Evidence grouping method and system for litigation cases, terminal and medium
By performing content analysis and graph structure feature learning on evidence in litigation cases, evidence grouping representation vectors are generated, and the grouping structure is iteratively optimized. This solves the problem of inconsistent evidence grouping in litigation cases with a large number of pieces of evidence and complex relationships, and realizes the structured expression and logically consistent grouping of evidence.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSPUR YUNZHOU (SHANDONG) IND INTERNET CO LTD
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-10
AI Technical Summary
When faced with litigation cases involving a large amount of evidence and complex relationships, existing technologies struggle to produce evidence grouping results that are consistent with legal logic.
By acquiring multiple types of evidence in a case, content analysis is performed to generate unified feature representations, factual elements and related information are identified, a graph structure is constructed and feature learning is performed to generate evidence grouping representation vectors, the grouping structure is iteratively optimized, and the relationships between evidence are inferred.
It enables structured expression and logically consistent grouping of evidence, improves the efficiency of evidence organization and case fact sorting, and enhances the accuracy of evidence relevance identification and grouping stability.
Smart Images
Figure CN121835652A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of evidence grouping, and particularly relates to an evidence grouping method, system, terminal and medium for litigation cases. BACKGROUND
[0002] With the continuous advancement of the digitization process of judicial trials and procuratorial business, the number of evidence materials formed in the case handling process is rapidly increasing, and the types of evidence cover witness testimony, interrogation records, call records, electronic documents, communication data and other forms. There are often complex correlations between a large number of evidences, such as time sequence, behavior causality and subject participation, and these correlations will not be presented in a structured form in the original materials. The case handling personnel need to manually sort out the fact chain and judge the logical relationship between the evidences after reading a large number of materials, which is high in work intensity and easy to miss.
[0003] In the prior art, some studies extract events, subjects and time information from evidence content based on text analysis methods, and assist evidence classification through similarity calculation or rule matching methods; some technologies also attempt to construct a relationship graph based on case records to show the subject behavior relationship and case context.
[0004] However, these technologies focus on information extraction or visual display, and usually rely on general clustering algorithms or simple association rules in evidence grouping. Since the unique logical chain, causal structure and subject participation structure of litigation evidence are not considered, these methods are difficult to output evidence grouping results with consistent legal logic when facing cases with large number of evidences and complex relationships. SUMMARY
[0005] The application provides an evidence grouping method, system, terminal and medium for litigation cases to solve the problem that the prior art is difficult to output evidence grouping results with consistent legal logic when facing cases with large number of evidences and complex relationships in the background art.
[0006] The technical solution adopted by the application is as follows: In a first aspect, the application provides an evidence grouping method for litigation cases, which includes the following steps: Obtaining multiple types of evidence in a case, performing content analysis on the evidence to generate a unified feature representation for representing evidence information; Identifying fact elements and associated information involved in the evidence from the obtained unified feature representation, and constructing a graph structure for representing the relationship between the evidences; Performing feature learning on the constructed graph structure to obtain a representation vector capable of representing the correlation between the evidences, and generating a preliminary evidence grouping result according to the evidence grouping representation vector; Iterative optimization is performed on the preliminary evidence grouping result, and the grouping structure is adjusted according to a preset strategy to obtain an optimized evidence grouping; Based on the optimized grouping, the relationship between the evidences is analyzed by reasoning to obtain interpretation information for characterizing the evidence grouping basis.
[0007] Further, the obtained unified feature representation identifies the fact elements and related information involved in the evidence, including: The unified feature representation is structurally decomposed to obtain event features for characterizing event actions, time features for characterizing occurrence times, and subject features for characterizing involved subjects; The fact correlation is determined by comparing the causal pointing relationship of the event features of different evidences, the time sequence correlation is determined by time sequence calculation on the time features of different evidences, and the content correlation is determined by similarity calculation on the subject features, behavior semantic features and content similarity features.
[0008] Further, the structured decomposition of the unified feature representation includes: dividing the unified feature representation of the evidence into feature channels, separating the semantic channel for characterizing action content, the time sequence channel for characterizing time description, and the role channel for characterizing subject attributes from the unified feature representation; The feature distribution in each channel is pattern clustered to obtain independent representations of event features, time features and subject features.
[0009] Further, the feature learning on the constructed graph structure includes: performing multi-round message passing on the feature vectors of the nodes in the graph structure and the connection relationship between the nodes, weighting and aggregating the features of adjacent nodes according to the weight of the connection relationship, and combining and updating the features of the current node; When the node features can reflect both their own attributes and the association information of adjacent nodes after multi-round iteration, the node representation vector that can represent the global correlation of the evidence is obtained; In the evidence grouping message passing and feature updating process, the feature update of node v in the lth iteration satisfies the following formula:
[0010] Wherein, denotes the feature vector of node v in the lth iteration; denotes the updated feature vector of node v in the (l+1) th iteration; denotes the set of adjacent nodes of node v; denotes the feature vector of adjacent node u in the lth iteration; denotes the edge feature of the connection edge between node u and node v, which is used to represent the association type between the evidences; a linear transformation matrix for updating the characteristics of the node itself; a linear transformation matrix for processing the edge characteristics; is the connection weight between the node u and the node v, which is used to reflect the association strength between different evidences; is a nonlinear activation function, which is used to perform nonlinear mapping on the linearly transformed characteristics.
[0011] Further, generating the preliminary evidence grouping result according to the evidence grouping representation vector comprises: calculating the mutual distance of the representation vectors corresponding to all evidences, identifying the aggregation region of the representation vectors according to the distance distribution, and determining the aggregation center to which multiple evidences can be grouped; The evidence grouping divides the evidences into multiple preliminary groupings according to the corresponding relationship between the evidence representation vectors and the evidence grouping aggregation centers.
[0012] Further, the iterative optimization of the preliminary evidence grouping result comprises: calculating the intra-group similarity of each evidence in the current grouping and the inter-group difference with other groupings, and identifying the evidence or grouping that needs to be adjusted according to the case of reducing the intra-group similarity or reducing the inter-group difference; The evidence grouping updates the grouping attribution or grouping boundary of the evidence grouping evidence in each iteration until the grouping result meets the preset stability condition.
[0013] Further, the evidence grouping preset strategy adjusts the grouping structure, which comprises: determining the corresponding strategy instruction according to the intra-group similarity of each evidence in the current grouping, the boundary distance with the adjacent grouping, and the structural balance within the grouping; The evidence grouping strategy instruction comprises merging adjacent groupings, splitting the groupings, adjusting the evidence from the current grouping to other groupings, or creating a new grouping for the isolated evidence; The evidence grouping updates the grouping structure according to the evidence grouping strategy instruction in each iteration; The reasoning analysis of the relationship between the evidences based on the optimized grouping comprises: generating the association path according to the connection relationship between the evidences in the optimized grouping; The time sequence, causal direction and content correlation in the association path are reasoned level by level to construct the reasoning result reflecting the evidence chain, and the evidence grouping extracts the explanation information capable of representing the logic of the evidence grouping according to the evidence grouping reasoning result.
[0014] In a second aspect, the present application provides an evidence grouping system for litigation cases, which is used to implement the evidence grouping method for litigation cases as described in the first aspect, and the system comprises: An evidence characteristic representation generation unit is configured to obtain multiple types of evidence in a case, perform content analysis on the evidence, and generate a unified characteristic representation for representing the evidence information. a fact element identification unit configured to structureally decompose the uniform feature representation of the evidence groups, identify event features for representing event actions, time features for representing occurrence times, and subject features for representing subject relationships involved; a relationship construction unit configured to determine fact correlations, time sequence correlations, and content correlations between the evidences according to the event features, the time features, and the subject features of the evidence groups, and construct a graph structure for representing relationships between the evidences according to the correlations; a graph structure feature learning unit configured to perform multiple rounds of message passing, weighted aggregation, and feature updating on feature vectors of nodes in the graph structure of the evidence groups and connection relationships between the nodes, to obtain node representation vectors capable of representing correlation between the evidences; a preliminary grouping generation unit configured to perform distance calculation, clustered region identification, and clustered center determination according to the node representation vectors of the evidence groups, to generate a preliminary evidence grouping result; a grouping iteration optimization unit configured to calculate intra-group similarity and inter-group difference of the preliminary evidence grouping, and adjust grouping attribution or grouping boundaries of the evidence groups according to a preset strategy, to obtain an optimized evidence grouping; a reasoning analysis unit configured to generate correlation paths according to connection relationships between the evidences based on the optimized evidence grouping, and perform step-by-step reasoning on time sequence, cause-effect direction, and content relevance in the correlation paths, to obtain explanation information for representing a basis of the evidence grouping.
[0015] In a third aspect, the present application provides a terminal, comprising: a memory configured to store an evidence grouping procedure for litigation cases; a processor configured to execute the evidence grouping procedure for litigation cases to implement steps of the evidence grouping method for litigation cases according to the first aspect.
[0016] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and when a computer reads the computer instructions in the storage medium, the computer executes the evidence grouping method for litigation cases according to the first aspect.
[0017] It can be seen from the above technical solutions that the present application has the following advantages: By sequentially constructing evidence features, identifying fact elements, forming a relationship graph structure, and performing feature learning, the present application can uniformly model event connections, time sequences, and subject participation between evidences in the case of a large number of evidences, various evidence sources, and unclear relationship structures, so that originally scattered and disordered evidence information can be presented in a structured form, laying a foundation for subsequent evidence relationship analysis and significantly improving the efficiency of evidence organization, induction, and case fact analysis.
[0018] By structurally decomposing the unified feature representation and identifying event features, time features, and subject features, the behavior content, time sequence, and involved subjects in the evidence can be expressed in an abstract and computable manner. Further, by determining the fact, time, and content correlation between different pieces of evidence based on the causal pointing relationship, time sequence changes, and semantic and subject feature similarity, the structured information of the evidence can be presented in a clearer logical relationship, which is conducive to automatically discovering key evidence chains and identifying support and contradiction relationships between pieces of evidence from a large amount of materials, and improving the accuracy of evidence relevance identification.
[0019] By dividing the unified feature representation into feature channels and performing mode clustering in the semantic channel, time sequence channel, and subject channel, the information in different dimensions in the evidence can be decoupled in the representation space, and the features of the event action, time sequence, and subject participation structure can be modeled in an independent manner. Through this decomposition and clustering manner, the expression of different types of evidence information is more stable and more distinguishable, thereby enhancing the reliability of subsequent correlation mining between pieces of evidence and making the basis for evidence feature learning and relationship construction more consistent and clear.
[0020] By performing multiple rounds of message passing, weighted aggregation, and feature updating on the evidence nodes in the graph structure, each evidence node can simultaneously fuse its own features and the semantic relationship, time relationship, and subject relationship of its adjacent evidence nodes, and structured expression of the complex evidence network is achieved. The node representation vector obtained after multiple iterations can comprehensively reflect the overall correlation between pieces of evidence, and the relationship between pieces of evidence is promoted from a local perspective to a global perspective, which helps to provide stable and comprehensive numerical expression for evidence grouping and improves the accuracy of subsequent grouping processing.
[0021] By calculating the mutual distance of the representation vectors and identifying the clustering area, the evidence can be preliminarily divided according to the similarity of its correlation structure, so that the search range can be automatically reduced and several preliminary evidence sets with internal close correlation can be formed in a large number of evidence. Based on the clustering center, the evidence is allocated, so that the preliminary grouping can better reflect the structural features and correlation rules between pieces of evidence, thereby providing a reasonable initial structure for subsequent grouping optimization and improving the efficiency and controllability of the evidence organization process.
[0022] By calculating the intra-group similarity and inter-group difference of the preliminary evidence grouping, the quality of the evidence grouping can be quantitatively judged, and when the evidence attribution is unreasonable, the intra-group difference is too large, or the grouping boundary is not clear, the evidence or grouping structure that needs to be adjusted can be automatically identified, and the grouping attribution of the evidence is gradually optimized in the iteration process. The iteration update mechanism can effectively improve the stability and consistency of the evidence grouping, and the final grouping structure is more consistent with the requirements of the case fact logic and the evidence chain structure.
[0023] By determining the grouping strategy instruction according to the similarity within the group, the distance of the grouping boundary and the structural balance, the operations such as merging grouping, splitting grouping, adjusting evidence attribution and creating new grouping can be flexibly performed in the evidence grouping process, so as to adapt to the support relationship, contradictory relationship and logical structure change among different evidences in the case. The grouping structure is dynamically updated combined with the strategy instruction, and the evidence correlation path is generated based on the optimized grouping, which can further reason the time sequence, cause and content association among the evidences, so that the evidence grouping logic and the evidence chain structure can be presented in a clear reasoning result, and a structured basis is provided for fact analysis. BRIEF DESCRIPTION OF DRAWINGS
[0024] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.
[0025] Figure 1 The step diagram of the evidence grouping method for litigation cases in the embodiment; Figure 2 The reasoning analysis diagram of the relationship between the evidences in the embodiment; Figure 3 The architecture diagram of the evidence grouping system for litigation cases in the embodiment. DETAILED DESCRIPTION
[0026] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.
[0027] Please refer to Figure 1 The present application provides an evidence grouping method for litigation cases, which comprises: Step S1, acquiring multiple categories of evidences in a case, performing content analysis on the evidences, and generating a uniform feature representation for representing evidence information; In practical applications, the evidence can come from transcripts, witness testimony excerpts, communication record summaries, chat record intercepts, electronic document fragments, etc. When performing content analysis, first preprocess the evidence text by dividing it into sentences, removing noise, and structuring it, identify sentence boundaries and key action words. Then, unify the content of evidence from different sources into a vectorized representation, and map event descriptions, subject titles, time words, etc. in the evidence text to the same feature space to form a unified feature representation for subsequent analysis.
[0028] In a case involving an economic dispute, the evidence includes two WeChat chat records, a call record summary, and a contract summary. The chat records are semantically decomposed, and key action words such as "transfer" "sign" "confirm" are extracted from the conversation between the two parties. The contract summary extracts the contract date and subject name. Finally, the above different forms of evidence information are uniformly encoded into a unified feature representation vector.
[0029] Step S2, identify the fact elements and related information involved in the obtained unified feature representation, and construct a graph structure for representing the relationship between the evidence; By performing semantic pattern analysis on the unified feature representation, the basic legal fact elements contained in the evidence are extracted, including event action, occurrence time, participating subject, and behavior object. Then, according to the behavior type similarity, time sequence correspondence, and subject overlap between different evidence, the connection relationship between nodes is constructed, and a preliminary evidence relationship graph structure is generated. Each node in the graph corresponds to a piece of evidence, and each connection edge corresponds to an explicit association between two pieces of evidence.
[0030] In a theft case, one piece of evidence records that the suspect "entered the warehouse at 2 a.m.", and another piece of evidence records that "the warehouse manager discovered that the goods were moved early in the day". According to the time sequence and action causal relationship, it is identified that there is a connection between the two pieces of evidence, and a connection edge is created between the two pieces of evidence in the graph structure.
[0031] Step S3, perform feature learning on the constructed graph structure to obtain a representation vector that can represent the association between the evidence, and generate a preliminary evidence grouping result based on the evidence grouping representation vector. In this step, the evidence relationship graph constructed in the previous stage is subjected to hierarchical information propagation, so that each evidence node in the graph can gradually integrate the structural relationships of the associated evidence around it. During execution, the semantic content, time annotation, and subject participation information of each node are read, as well as the corresponding features of its associated nodes. The representation of the node is continuously updated in multiple rounds of propagation, so that it can cover the overall association structure between multiple pieces of evidence in the case. When the node representation vector is updated to a stable state, the mutual distance, semantic proximity, and structural similarity between these vectors are used to determine the clustering relationship of the evidence in the vector space, and the evidence set with high density in the vector distribution is divided into a preliminary grouping.
[0032] For example, in a contract fraud case, different evidence records the behaviors of "suspect sending false bidding documents", "victim paying deposit as instructed", and "suspect going missing". In the feature learning stage, the action sequence and behavior subject consistency between these evidences are propagated, so that the node representations of these evidences gradually approach each other. Finally, they form a clear clustering area in the vector space and are automatically grouped into the same preliminary grouping, while unrelated evidence (such as "victim's report registration record") forms another clustering area.
[0033] Step S4, iteratively optimizing the preliminary evidence grouping result, adjusting the grouping structure according to the preset strategy, and obtaining the optimized evidence grouping; In this step, the rationality of the preliminary grouping is continuously evaluated. For each piece of evidence, the similarity between it and other evidence in the same group is calculated to determine whether the evidence is suitable to remain in the existing grouping; at the same time, the difference between it and other groupings is calculated to identify whether the evidence is more suitable for migration to other evidence set. If there is a theme inconsistency or evidence content span that is too large within a grouping, the grouping will be split according to the strategy; on the contrary, if multiple groupings present high similarity, a merge operation will be performed. The entire optimization process is executed in multiple iterations, and after each execution, the grouping stability is recalculated. When all evidence presents a stable structure, the final grouping result is obtained.
[0034] In a violent conflict case, a preliminary grouping includes "dispute before the conflict", "physical contact during the conflict", and "post-conflict police report". During the iteration process, it is identified that "police report" has significant differences in time location and event nature from the first two, resulting in low intra-group similarity. Therefore, the evidence is automatically moved to another dominant event group. The final optimized grouping corresponds to "conflict behavior chain" and "post-treatment behavior chain" respectively, with clear structure and logical consistency.
[0035] Step S5, based on the optimized grouping, reasoning analysis is performed on the relationship between the evidence to obtain explanation information for characterizing the evidence grouping basis.
[0036] In this step, the evidence relationship is comprehensively reasoned according to the stabilized grouping structure. First, the time sequence, behavior logic structure and subject participation trajectory between the evidence in each group are identified to construct a complete evidence chain. Then, the possible indirect causal relationship, behavior connection relationship or information reinforcement relationship between multiple evidence chains are further analyzed. In the reasoning process, based on the key descriptions in the evidence content, such as action verbs, time words, subject names, etc., structured explanation information is generated, which can reflect the logical path between the evidence, why the evidence is classified into the same group, the key nodes in the chain, etc. Finally, the explanation information can be used as auxiliary material for the reconstruction of the case facts.
[0037] For example, in the final optimized grouping of a credit card fraud case, a group of evidence includes "suspect sends fake web link", "victim fills in bank card information on the link page", and "suspect uses the information for fraudulent transactions". According to the time sequence and behavior causal relationship between these evidences, an explanation path is automatically generated: "fake link sending → information filling → fraudulent behavior", and the structured explanatory text is output: "the evidence in this group collectively describes the continuous behavior chain of the suspect obtaining bank information through deceptive means and implementing fraudulent transactions", thus forming a clear evidence grouping basis.
[0038] In some embodiments, identifying the obtained unified feature representation involves identifying fact elements and associated information involved in the evidence, including: Structurally decomposing the unified feature representation to obtain event features for characterizing event actions, time features for characterizing occurrence times, and subject features for characterizing involved subjects; Determining fact associations by comparing the causal direction relationships of event features of different evidence, determining time sequence associations by performing time sequence calculations on the time features of different evidence, and determining content associations by performing similarity calculations on subject features, behavior semantic features and content similarity features.
[0039] The unified feature representation is decomposed into three dimensions of event, time and subject. The event dimension mainly identifies behavior categories, action targets and action directions; the time dimension identifies absolute or relative time expressions in the evidence; and the subject dimension identifies individual or organizational names involved in the evidence. Then, possible causal relationships are inferred according to the sequence of event features, and the evidence is sorted according to the time points in the time features, and the degree of closeness in content is determined through subject name coincidence and keyword semantic similarity analysis.
[0040] For example, two pieces of evidence respectively express "Zhang asks Li for money" and "Li pays cash to Zhang on the same day". It is identified that "asking for money" and "paying cash" have a causal corresponding relationship, and combined with the fact that the subjects in the two pieces of evidence are Zhang and Li and the time is within two days, it is determined that the two pieces of evidence are closely related.
[0041] In some embodiments, the structured decomposition of the unified feature representation comprises: performing feature channel division on the unified feature representation of the evidence, and separating a semantic channel for representing action content, a time channel for representing time description, and a role channel for representing subject attribute from the unified feature representation respectively; Performing mode clustering on the feature distribution in each channel to obtain independent representations of event features, time features, and subject features.
[0042] In actual implementation, first, the semantic information, time expression information, and subject-related information in the unified feature representation are allocated to different internal feature channels, so that different types of evidence information are not mixed. Then, the evidence features are classified in each channel respectively, and the evidence patterns with similar time, similar behavior description, or similar subject are identified through clustering, so that different dimensional information of the evidence can be expressed in an independent manner.
[0043] In a contract dispute case, for example, different evidence may contain behavior descriptions such as "signing a contract", "fulfilling a contract", and "terminating a contract", which are clustered as the same action pattern; and time information such as "March 2023" and "the day of signing" is clustered as a time feature pattern; and subject information such as "Company A", "Company B", and "Zhang (legal person)" is clustered as a subject pattern.
[0044] In some embodiments, the feature learning on the constructed graph structure comprises: performing multiple rounds of message passing on the feature vectors of the nodes in the graph structure and the connection relationship between the nodes, weighting and aggregating the features of adjacent nodes according to the weight of the connection relationship, and combining and updating the features of the current node; When the node features can reflect both their own attributes and the association information of adjacent nodes after multiple iterations, the node representation vector that can represent the global association of the evidence is obtained; In the message passing and feature updating process of evidence grouping, the feature update of node v in the lth iteration satisfies the following formula:
[0045] Wherein, represents the feature vector of node v in the lth iteration; represents the updated feature vector of node v in the (l+1)th iteration; represents the set of adjacent nodes of node v. represents the eigenvector of the adjacent node u at the lth iteration; represents the edge feature of the connection edge between node u and node v, used to represent the association type between evidences; is a linear transformation matrix used to update the node's own feature; is a linear transformation matrix used to process the edge feature; is the connection weight of node u and node v, used to reflect the association strength between different evidences; is a nonlinear activation function used to perform nonlinear mapping on the linearly transformed features.
[0046] When performing feature learning, the association information between nodes is propagated layer by layer in the evidence relationship graph, so that each node gradually integrates the peripheral evidence information related to it after multiple iterations. For example, if multiple evidences are related to a certain transaction, the connection relationship between these evidences will enhance their feature correlation through the message passing mechanism. After multiple updates, each evidence node obtains a comprehensive vector representation that better reflects the overall evidence structure of the case.
[0047] For example, three evidences describe "negotiation begins", "both parties confirm the delivery amount", and "finally complete the payment" respectively. After multiple rounds of graph propagation, the continuous relationship between the three evidences can be captured, so that the final node vector can reflect the complete behavior chain.
[0048] In some embodiments, generating a preliminary evidence grouping result according to the evidence grouping representation vectors includes: calculating the mutual distance of all evidence corresponding representation vectors, identifying the clustering area of the representation vectors according to the distance distribution, and determining the clustering center to which multiple evidences can be grouped; The evidence grouping divides the evidences into multiple preliminary groupings according to the corresponding relationship between the evidence representation vectors and the evidence grouping clustering centers.
[0049] When performing grouping, the distance of all evidence representation vectors is calculated to determine the similarity between evidences. Evidences with close distances are considered to belong to the same evidence mode area, and further form a clustering center. Finally, each evidence is assigned to different preliminary groupings according to the distance relationship with different clustering centers.
[0050] For example, a case has multiple evidences describing the same transfer process, including "apply for transfer", "submit credentials", and "receive fund confirmation". According to the vector similarity of these evidences, they are classified into the same preliminary grouping, while unrelated evidences such as "contract signing photo sorting instructions" are classified into another grouping.
[0051] In some embodiments, the iterative optimization of the preliminary evidence grouping result comprises: calculating the intra-group similarity of each evidence in the current grouping and the inter-group difference with other groupings, and identifying the evidence or grouping that needs to be adjusted according to the reduction of the intra-group similarity or the decrease of the inter-group difference; The evidence grouping updates the grouping attribution or grouping boundary of the evidence grouping evidence in each iteration until the grouping result meets the preset stability condition.
[0052] In some embodiments, the adjustment of the grouping structure by the evidence grouping preset strategy comprises: determining the corresponding strategy instruction according to the intra-group similarity of each evidence in the current grouping, the boundary distance with adjacent groupings, and the structural balance within the grouping; The evidence grouping strategy instruction comprises merging adjacent groupings, splitting the grouping, adjusting the evidence from the current grouping to other groupings, or creating a new grouping for the isolated evidence; The evidence grouping updates the grouping structure according to the evidence grouping strategy instruction in each iteration; The reasoning analysis of the relationship between the evidence based on the optimized grouping comprises: generating an association path according to the connection relationship between the evidence in the optimized grouping; The time sequence, causal direction, and content relevance in the association path are reasoned step by step to construct a reasoning result reflecting the evidence chain, and the evidence grouping extracts interpretation information that can represent the logic of the evidence grouping according to the evidence grouping reasoning result.
[0053] When performing grouping optimization, it is continuously evaluated whether the current grouping structure can accurately reflect the real logical relationship between the evidence. When the similarity of a certain evidence in the current grouping is low, the system attempts to reassign it to a grouping with higher correlation. If the evidence topics within a grouping are mixed, the system can split the grouping. If multiple groupings have highly consistent content, a merging operation can be performed. Through continuous adjustment and optimization, the system enables the final grouping structure to stably describe the logical chain of evidence within the case. In the reasoning phase, the system gradually identifies the evidence chain from the time sequence and causal sequence directions based on the optimized evidence relationship, generating a clear behavior sequence and logical chain.
[0054] In a case of embezzlement, a group of evidence contains both records of "accounting discovery of account anomalies" and content of "suspect taking away funds", and the two topics are inconsistent. During the iteration process, the system moves the "accounting anomaly" evidence to another grouping. Finally, the system infers the evidence chain of "taking away funds → accounting anomaly → tracing the flow of funds" according to the time sequence, making the logical order of the key evidence of the entire case more clear.
[0055] As Figure 2As shown, in some embodiments, the litigation case-oriented evidence grouping method can also include a multi-modal evidence preprocessing and structured modeling step, which is used to provide a richer input basis for the subsequent unified feature representation generation and graph structure construction. Specifically, the system first performs a multi-modal evidence input step to receive different forms of evidence data from the case management system, the electronic file system, or external storage media. The evidence data can at least include text evidence (contract documents, interrogation records, witness testimonies, legal opinions, etc.), image evidence (on-site photos, physical evidence pictures, monitoring screenshots, etc.), audio evidence (recording files, call record collation audio, on-site recordings, etc.), and video evidence (monitoring videos, law enforcement record videos, court trial videos, etc.). The system performs unified numbering and index management on evidence of different sources and different formats when accessed, so as to facilitate subsequent correlation analysis and backtracking.
[0056] In some embodiments, after completing the evidence access, the system performs type identification processing on the evidence. Type identification can be completed based on file format, metadata information, and content features, for example, distinguishing basic categories such as text, image, audio, and video according to file extension, MIME type, and file header information; for mixed content files, further determination can be made as to whether they contain pictures, sound tracks, or subtitle information through sampling decoding and content detection. Through the above type identification step, the system can assign corresponding feature extraction processes and parameter configurations to different categories of evidence, providing a basis for subsequent multi-modal feature extraction.
[0057] In some embodiments, the system performs multi-modal feature extraction on different categories of evidence. For text evidence, natural language processing procedures are used to perform word segmentation, syntax analysis, and entity recognition on the text, and extract text features such as action verbs, subject names, time expressions, place descriptions, and semantic relationships related to the case. For image evidence, target detection and scene recognition methods can be used to identify key targets such as people, objects, vehicles, and bills, and call optical character recognition when needed to convert the text content in the picture into parseable text features. For audio evidence, speech recognition can be used to transcribe speech content into text, while extracting speaker features, tone changes, and background environment sound features to assist in identifying the speaking subject and scene attributes. For video evidence, image analysis and text recognition can be performed on key frames, while combining time axis to extract shot change information and event duration and other timing features, thereby obtaining multi-dimensional feature representation corresponding to the video behavior process.
[0058] In some embodiments, the system performs a multi-modal fusion process after completing different modal feature extraction, aligns and integrates the features obtained from text, image, audio, video and time series analysis according to the evidence number and timeline. During the fusion process, multiple evidence features corresponding to the same event or the same forensic behavior can be aggregated based on timestamps, case scenario labels or scene labels, and features with high semantic correlation but different sources can be combined into a unified representation vector, thereby forming a multi-modal fusion result corresponding to the "unified feature representation for representing evidence information" in step S1, and providing more comprehensive basic information for subsequent fact element recognition and correlation construction.
[0059] In some embodiments, the system performs structured modeling on the evidence information based on the feature representation after multi-modal fusion. Specifically, the system can classify the identified fact elements, spatio-temporal elements, relationship elements and metadata information in each evidence into a unified data structure according to a preset evidence structure template. Fact element extraction is used to extract the behavior type, behavior object and behavior result involved in the evidence; spatio-temporal information extraction is used to extract the event occurrence time, evidence collection time, place description and spatial position clues; relationship information extraction is used to identify the subject relationship, behavior before and after relationship, and proof or counterproof relationship between different evidences; metadata generation is used to record the evidence source, formation process, carrier type and integrity mark attributes. Through the above structured modeling, the evidence information originally existing in natural language or unstructured form is converted into a node and attribute set that can be directly used in a graph structure.
[0060] In some embodiments, the system can construct an evidence knowledge graph based on the above structured evidence information. The nodes in the knowledge graph not only include each piece of evidence itself, but also include entity nodes such as involved subjects, key events and important items; edges are used to represent the pointing relationship between evidence and entities, the mutual confirmation or contradiction relationship between evidences, and the correlation relationship between different entities. When constructing the knowledge graph, the system can automatically generate nodes and edges according to fact elements, spatio-temporal elements and relationship elements, and attach weights, labels and time attributes to the nodes and edges, so that the entire graph can reflect the global view of the evidence structure in the case. The evidence knowledge graph can be used as the input of the subsequent graph structure feature learning, evidence grouping and reasoning analysis steps, so that the method of the present application has better expandability and adaptability in complex cases and multi-source evidence scenarios.
[0061] In some embodiments, the present application provides an evidence grouping system for litigation cases, which is used to implement the evidence grouping method for litigation cases. The system comprises: an evidence feature representation generation unit, configured to obtain multiple types of evidence in a case, perform content analysis on the evidence, and generate a unified feature representation for representing evidence information; The fact element identification unit is configured to perform structural decomposition on the uniform feature representation of the evidence group, identify event features for representing event actions, time features for representing occurrence times, and subject features for representing subject relationships involved in the event; The relationship construction unit is configured to determine fact correlations, time sequence correlations and content correlations between the evidences according to the event features, the time features and the subject features of the evidence group, and construct a graph structure for representing the relationships between the evidences according to the correlations; The graph structure feature learning unit is configured to perform multiple rounds of message passing, weighted aggregation and feature updating on the feature vectors of the nodes in the graph structure of the evidence group and the connection relationships between the nodes, to obtain node representation vectors capable of representing the correlation between the evidences; The preliminary grouping generation unit is configured to perform distance calculation, gathered region identification and gathered center determination according to the node representation vectors of the evidence group, to generate a preliminary evidence grouping result; The grouping iteration optimization unit is configured to calculate the intra-group similarity and the inter-group difference of the preliminary evidence grouping of the evidence group, adjust the grouping attribution or the grouping boundary of the evidence group according to a preset strategy, and obtain an optimized evidence grouping; The reasoning analysis unit is configured to generate a correlation path according to the connection relationships between the evidences based on the optimized evidence grouping, perform step-by-step reasoning on the time sequence, the cause-effect direction and the content correlation in the correlation path, and obtain explanation information for representing the basis of the evidence grouping.
[0062] In some embodiments, the evidence analysis can be extended to unified processing of multi-modal evidences. The system can not only receive text evidences such as transcripts and legal documents, but also receive image evidences such as scanned copies and on-site photos, and audio-visual materials such as audio recordings and audio transcription texts. For image evidences, the text and structural information in the pictures can be converted into analyzable texts through text recognition and layout analysis. For audio and video evidences, the dialogue content can be organized into time-labeled text segments through speech recognition and speech transcription. Through the above processing, different types of evidence information are uniformly mapped to the same feature space, which facilitates subsequent fact element extraction and correlation analysis.
[0063] In some embodiments, the system can construct a dynamic evidence knowledge graph based on the evidence resolution result, extract the elements such as facts, characters, time, place, and physical evidence appearing in the evidence, and identify various legal semantic relationships such as causal relationship, chronological relationship, proof relationship, and contradictory relationship. In the process of constructing the knowledge graph, the system not only generates corresponding nodes for each piece of evidence, but also separately establishes nodes and edges for important entities and relationships, so that the implicit connection between evidences can be discovered through multiple paths. With the addition of new evidence or the modification of original evidence in the case handling process, the knowledge graph can be incrementally updated according to the version, and the historical versions are preserved for comparative analysis of the evidence structure changes at different stages.
[0064] In some embodiments, the system can access a legal knowledge base or a law article knowledge graph, and associate the behavior types involved in the evidence with the corresponding legal concepts and legal provisions while constructing the evidence relationship. In this way, the evidence nodes not only carry the description at the fact level in the graph structure, but also carry the possible legal evaluation information, providing a more legal practice-oriented semantic background for subsequent grouping optimization and reasoning analysis. For example, evidence related to behaviors such as “transfer” “receive payment” “sign a contract” can be associated with corresponding contract law or criminal law provision nodes in the graph, so that the system can consider the key behavior links in the legal sense when analyzing the evidence chain.
[0065] In some embodiments, the system can combine the user's analysis intent to guide evidence screening and grouping. When using the system, the case handling personnel can express the current focus point through search conditions, problem descriptions, or natural language inputs, such as “evidence around the first fund flow” “communication related to a certain suspect”. The system performs semantic understanding on the user input based on the existing evidence knowledge graph and graph structure representation, and locates the most relevant evidence subgraph and evidence grouping in the graph according to the intent, and preferentially performs feature learning and grouping optimization on the evidence in the subgraph, thereby realizing the intent-oriented evidence grouping process.
[0066] In some embodiments, the system can carry out adversarial reasoning analysis from the perspectives of both parties based on the evidence relationship graph. Specifically, the system can construct an evidence chain that “supports the establishment of a certain key fact” and an evidence chain that “questions or weakens the fact”, mark the two types of evidence as different stands in the graph, and mark the conflict points, broken points, and weak links between the chains. In this way, the system can assist in identifying the contradictory relationship between the evidences and the gaps in the evidence chain while grouping the evidences, providing structured basis for debate strategy formulation, evidence presentation and evidence preparation.
[0067] In some embodiments, the system can provide cross-evidence correlation analysis and knowledge discovery capabilities. After grouping and graph structure construction of the evidence, the system can compare the evidence knowledge graphs in different cases and different case files, and identify frequently co-occurring behavior patterns, similar evidence structures, or high-risk event patterns. For example, when similar transaction paths, signing patterns, or fund flow patterns repeatedly appear in multiple cases, the system can provide prompts through cross-case correlation analysis to assist investigators in discovering potential associated case clues or habitual crime patterns.
[0068] In some embodiments, the system can provide hierarchical display and collaborative analysis capabilities for different roles. For professional users such as judges, prosecutors, and lawyers, the system can provide professional interfaces containing evidence relationship graphs, evidence chain views, and grouping details, supporting viewing according to time axis, subject relationship, or controversy dimensions; for parties or non-professional participants, the system can provide simplified evidence grouping display and textual descriptions, highlighting key evidence and core nodes. The viewing results of the same case by different roles can be based on the same grouping and knowledge graph, only the presentation method and information granularity are different, facilitating collaborative analysis.
[0069] In some embodiments, the system can introduce a decision optimization model in the grouping iteration optimization process, taking the current evidence grouping state as the environment state, and merging groups, splitting groups, adjusting single evidence attribution, etc. as optional actions, and by setting reward signals related to group consistency, inter-group distinction, evidence chain integrity, and conflict annotation rationality, the system can optimize the grouping strategy in the long term. The system can use historical case data and artificially annotated high-quality grouping results for strategy training, so that it can more accurately perform adjustment operations in new cases and improve the reliability of automatic grouping.
[0070] In some embodiments, the system can record the interactive operations and feedback results of investigators during use, such as manual grouping adjustment of certain evidence, confirmation or negation of part of the grouping results, etc., as samples for the system model to continuously learn. The system can periodically update the parameters of feature extraction, correlation identification, and grouping strategy according to these feedbacks, so that the evidence grouping behavior in subsequent cases is more consistent with the experience and preferences of investigators, achieving individualized adaptive optimization for organizations or institutions.
[0071] In some embodiments, the system can provide explanatory information for the evidence grouping results, not only a general description of the grouping logic, but also a breakdown from feature and relationship dimensions. The system can identify features that significantly influence the inclusion of a piece of evidence in the current group, such as key behavioral descriptions, adjacent time points, common participating entities, or shared transaction targets, and list several typical evidence pairs supporting the grouping decision. Furthermore, the system can generate a brief textual description for each evidence group, summarizing the factual events or behavioral chain corresponding to that group, facilitating investigators' quick understanding of the grouping's meaning.
[0072] In some embodiments, the method and system can be deployed in various application environments such as law firms, court information platforms, procuratorial case-handling systems, and corporate legal management platforms. The system can interface with existing case management systems to automatically import evidence materials from case files and provide evidence grouping, evidence chain display, and group interpretation functions as a service. For different types of cases, such as criminal cases, civil and commercial cases, and administrative dispute cases, the system can differentiate the parameters for feature extraction and association identification based on preset case type templates to adapt to the characteristics of evidence structure under different legal relationships and standards of proof.
[0073] In some embodiments, the system may include a visual grouping display function, presenting the evidence grouping results graphically. Specifically, the system can display evidence and its relationships on the interface in the form of nodes and connections, and use differences in color, shape, or layout to distinguish different groups and types of evidence. Users can also click on a node to view the original evidence text, a summary of key information, and its position in the chain of evidence. In this way, users can intuitively grasp the evidence structure of the case and the content composition of each group, accelerating their understanding of the factual framework of complex cases.
[0074] like Figure 3 As shown, in one embodiment, the evidence grouping system for litigation cases adopts a layered architecture, comprising a data layer, a knowledge graph layer, a computation layer, and an application layer from bottom to top. Each layer interacts with other layers through pre-defined interfaces to exchange data and control commands, supporting the entire process of evidence collection, preprocessing, relationship modeling, grouping calculation, and result display. In this embodiment, the system can be deployed in a dedicated server environment of a court, procuratorate, or law firm, or it can be deployed in a private cloud within an organization's internal network, accessible via a browser or dedicated client.
[0075] In the data layer, the system includes a multi-modal evidence access module, an evidence preprocessing module, and an evidence verification module. The multi-modal evidence access module is used to receive text evidence, image evidence, audio evidence, and video evidence from a case management system, an electronic file system, or external storage, and uniformly encapsulate evidence of different sources and different formats into internal data objects; the evidence preprocessing module performs operations such as format standardization, quality inspection, deduplication processing, and metadata extraction on the received evidence, for example, performs character recognition on scanned copies, transcribes audio and video, and supplements information such as time, source, and file number for evidence with missing metadata; the evidence verification module performs basic checks on the authenticity, integrity, and legality of the evidence, for example, verifies the signature or hash value of an electronic file, and marks evidence that cannot pass the verification, so as to be distinguished or reduced in weight in subsequent grouping and reasoning analysis.
[0076] In the knowledge graph layer, the system includes an evidence graph storage module, a graph evolution management module, and a reasoning chain construction module. The evidence graph storage module can be implemented based on a graph database cluster, stores event nodes, subject nodes, time nodes, and evidence nodes extracted in a case, and saves edge information such as causal relationship, time sequence relationship, proof relationship, and conflict relationship between nodes, while supporting index optimization to improve query efficiency. The graph evolution management module is used to perform incremental update and version control on the evidence graph when new evidence is imported, old evidence is revised, or the case progresses, to maintain the consistency and traceability of the graph. The reasoning chain construction module automatically generates an evidence chain describing a disputed fact or a key behavior process based on the node and edge information in the evidence graph, links closely related evidence nodes into a directed path, and provides structured input for the adversarial analysis and grouping decision of the computing layer.
[0077] In the computing layer, the system includes a graph retrieval and intelligent reasoning engine, an adversarial analysis engine, and a grouping decision engine. The graph retrieval and intelligent reasoning engine is used to perform graph structure-based retrieval and reasoning on the evidence graph, combined with graph structure feature learning results, to support subgraph extraction and evidence chain completion according to case points of contention, behavior types, or subject relationships, and to return the most relevant evidence set and its relationship structure according to user input questions or analysis intentions. The adversarial analysis engine is used to analyze evidence from both positive and negative perspectives, to model evidence chains supporting a fact and evidence chains questioning or weakening the fact in parallel, to identify contradictions between different evidences in time, content, or source through conflict detection, and to propose possible coping strategies or focus points in combination with a strategy generation module. The grouping decision engine performs clustering operations and quality evaluation on the evidence representation vectors from the graph structure feature learning, generates preliminary evidence grouping results according to the similarity within the group, the difference between groups, and the preset grouping strategy rules, and performs grouping merging, splitting, and evidence migration operations in the iteration process until a stable optimized evidence grouping is formed, while the quality of the evidence chain structure corresponding to each grouping is evaluated and marked.
[0078] In the application layer, the system includes a user interface system, a permission management system, and a visual grouping panel. The user interface system can provide evidence import, retrieval, grouping viewing, and reasoning result browsing functions to case handling personnel through a Web interface, a mobile interface, or an open API interface, enabling users to complete the organization and analysis of case evidence in a unified interface. The permission management system is used for identity authentication, role management, and access control of different role users, ensuring that only judges, prosecutors, lawyers, or auxiliary personnel with corresponding permissions can access the evidence grouping results and graph information of the corresponding case. The visual grouping panel is used to display evidence grouping results and evidence relationship graphs in a graphical manner, distinguishing different groupings by color or area, and laying out evidence nodes on a time axis, subject relationship axis, or contention dimension, and supporting users to click any evidence node to view the original evidence content, the evidence chain it belongs to, and its position in the grouping, thereby facilitating case handling personnel to intuitively understand and deeply analyze the evidence structure of complex cases in actual use.
[0079] In some embodiments, the present application provides a terminal, comprising: a memory for storing an evidence grouping program for litigation cases; a processor for executing the steps of the evidence grouping method for litigation cases when the evidence grouping system for litigation cases is executed.
[0080] In some embodiments, the present application provides a computer readable storage medium storing computer instructions, when a computer reads the computer instructions in the storage medium, the computer executes the evidence grouping method for litigation cases.
[0081] The above description is merely one or more embodiments of the present disclosure, and is not intended to limit the one or more embodiments of the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the one or more embodiments of the present disclosure shall be included in the protection scope of the one or more embodiments of the present disclosure.
Claims
1. A method for grouping evidence for a litigation case, the method comprising: The method comprises the following steps: obtaining multiple types of evidence in a case, performing content analysis on the evidence to generate a unified feature representation for representing evidence information; identifying fact elements and related information involved in the evidence from the obtained unified feature representation, and constructing a graph structure for representing the relationship between the evidence; performing feature learning on the constructed graph structure to obtain a representation vector capable of representing the correlation between the evidence, and generating a preliminary evidence grouping result according to the evidence grouping representation vector; iteratively optimizing the preliminary evidence grouping result, adjusting the grouping structure according to a preset strategy, and obtaining an optimized evidence grouping; based on the optimized grouping, performing reasoning analysis on the relationship between the evidence to obtain explanation information for representing the basis of the evidence grouping.
2. The litigation case oriented evidence grouping method of claim 1, wherein, The identification of fact elements and related information involved in the evidence from the obtained unified feature representation comprises: performing structural decomposition on the unified feature representation to obtain event features for representing event actions, time features for representing occurrence times, and subject features for representing involved subjects; determining fact correlation by comparing the causal direction relationship of the event features of different evidence, determining time sequence correlation by performing time sequence calculation on the time features of different evidence, and determining content correlation by performing similarity calculation on the subject features, behavior semantic features and content similarity features.
3. The litigation case oriented evidence grouping method of claim 2, wherein, The structural decomposition on the unified feature representation comprises: dividing the unified feature representation of the evidence into feature channels, separating the semantic channel for representing action content, the time sequence channel for representing time description, and the role channel for representing subject attributes from the unified feature representation; and performing pattern clustering on the feature distribution in each channel to obtain independent representations of the event features, the time features and the subject features. The feature learning on the constructed graph structure comprises: performing multi-round message passing on the feature vectors of the nodes in the graph structure and the connection relationship between the nodes, weighting and aggregating the features of adjacent nodes according to the weight of the connection relationship, and combining and updating the features with the features of the current node; 4. The litigation case oriented evidence grouping method of claim 1, wherein, when the node features can reflect both their own attributes and the correlation information of adjacent nodes after multiple rounds of iteration, the node representation vector capable of representing the global correlation of the evidence is obtained; generating a preliminary evidence grouping result according to the evidence grouping representation vector comprises: calculating the mutual distance of the representation vectors corresponding to all evidence, identifying the gathering area of the representation vectors according to the distance distribution, and determining the gathering center to which multiple evidence can be grouped; In the evidence grouping message passing and feature update process, the feature update of node v in the tth iteration satisfies the following equation: xt+1= xt+ 1 2∑y∈N(v) (xt+1- y) wherein, denotes the feature vector of node v at the lth iteration; denotes the updated feature vector of node v at the (l+1)th iteration; denotes the set of nodes adjacent to node v; denotes the feature vector of adjacent node u at the lth iteration; denotes the edge feature of the connecting edge between node u and node v, used to represent the type of association between evidences; is a linear transformation matrix used to update the feature of a node itself; is a linear transformation matrix used to process the edge feature; is the connection weight between node u and node v, used to reflect the strength of association between different evidences; is a nonlinear activation function used to perform nonlinear mapping on the linearly transformed features.
5. The litigation case oriented evidence grouping method of claim 4, wherein, the evidence grouping divides the evidence into multiple preliminary groupings according to the corresponding relationship between the evidence representation vectors and the evidence grouping gathering centers. The iterative optimization of the preliminary evidence grouping result comprises: calculating the intra-group similarity of each evidence in the current grouping and the inter-group difference with other groupings, and identifying the evidence or grouping that needs to be adjusted according to the reduction of the intra-group similarity or the decrease of the inter-group difference; 6. The litigation case oriented evidence grouping method of claim 5, wherein, the evidence grouping updates the grouping attribution or grouping boundary of the evidence grouping evidence in each iteration until the grouping result meets the preset stability condition. 7. The litigation case oriented evidence grouping method of claim 1, wherein, The preset strategy for evidence grouping adjusts the grouping structure, including determining a corresponding strategy instruction according to the in-group similarity of each piece of evidence in the current group, the distance from the adjacent group boundary, and the structural balance within the group; The evidence grouping strategy instruction includes merging adjacent groups, splitting the group, adjusting the evidence from the current group to other groups, or creating a new group for isolated evidence; The evidence grouping updates the grouping structure according to the evidence grouping strategy instruction in each iteration; The reasoning analysis based on the optimized grouping includes generating a correlation path according to the connection relationship between the evidences in the optimized grouping; The time sequence, causal direction, and content relevance in the correlation path are reasoned level by level to construct a reasoning result reflecting the evidence chain, and the evidence grouping extracts interpretation information that can represent the logic of the evidence grouping according to the evidence grouping reasoning result.
8. A litigation case-oriented evidence grouping system for implementing the litigation case-oriented evidence grouping method as claimed in claim 1, characterized by, The system includes: An evidence feature representation generation unit is configured to obtain multiple types of evidence in a case, perform content analysis on the evidence, and generate a unified feature representation for representing evidence information; A fact element identification unit is configured to structurally decompose the unified feature representation of the evidence grouping, identify event features representing event actions, time features representing occurrence times, and subject features representing subject relationships; A relationship construction unit is configured to determine fact correlations, time sequence correlations, and content correlations between the evidences according to the evidence grouping event features, time features, and subject features, and construct a graph structure for representing the relationships between the evidences; A graph structure feature learning unit is configured to perform multiple rounds of message passing, weighted aggregation, and feature updating on the feature vectors of the nodes in the evidence grouping graph structure and the connection relationships between the nodes to obtain node representation vectors that can represent the correlations between the evidences; A preliminary grouping generation unit is configured to perform distance calculation, cluster area identification, and cluster center determination based on the evidence grouping node representation vectors to generate a preliminary evidence grouping result; A grouping iteration optimization unit is configured to calculate the in-group similarity and inter-group difference of the preliminary evidence grouping of the evidence grouping, adjust the grouping attribution or grouping boundary of the evidence according to a preset strategy, and obtain an optimized evidence grouping; A reasoning analysis unit is configured to generate a correlation path according to the connection relationship between the evidences based on the optimized evidence grouping, reason the time sequence, causal direction, and content relevance in the correlation path level by level, and obtain interpretation information representing the basis of the evidence grouping.
9. A terminal, characterized by comprising: It includes: A memory is configured to store an evidence grouping program for litigation cases; A processor is configured to execute the evidence grouping program for litigation cases to implement the steps of the evidence grouping method for litigation cases as claimed in claim 1.
10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions, and when a computer reads the computer instructions in the storage medium, the computer executes the evidence grouping method for litigation cases as claimed in claim 1.