triggering engine employs a weighted decision-making mechanism, which can dynamically adjust the weight ratio of compliance indicators and abnormal behavior intensity according to the characteristics of different types of cases, thereby improving the accuracy of triggering decisions. For example, for commercial cases, the weight of compliance indicators can be set to 0.6, and the weight of abnormal behavior intensity can be set to 0.4; while for civil cases, they can be set to 0.5 and 0.5 respectively.

[0103] The rule self-optimization module is the system's learning and evolution unit. This module includes a supervision result analysis unit, a legal compliance loss function unit, and a rule dynamic update unit. The supervision result analysis unit collects the execution results of the supervision process, including information such as whether the supervision was effective, whether malicious litigation was detected, and the effectiveness of the supervision measures. The legal compliance loss function unit calculates the loss value of the system rules based on the supervision results. The loss function is in the form L=α·(1-P)²+β·F, where P is the prediction accuracy, F is the cost of misjudgment, and α and β are weighting coefficients. The rule dynamic update unit dynamically updates the rule base according to the calculation results of the loss function. The update direction is constrained by the judicial interpretation vector to ensure that the rule updates comply with legal norms and the spirit of judicial interpretations. The rule update cycle is once a week, with special updates occurring within 24 hours of the release of major judicial interpretations.

[0104] In a preferred embodiment, the rule self-optimization module also includes a multi-source feedback integration mechanism, which not only considers the internal supervision results of the system, but also integrates external information sources such as court judgments, expert opinions and practical feedback to form a more comprehensive optimization basis and improve the adaptability and accuracy of the rule base.

[0105] This system, through the coordinated operation of four modules—evidence statute of limitations processing, entropy control early warning, dual-path triggering, and rule self-optimization—achieves intelligent supervision of malicious litigation behavior, effectively improving judicial efficiency, reducing waste of judicial resources, and safeguarding the legitimate rights and interests of litigants. The system adopts a modular design, allowing each functional unit to be independently upgraded to adapt to constantly changing legal environments and litigation models.

[0106] Third Embodiment

[0107] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method of any of the above embodiments.

[0108] The computer device includes a processor and a memory, and may also include an input system and an output system. The processor, memory, input system, and output system can be connected via a bus or other means. The input system can receive input digital or character information, as well as signal inputs that generate the method of this invention. The output system may include a display device such as a screen.

[0109] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as program instructions / modules corresponding to the methods in the embodiments of this application. Memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required for at least one function; the data storage area may store data created by the use of the method of this invention, etc. Furthermore, memory may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the local module via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0110] In some embodiments, the processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor is typically used to control the overall operation of a computer device. In this embodiment, the processor is used to run program code stored in memory or process data. In this embodiment, the processors of multiple computer devices execute various server functions and data processing by running non-volatile software programs, instructions, and modules stored in memory, thereby implementing the steps of the above-described method embodiments.

[0111] Fourth embodiment

[0112] This invention also provides a computer-readable storage medium storing computer program instructions that, when executed, implement the method of any of the above embodiments.

[0113] It should be understood that, without conflict, all the embodiments, features and advantages described above for the method according to the present invention are equally applicable to the systems, devices and storage media of the present invention.

[0114] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the functionality of various illustrative components, blocks, modules, circuits, and steps has been generally described. Whether this functionality is implemented as software or as hardware depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the embodiments disclosed herein.

[0115] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.

[0116] It should be understood that, as used herein, the singular form "a" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the associatedly listed items. The embodiment numbers disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0117] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A method for intelligent monitoring of malicious litigation based on dynamic legal knowledge graphs, characterized in that, Includes the following steps: The system acquires dynamic judicial data and identifies the type of judicial data. Based on the data type, it sets a corresponding attenuation coefficient and automatically calculates the timeliness weight of the judicial data using an exponential attenuation model. The timeliness weight is then marked on the judicial data to generate a dynamic data vector. The timeliness weight attenuates as the data timeliness interval increases. A knowledge graph is constructed based on the dynamic data vectors, and the weight values ​​of the relationships between entities in the graph are updated in real time. The intensity value of behavioral anomalies is calculated using the information entropy model. An early warning is triggered when the intensity value of behavioral anomalies exceeds a preset threshold. The system calculates the matching degree between the case characteristics of the judicial data and the legal case database in parallel as a compliance indicator value, and simultaneously receives the abnormal behavior intensity value; when the compliance indicator value is lower than the first threshold or the abnormal behavior intensity value is higher than the second threshold, the supervision process is initiated. Based on the monitoring results, the rule base is dynamically updated in conjunction with the legal compliance loss function, wherein the update direction is constrained by the judicial interpretation vector; The generation of the dynamic data vector further includes: parsing the input judicial data and classifying it into time-sensitive, time-stable, and long-term valid types according to its content and format; setting a basic attenuation coefficient for each type of data and dynamically adjusting it based on the credibility of the data source and the degree of structure; calculating the time interval Δt based on the data generation time and the current time, and applying it according to an exponential function. The time-sentiment weight w is automatically calculated, so that the time-sentiment weight w decreases exponentially with the increase of Δt; where k is the normalization coefficient and d is the dynamically adjusted decay coefficient; the time-sentiment weight w is used as metadata to mark the original judicial data, and the dynamic data vector marked with the time-sentiment weight is output. The calculation of the abnormal behavior intensity value further includes: mapping a dynamic data vector carrying time-limited weights into a knowledge graph, where nodes represent judicial entities and edges represent relationships between entities; dynamically calculating the change in edge weights through a neural network model, so that the relationship weights are adjusted in real time as the data vector changes; calculating the abnormal behavior intensity value through an information entropy model based on the probability distribution of edge weight changes; and triggering a malicious litigation supervision process when the abnormal behavior intensity value exceeds a preset threshold.

2. The intelligent supervision method for malicious litigation based on dynamic legal knowledge graph as described in claim 1, characterized in that, The compliance indicator values ​​are obtained by comparing the similarity between dynamic data vectors and feature vectors in a legal case database, specifically including: Extract the n-dimensional feature vector from dynamic data; The n-dimensional feature vector is matched with the feature vectors in the legal case database to obtain the number of matched features; The matching degree is obtained by calculating the proportion of the number of matched features to the total feature dimension n. The matching degree will be used as a compliance indicator value.

3. The intelligent supervision method for malicious litigation based on dynamic legal knowledge graph as described in claim 1, characterized in that, The dynamic updating of the rule base further includes the following steps: Obtain the output results of the malicious litigation supervision process, and extract the adopted supervision decisions and related data features; Based on the deviation between the predicted vector and the legal standard calculated using the judicial interpretation vector, a legal compliance loss value is generated. The rule vector is modified along the negative gradient direction of legal compliance loss, and the update amount is constrained by the learning rate and the judicial compliance weight coefficient. The newly added data features are input into the knowledge graph nodes to complete the closed-loop update.

4. The intelligent supervision method for malicious litigation based on dynamic legal knowledge graph as described in claim 1, characterized in that, It also includes generating a two-dimensional heat map by integrating legal rules and feature contribution algorithms, where the horizontal axis maps the subject association hierarchy and the vertical axis maps the compliance index value, and a continuous heat field representing the risk value is generated in the coordinate system with color depth; the feature contribution algorithm uses SHAP value to calculate the risk value, and the risk value is generated by weighted aggregation of the contribution of each feature, with the weight dynamically adjusted by the partial derivative of the legal rules.

5. The intelligent supervision method for malicious litigation based on dynamic legal knowledge graph according to claim 4, characterized in that, The feature contribution algorithm further includes: Monte Carlo sampling is performed on the feature set of the input judicial data to generate a combination of feature subsets; Each feature subset is input into the pre-trained risk prediction model, the output predicted value is compared with the full feature predicted value to obtain the feature marginal contribution; The SHAP contribution of each feature is output by weighting all marginal contribution values ​​according to the size of the feature subset. The partial derivative of the judicial interpretation vector with respect to the features is used as a weight to dynamically adjust the contribution of SHAP.

6. A malicious litigation intelligent supervision system based on a dynamic legal knowledge graph, characterized in that, include: The evidence statute of limitations processing module is used to acquire dynamic judicial data, identify the type of judicial data, set a corresponding decay coefficient according to the data type, automatically calculate the statute of limitations weight of the judicial data through an exponential decay model, mark the statute of limitations weight on the judicial data, and generate a dynamic data vector; wherein, the statute of limitations weight decays as the data statute of limitations interval increases; The entropy control early warning module constructs a knowledge graph based on the dynamic data vector, updates the weight values ​​of the relationships between entities in the graph in real time, and calculates the intensity value of behavioral anomalies through the information entropy model; when the intensity value of behavioral anomalies exceeds a preset threshold, an early warning is triggered. The dual-path trigger engine calculates in parallel the matching degree between the case features of the judicial data and the legal case database as a compliance indicator value, while simultaneously receiving the abnormal behavior intensity value; when the compliance indicator value is lower than the first threshold or the abnormal behavior intensity value is higher than the second threshold, the supervision process is initiated. The rule self-optimization module dynamically updates the rule base based on the supervision results and the legal compliance loss function, wherein the update direction is constrained by the judicial interpretation vector. The generation of the dynamic data vector further includes: parsing the input judicial data and classifying it into time-sensitive, time-stable, and long-term valid types according to its content and format; setting a basic attenuation coefficient for each type of data and dynamically adjusting it based on the credibility of the data source and the degree of structure; calculating the time interval Δt based on the data generation time and the current time, and applying it according to an exponential function. The time-sentiment weight w is automatically calculated, so that the time-sentiment weight w decreases exponentially with the increase of Δt; where k is the normalization coefficient and d is the dynamically adjusted decay coefficient; the time-sentiment weight w is used as metadata to mark the original judicial data, and the dynamic data vector marked with the time-sentiment weight is output. The calculation of the abnormal behavior intensity value further includes: mapping a dynamic data vector carrying time-limited weights into a knowledge graph, where nodes represent judicial entities and edges represent relationships between entities; dynamically calculating the change in edge weights through a neural network model, so that the relationship weights are adjusted in real time as the data vector changes; calculating the abnormal behavior intensity value through an information entropy model based on the probability distribution of edge weight changes; and triggering a malicious litigation supervision process when the abnormal behavior intensity value exceeds a preset threshold.

7. A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, performs the intelligent monitoring method for malicious litigation based on a dynamic legal knowledge graph as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing computer program instructions, which, when executed, implement the intelligent supervision method for malicious litigation based on a dynamic legal knowledge graph as described in any one of claims 1 to 5.