A method and system for online mediation

Abstract

The application discloses a method and system for online mediation, and relates to data processing, comprising: receiving case data through an online system, wherein the case data comprises basic information of parties, dispute facts and related evidence; performing natural language processing and feature extraction on the case data to obtain a set of dispute elements; performing conditional relationship judgment on the set of dispute elements through causal relationship analysis to obtain a first determination result and a second determination result, wherein the first determination result represents technical determination facts, and the second determination result represents legal determination facts; generating a mediation report according to the first determination result and the second determination result; and performing online mediation according to the mediation report. In view of the inaccuracy of responsibility determination in online mediation, the application improves the accuracy of mediation responsibility determination by analyzing technical responsibility and legal responsibility, etc. through reasoning algorithms and similarity matching.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method and system for online mediation. Background Technology

[0002] The existing methods for handling civil and commercial disputes include negotiation, people's mediation, administrative mediation, and litigation. However, these four methods are independent of each other and lack coordination, leading many parties to ultimately choose litigation, which increases the burden on the judiciary.

[0003] Current online mediation systems face the following technical challenges when handling complex civil and commercial disputes:

[0004] First, it is impossible to accurately distinguish and identify technical and legal liabilities. The existing system lacks a mechanism for classifying and identifying technical and legal factors in disputed facts, which leads to the inability to accurately determine the causal relationship at the technical level and the legal level in civil and commercial disputes involving technical issues (such as medical malpractice, engineering quality, product liability, etc.), resulting in confusion and ambiguity in liability determination.

[0005] Second, there is a lack of scientific methods for analyzing causal relationships. Traditional mediation mainly relies on experience-based judgment, lacking quantitative analysis based on objective data and technical standards for the determination of technical facts; and lacking standardized matching based on judicial precedents and legal provisions for the determination of legal facts, resulting in strong subjectivity and poor consistency in liability determination.

[0006] Third, there is a disconnect between technical assessment and legal mediation. In practice, technical assessment often focuses on determining "whether a technical problem exists," while legal mediation focuses on "what responsibility should be borne," and the two lack effective integration. Especially for mixed disputes, which involve both technical judgments about whether technical parameters meet standards and the determination of the legal causal relationship between the behavior and the damage, the existing system cannot handle these two types of issues simultaneously.

[0007] Therefore, there is an urgent need for an online mediation method that can identify technical causality and legal causality separately, and conduct quantitative analysis of liability through scientific algorithms, in order to improve the accuracy of liability determination in mediation and the acceptance by the parties involved. Summary of the Invention

[0008] To address the inaccuracy in determining liability in online mediation, this application uses reasoning algorithms and similarity matching to analyze technical and legal liabilities, thereby improving the accuracy of liability determination in mediation.

[0009] One aspect of this application provides a method for online mediation, comprising: S1, receiving case data through an online system, the case data including basic information of the parties, disputed facts, and relevant evidence; S2, performing natural language processing and feature extraction on the case data to obtain a set of disputed elements; S3, performing conditional relationship judgment on the set of disputed elements through causal relationship analysis to obtain a first determination result and a second determination result, the first determination result representing technically determined facts, and the second determination result representing legally determined facts; S4, generating a mediation report based on the first determination result and the second determination result; and S5, conducting online mediation based on the mediation report.

[0010] Furthermore, S2 performs natural language processing and feature extraction on the case data to obtain a set of disputed elements, including: performing multimodal recognition and preprocessing on the case data to obtain structured data containing text data, image data, and tabular data; based on the structured data, extracting entities containing names, places, times, amounts, and events through named entity recognition algorithms, and marking the original evidence sources of each entity; based on the structured data, identifying semantic information containing subject, object, causal relationship words, and liability keywords in the disputed facts through semantic analysis; and performing association mapping between the extracted entities and semantic information based on a pre-constructed knowledge graph of civil and commercial disputes.

[0011] Based on causal relationship terms, the extracted entities are classified into causal elements according to time sequence, resulting in causal behavior elements, conditional elements, and outcome elements. Specifically, the time sequence is as follows: time axis: t1→t2→t3→...→tn; event chain: event A→event B→event C→...→outcome event. Causal behavior elements are entities that occur earliest in time and are marked as "cause" by causal relationship terms; conditional elements are entities that are in the middle in time and are marked as "condition" or "influencing factor"; outcome elements are entities that occur latest in time and are marked as "outcome" or "consequence" by causal relationship terms. Classification rule: IF timestamp of entity E1 < timestamp of entity E2 AND there is a causal relationship term connection (E1, E2), THENE1∈causal behavior elements, E2∈outcome elements.

[0012] Based on the association mapping between entities and semantic information, the causal behavioral elements, conditional elements, and result elements are labeled as technical and legal categories, respectively; the labeled causal behavioral elements, conditional elements, and result elements are then used as a set of disputed elements.

[0013] Named Entity Recognition (NER) algorithms are a fundamental technology in natural language processing, used to automatically identify and classify entities with specific meanings from unstructured text. An actor refers to a legal entity that performs a specific act or assumes a specific obligation in a civil or commercial dispute, including natural persons, legal persons, and other organizations. The object of the act refers to the target of the actor's actions, including goods, rights, interests, or other legal objects.

[0014] Causal relation vocabulary: Language markers that indicate the logical relationship between cause and effect, used to identify causal links between events; in this application, direct causal words: because, due to, leading to, causing, resulting, producing, etc.; indirect causal words: prompting, influencing, related, relevant, involving, etc.; negative causal words: unrelated to, not the cause of, excluding the influence of, etc.

[0015] Keywords related to liability: Professional terms indicating the attribution, manner, and degree of legal liability. In this application, terms related to the degree of liability include: full liability, primary liability, secondary liability, minor liability, etc.; terms related to the type of liability include: liability for breach of contract, tort liability, liability for pre-contractual negligence, etc.; terms related to the manner of liability include: compensation for losses, payment of liquidated damages, restoration to the original state, and apology, etc.

[0016] Furthermore, in S3, through causal relationship analysis, conditional relationships are judged on the set of disputed elements to obtain a first determination result and a second determination result. The first determination result represents the technical determination fact, and the second determination result represents the legal determination fact. This includes: extracting causal behavioral elements, conditional elements, and result elements marked as technical and legal from the set of disputed elements respectively, forming a technical causal relationship chain and a legal causal relationship chain; verifying the conditional relationships using a reasoning algorithm based on the technical causal relationship chain to obtain a technical causal relationship strength index; verifying the conditional relationships using a similarity matching algorithm based on the legal causal relationship strength index; performing a technical standard retrieval using a matching algorithm based on the technical causal relationship strength index and a pre-set technical specification library to obtain a technical determination result, which serves as the first determination result; and performing a legal retrieval using a matching algorithm based on the legal causal relationship strength index and a pre-set legal provision library to obtain a legal determination result, which serves as the second determination result.

[0017] Among them, conditional relationship judgment refers to the process of analyzing whether there is a conditional dependency relationship between disputed elements through algorithms, that is, judging whether a certain element is a necessary or sufficient condition for the occurrence of another element. In this application, the judgment logic is as follows: Necessary condition judgment: verifying "without A, there is no R", that is, whether the result element R will not occur when the causal element A does not exist; Sufficient condition judgment: verifying "with A, there is R", that is, whether the result element R will necessarily occur when the causal element A exists; Influence of conditional element: judging the moderating effect of conditional element C on the causal relationship A→R. Judgment results are classified as follows: Strong conditional relationship: A is a necessary and sufficient condition for R; Medium conditional relationship: A is one of the necessary or sufficient conditions for R; Weak conditional relationship: A has an influence on R but is not a critical condition; No conditional relationship: There is no conditional dependency between A and R.

[0018] Verifying conditional relationships using inference algorithms is a technique specifically designed for technical causal chains. Based on the fundamental cause-and-effect logic of "no A, no R," it employs probabilistic inference algorithms for objective verification. The logical basis is the "no A, no R" proof-by-contradiction logic, which confirms causal relationships by verifying whether the result does not occur when the cause is missing. This is based on Bayesian probabilistic inference. At that time, A was considered a necessary condition for R. In this application, the verification process involves constructing a technical parameter probability model: P(technical result | technical reason, technical condition); and calculating the counterfactual probability: Calculate the strength of the conditional relationship: Threshold judgment: When ΔP > θ, it is confirmed that a conditional relationship exists.

[0019] Similarity matching algorithms for verifying conditional relationships are a technical process specifically designed for legal causal chains. Based on the legal judgment logic of "proportional causation," it employs similarity matching algorithms for subjective evaluation. In this application, the calculation of similarity includes direct similarity: Sim(LA) i ,LR j ) = cos(LA i ,LR j Conditional weight calculation:

[0020] w k =Sim(LC k ,LA i )×Sim(LC k ,LR j Overall similarity: a weighted similarity that takes into account the influence of conditional factors.

[0021] Furthermore, based on the technological causal relationship chain, an inference algorithm is used to verify the conditional relationship and obtain a technological causal relationship strength index, including: marking the technological cause-behavioral element as node A. i Technical result elements are labeled as node R. jTechnical condition elements are marked as node C. k Construct a directed causal graph G(V,E) for technology; identify the technology parameter V from the behavioral elements of technology-related causes. i The technical parameters include medical technical parameters, engineering technical parameters, product quality parameters, or operating procedure parameters; technical parameter V is retrieved from a preset technical specification database. i The corresponding parameter threshold θ i Construct a technical parameter-threshold mapping set Θ = {(V1,θ1),(V2,θ2),......,(V n ,θ n )};Calculate the technical parameter V respectively i Below the parameter threshold θ i The conditional probability P(R) at that time j |A i <θ i and exceeding the parameter threshold θ i The conditional probability P(R) at that time j |A i ≥θ i ); Calculate the probability change ΔP = P(R) j |A i ≥θ i )-P(R j |A i <θ i ), as an indicator of the strength of technical causal relationships.

[0022] Technical factors in civil and commercial disputes often exhibit a critical threshold phenomenon, meaning that technical parameters only have a significant impact when they reach or exceed a specific threshold, while changes within that threshold have a negligible effect on the outcome. For example: in medical disputes, adverse reactions only occur when drug dosage exceeds a safety threshold; in engineering accidents, structural failure only occurs when material strength falls below the design threshold; in product quality disputes, usage requirements are only met when purity indicators reach a qualified threshold; and in operational accidents, dangerous consequences only arise when the safe distance is insufficient to meet the critical value.

[0023] Existing technologies generally employ linear interpolation methods and continuous probability models for technical causal relationship analysis. However, linear models cannot capture abrupt changes in technical parameters near a threshold, leading to inaccurate causal relationship judgments at critical states. Continuous probability models smooth out threshold boundaries but cannot accurately quantify the differential impact on results before and after crossing the threshold. Therefore, this application uses a threshold θ to define the continuous technical parameter space. i The threshold boundary is divided into two discrete intervals, and the conditional probability within each interval is calculated to accurately capture the threshold boundary effect.

[0024] Furthermore, based on the legal causal relationship chain, a similarity matching algorithm is used to verify the conditional relationship, obtaining a legal causal relationship strength index, including: using a semantic vectorization algorithm, marking legal causal behavioral elements as a legal behavioral vector set LA = {LA1, LA2, ..., LA...}. m Legal outcome elements are tagged as a set of legal outcome vectors LR = {LR1, LR2, ..., LR...} n Legal condition elements are tagged as a set of legal condition vectors LC = {LC1, LC2, ..., LC...} k}; Extract legal parameter L from LA i Legal parameter L i This includes data on subjective fault, objective conduct, damage results, and causal relationships; subjective fault includes the degree of intent and the degree of negligence; objective conduct includes the illegality of the conduct; damage results include the degree and scope of damage; and causal relationships include temporal and spatial correlations.

[0025] According to legal parameter L i By using a pre-set database of judicial precedents, the corresponding judicial determination standard S is retrieved. i Construct legal parameters L i - Judicial Determination Standard S i The mapping set Ψ = {(L1,S1),(L2,S2),......,(L n ,S n For each legal action vector LA i and legal outcome vector LR j Calculate vector similarity Sim(LA) i ,LR j ) = cos(LA i ,LR j Based on the principle of adequate causation, calculate the legal condition vector LC. k The weight w of the similarity k w k =Sim(LC k ,LA i )×Sim(LC k ,LR j According to the influence weight w k and judicial determination standards S i Calculate legal liability α i α i =Σ(S i ×w k ); Calculate the strength index of legal causal relationships

[0026] Semantic vectorization algorithms refer to natural language processing techniques that convert semantic information in legal texts into high-dimensional numerical vector representations, making semantically similar legal concepts appear closer together in the vector space. This application preferably employs a BERT-based pre-trained language model, specifically fine-tuned for the legal domain.

[0027] Temporal and spatial correlation are indicators used in causal relationship data to quantify the degree of correlation between legal elements in the temporal and spatial dimensions, reflecting the spatiotemporal closeness of causal relationships. Temporal correlation: Where t1 and t2 are the timestamps of the causal behavior and the result event, respectively, and λ is the time decay coefficient. The value range of λ is: strong time correlation (0.8-1.0): time interval less than 24 hours; medium time correlation (0.5-0.8): time interval 1 day to 1 month; weak time correlation (0.2-0.5): time interval 1 month to 1 year; no time correlation (0.0-0.2): time interval more than 1 year.

[0028] Spatial correlation: Where distance represents geographical distance, and 'a' is the spatial attenuation coefficient. The value range of 'a' is as follows: strong spatial correlation (0.8-1.0): within the same building or within 100 meters; medium spatial correlation (0.5-0.8): within the same area, 1-10 kilometers apart; weak spatial correlation (0.2-0.5): within the same city, 10-100 kilometers apart; no spatial correlation (0.0-0.2): across cities or more than 100 kilometers apart. The comprehensive causal correlation degree = w1 × temporal correlation degree + w2 × spatial correlation degree, where w1 + w2 = 1, and the weights are dynamically adjusted according to the specific dispute type.

[0029] Judicial Determination Standard S i This refers to authoritative standards for determining legal liability, retrieved from a pre-defined database of judicial precedents and corresponding to specific legal parameters Li, used to guide the quantitative calculation of legal responsibility. In this application, data sources include: guiding cases of the Supreme People's Court, typical precedents from courts at all levels, and judicial interpretations. Fault determination standards: Intentional act standard value 0.9, gross negligence 0.7, ordinary negligence 0.5; Illegality standards: Serious violation 0.8, ordinary violation 0.6, minor violation 0.3; Degree of damage standards: Major damage 0.9, ordinary damage 0.6, minor damage 0.3; Causation standards: Direct causation 0.9, indirect causation 0.6, weak causation 0.2.

[0030] Legal liability α i This refers to the influence weight w k and judicial determination standards S i The calculated quantitative legal liability value reflects the proportion of specific legal parameters in the overall liability assumption.

[0031] In particular, the determination of "proportional causation" is a common challenge in the legal causal relationship determination in civil and commercial disputes. The theory of proportional causation requires judging whether a causal act, under general social norms, would "usually" or "should" lead to a specific harmful result. This judgment involves the mediating influence of legal conditions on the causal relationship between the causal act and the harmful result; that is, the legal conditions must be related to both the causal act and the harmful result to constitute a valid chain of proportional causation.

[0032] Existing technologies generally employ simple binary judgments and subjective expert assessments to analyze causal relationships. Specifically, this includes: calculating only the direct similarity between the causal behavior and the damage result, ignoring the influence of mediating factors; and relying on the subjective experience of legal experts to score and assess causality, lacking a unified quantitative standard.

[0033] Therefore, this application uses the dual similarity weight calculation formula w k =Sim(LC k ,LA i )×Sim(LC k ,LR j ), legal condition vector LC k Respectively related to the legal behavior vector LA i and legal outcome vector LR j Calculating similarity, through product operations, ensures that the conditional elements are strongly correlated with both the causal behavior and the harmful result, reflecting the essential characteristic of "bidirectional association" in adequate causation. When Sim(LC) k ,LA i ) higher and Sim(LC k ,LR j When the product value w is also relatively high, k A significant increase indicates that the conditional element has a strong mediating effect between cause and effect; when any similarity is low, the product value decays rapidly, automatically filtering out weakly correlated interfering elements.

[0034] Furthermore, based on the technical causal relationship strength index and a pre-set technical specification library, a matching algorithm is used to retrieve technical standards and obtain technical identification results, which serve as the first identification result. This includes: classifying the technical causal relationship strength index into three levels—strong correlation, moderate correlation, and weak correlation—based on the magnitude of the probability change ΔP; and classifying it based on the technical parameter V... i and parameter threshold θ i Calculate the deviation Based on the causal relationship level and deviation δ i Construct a technical responsibility determination matrix M; based on matrix M, generate a technical identification result, which serves as the first identification result.

[0035] Furthermore, based on the legal causation strength index and a pre-set legal provisions database, a matching algorithm is used to perform a legal retrieval, obtaining a legal determination result as the second determination result. This includes: classifying the legal causation strength index into three levels—direct causation, indirect causation, and no causation—based on β; and from the legal parameter L... i Extract objective behavioral feature vectors and subjective fault feature vectors to construct a legal behavior feature matrix F = [objective behavioral features; subjective fault features]. Based on the legal behavior feature matrix F, calculate the vector similarity and retrieve the top k most relevant legal provisions from the legal provision database to obtain the set of legal provision numbers and the relevance score. Based on legal responsibility α... i And the causal relationship type label, calculate the proportion of responsibility R: when the causal relationship type is direct, R = α i When the causal relationship is indirect, R = α i ×β; When the causal relationship type is none, R = 0; Based on the liability-bearing ratio R, the type of compensation liability is determined, and a legal determination result is generated as the second determination result.

[0036] In the legal determination of civil and commercial disputes, a common challenge lies in calculating the proportion of liability based on the different types of causal relationships. Legal theory clearly distinguishes between three types of causal relationships: direct causation, indirect causation, and no causal relationship. Different types of causal relationships correspond to different degrees of liability. Direct causation requires full or primary liability, indirect causation requires a reduction in liability based on the degree of mediation, and no liability is incurred in the absence of a causal relationship.

[0037] Existing technologies generally use a uniform liability coefficient and a simple proportional allocation to calculate legal liability, applying the same liability calculation formula to all types of causal relationships without reflecting the differences between types; and using a fixed reduction ratio (such as 0.5 or 0.7) for indirect causal relationships, which lacks relevance to the specific circumstances of the case.

[0038] This application employs differentiated methods for calculating liability ratios based on the type of causal relationship. When the causal relationship is direct: R = α i (Full liability); When the causal relationship is indirect: R = α i ×β (reduced liability); when the causal relationship type is none: R = 0 (no liability).

[0039] Furthermore, S4, based on the first and second determination results, generates a mediation report, including: the level of technical responsibility and deviation δ based on the first determination result. iCalculate the technical responsibility score T1; calculate the legal responsibility score T2 based on the responsibility ratio R and the type of compensation liability in the second determination result; calculate the range of compensation liability amount based on T1 and T2, as well as the amount in dispute in the case; and generate a mediation report based on the range of compensation liability amount and the preset mediation report template.

[0040] Another aspect of this application provides an online mediation system, comprising: a data receiving module for receiving case data through an online system; a feature extraction module for performing natural language processing and feature extraction on the case data to obtain a set of disputed elements; a causal relationship module for determining the causal relationship of the set of disputed elements to obtain a first determination result and a second determination result, including a technical analysis unit and a legal analysis unit; a structure generation module for generating a mediation report based on the first determination result, the second determination result, and a preset legal provisions library; and an online mediation module for conducting online mediation based on the mediation report.

[0041] Furthermore, the technical analysis unit performs reasoning and verification on the disputed elements, and matches technical standards with the technical specification library to generate a technical determination result, which serves as the first determination result; the legal analysis unit performs similarity verification on the disputed elements, and matches legal provisions with a pre-set legal provision library to generate a legal determination result, which serves as the second determination result.

[0042] Compared to existing technologies, the advantages of this application are:

[0043] Technical factors in civil and commercial disputes often exhibit a critical threshold phenomenon, meaning that technical parameters only have a significant impact when they reach or exceed a specific threshold, while changes within that threshold have a negligible effect on the outcome. This application calculates the conditional probability P(Ri) of the technical parameter Vi under different threshold conditions. j |A i <θ i ) and P(R j |A i ≥θ i ), and calculate the probability change ΔP = P(R) j |A i ≥θ i )-P(R j |A i <θ i It can capture nonlinear causal mutations near the threshold of technical parameters. In particular, when ΔP is large, it indicates that exceeding the threshold of technical parameters will significantly increase the probability of adverse results, thereby accurately identifying key technical responsibility points.

[0044] This scheme converts legal elements into high-dimensional semantic vectors LA, LR, LC, and uses cosine similarity Sim(LA) to... i ,LR j) = cos(LA i ,LR j Calculate the semantic correlation between behavior and result, and introduce conditional vector LC. k Influence weight w k =Sim(LC k ,LA i )×Sim(LC k ,LR j This constructs a complete vector space for legal causal relationships. This vectorized representation method can capture the implicit semantic relationships between legal concepts, and in particular, the semantic vectors trained through a judicial case database can encode a large amount of judicial practice experience. Meanwhile, legal responsibility α... i =Σ(S i ×w k The calculation comprehensively considers the influence weights of multiple conditions and factors, realizing multi-factor weighted analysis of complex legal causal relationships. This enables the determination of legal liability to reflect the comprehensive judgment logic in real judicial precedents, and greatly improves the legal professionalism of online mediation.

[0045] By calculating the technical responsibility score T1 and the legal responsibility score T2 separately, and generating a comprehensive responsibility coefficient based on their respective weights, the complex determination of liability in civil and commercial disputes is transformed into a quantifiable numerical analysis. This dual causal relationship analysis method can accurately reflect the different degrees of influence of technical and legal factors in a case, providing a scientific basis for calculating compensation amounts in mediation reports. Attached Figure Description

[0046] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0047] Figure 1 This is an exemplary flowchart illustrating an online mediation method according to some embodiments of this application;

[0048] Figure 2 This is an exemplary flowchart illustrating the construction of a set of disputed elements according to some embodiments of this application;

[0049] Figure 3 This is an exemplary flowchart illustrating the generation of a first determination result according to some embodiments of this application;

[0050] Figure 4 This is an exemplary flowchart illustrating the generation of a second determination result according to some embodiments of this application;

[0051] Figure 5 This is an exemplary flowchart illustrating the generation of a mediation report according to some embodiments of this application. Detailed Implementation

[0052] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0053] like Figure 1 As shown, the system receives case data online, which includes basic information of the parties, disputed facts, and relevant evidence; it performs natural language processing and feature extraction on the case data to obtain a set of disputed elements; through causal analysis, it judges the conditional relationships of the set of disputed elements to obtain a first determination result and a second determination result, whereby the first determination result represents the facts determined by technology and the second determination result represents the facts determined by law; and a mediation report is generated based on the first determination result and the second determination result.

[0054] S1: The system receives case data from various channels via a web service interface. This data mainly includes three categories: basic information of the parties involved, descriptions of the disputed facts, and relevant evidence. The disputed facts are usually submitted in text format, while the relevant evidence is more diverse, including scanned contract documents, on-site photos, audio and video recordings, medical records, and technical testing reports.

[0055] like Figure 2 As shown in Figure S2, the system uses OCR technology to convert scanned documents into editable text, and performs text correction and format standardization. For documents commonly found in civil and commercial disputes, such as complaints, answers, and evidence, the system employs an OCR model based on the Transformer architecture.

[0056] The system employs the YOLO-v5 object detection algorithm to identify key evidentiary elements in images, including product defects, damage traces, and the scene environment. The system has pre-trained detection models covering categories such as vehicle damage, building cracks, medical devices, and product defects, and generates structured image feature descriptions in the format {target category, location coordinates, confidence level, degree of damage}.

[0057] The system employs a table structure recognition algorithm to extract numerical data and relationships within tables. Through row and column segmentation, cell content recognition, and data type determination, it constructs a table data matrix and identifies the reference relationships and calculation logic between tables.

[0058] By aligning text data, image feature descriptions, and tabular data matrices with timestamps and linking them with evidence chains, the system generates multimodal fused structured data. A unified evidence numbering system is established to ensure the consistency and traceability of data across different modalities.

[0059] The system uses a deep learning-based Named Entity Recognition (NER) model to extract key entities. The NER model adopts a BiLSTM-CRF architecture, which first converts the text into word vector representations, then captures contextual information through a bidirectional LSTM network, and finally performs sequence labeling through a CRF layer.

[0060] Entity type identification: The system identifies five core entities, including names (parties, witnesses, experts, etc.), place names (location of the incident, hospital, company address, etc.), time (time of the incident, time of medical treatment, time of contract signing, etc.), amount (amount of loss, medical expenses, contract amount, etc.), and events (medical treatment, accident process, breach of contract, etc.).

[0061] Original evidence source marker: Each identified entity will be marked with its exact location in the original evidence, establishing a mapping relationship between the entity and the evidence source, in the format of {entity content, entity type, evidence source file, page number / location, extraction time}.

[0062] The system employs dependency parsing technology to construct a syntax tree for sentences and identify syntactic components such as subject, verb, and object. It focuses on extracting four key semantic information categories: **Action Subject Identification:** By analyzing the agent of the verb, it determines who performed the action, distinguishing between the active actor and the passive recipient. **Action Object Identification:** It analyzes the recipient of the action, clarifying the goal and scope of the action's effect. **Causal Relationship Vocabulary Identification:** The system maintains a causal relationship vocabulary database, including words such as "cause," "lead," "cause," "because," "due to," "therefore," and "result," and labels the cause and effect parts of causal relationships. **Responsibility Keyword Identification:** It identifies words related to liability determination, such as "fault," "breach of contract," "negligence," "intentional," "shall," and "must," analyzing the attribution and degree of responsibility.

[0063] The system pre-constructs a knowledge graph for civil and commercial disputes, containing hundreds of thousands of nodes and millions of edges. Node types include legal entities, behavioral nodes, outcome nodes, legal provisions nodes, and case precedent nodes. The system employs the TransE graph embedding algorithm to map each node in the knowledge graph to a 128-dimensional vector representation. For newly extracted entities and semantic information, the cosine similarity between them and knowledge graph nodes is calculated; when the similarity exceeds a threshold of 0.85, an association mapping relationship is established. Implicit associations are discovered through a graph propagation algorithm, generating an association mapping set for each entity, containing {association node, association type, association strength, propagation path}, which serves as an important basis for subsequent labeling.

[0064] Based on the identified causal relationship terms and time information, the system classifies all entities according to causal elements in a time sequence: Causal behavioral elements: Behavioral entities that act before the outcome and are connected to it by causal relationship terms. Conditional elements: Mediating factors that do not directly cause the outcome but influence the probability or degree of its occurrence. Outcome elements: The endpoint of the causal chain, typically the manifestation of harmful consequences or breach of contract.

[0065] The system labels elements based on the association mapping results between entities and semantic information. It calculates the association mapping strength between each element and the technical specification library and the legal provision library: Technical category labeling condition: When the association mapping strength between an element and the technical specification library exceeds a threshold of 0.7, it is labeled as a technical category; Legal category labeling condition: When the association mapping strength between an element and the legal provision library exceeds a threshold of 0.6, it is labeled as a legal category. An element may be labeled as both technical and legal, reflecting its composite attributes in disputes. The causal behavioral elements, conditional elements, and result elements labeled as technical and legal are integrated to form the final set of disputed elements.

[0066] S3, through causal analysis, assesses the conditional relationships of the set of disputed elements to obtain a first determination and a second determination. The first determination represents the facts determined technically, and the second determination represents the facts determined legally, including:

[0067] like Figure 3 As shown, causal behavioral elements, conditional elements, and result elements labeled as technical and legal are extracted from the set of disputed elements to form technical causal relationship chains and legal causal relationship chains.

[0068] Label the technical cause-behavioral element data as input node dataset A = {A1, A2, ..., A...} m The technical result element data is labeled as the output node dataset R = {R1, R2, ..., R}. n The technical condition element data is labeled as an intermediate node dataset C = {C1, C2, ..., C}. k Construct the node relationship matrix and edge weight matrix to generate the technical causal directed graph data structure G(V,E);

[0069] Regular expression matching is used to parse the technical cause-behavioral elements in the input node dataset A, extracting and converting them into a standardized numerical technical parameter dataset V = {V1, V2, ..., V...}. n For medical technical parameters, these may include drug dosage, operation duration, and test indicators; for engineering technical parameters, these may include strength, dimensions, and temperature; for product quality parameters, these may include pass rate and service life; and for operational specifications, these may include operating procedures and safety measures.

[0070] Using database query and key-value pair matching algorithms, and with a standardized numerical technical parameter dataset V as the query key, batch searches are performed in a pre-set expert assistant technical specification database to obtain the corresponding professional standard threshold datasets. A parameter-threshold binary data structure mapping table Θ={(V1,θ1),(V2,θ2),......,(V n ,θ n )};

[0071] Each element V in the technical parameter dataset V i This represents a specific, quantifiable technical indicator, which is directly extracted from the technical causal behavioral elements of the case. For example: medical technical parameter V. i "Cephalosporin antibiotic dosage = 2.5g / day", "Surgery duration = 180 minutes", "Blood glucose level = 15.8mmol / L". Engineering technical parameters V i "Concrete compressive strength = 22.3 MPa", "Rebar diameter = 18 mm", "Construction ambient temperature = 38℃". Product quality parameters V i "Product qualification rate = 85.6%", "Lifespan test value = 1800 hours", "Formaldehyde content = 0.12mg / m³" 3 "etc." Professional standard thresholds refer to the benchmark values ​​for judging the compliance of technical parameters determined by industry norms, technical standards, expert consensus, etc.

[0072] Preferably, a conditional relationship verification algorithm is used, based on the root cause judgment logic of "no A, no R", according to the technical parameter V. i Through statistical analysis of historical cases, the necessary condition relationship of technical parameters as technical result elements Rj is verified.

[0073] Using conditional probability statistical calculation algorithms, based on validated data dependencies, the parameter-threshold mapping data is segmented and statistically processed to calculate the technical parameter data V separately. i Data less than the threshold θ i The sample subset and the data θ greater than or equal to the threshold i Results element R in the sample subset j The probability of occurrence P(R) j |A i <θ i ) and P(R j |A i ≥θ i Specifically, extract a set of cases H = {h1, h2, ..., h...} containing the same or similar technical parameter types from the historical case database. m For each case h in the case set H k Extract its technical parameter value vi,k Parameter threshold θ i,k and the result state r j,k Construct a set of technical parameter groups: G1 = {h k |v i,k <θ i,k} represents a subset of cases where the technical parameters are below the threshold; G2 = {h k |v i,k ≥θ i,k} represents a subset of cases where the technical parameters reach or exceed the threshold;

[0074] In the statistics of G1, the result R is caused by j The number of cases N(R) j A i <θ i The total number of cases N(A) and G1 i <θ i ); In statistics G2, the result R is caused by j The number of cases N(R) j A i ≥θ i The total number of cases N(A) and G2 i ≥θ i ); Calculate conditional probability: When the sample size N(A) i <θ i ) or N(A i ≥θ i When the sample size is less than the preset minimum sample size threshold, Laplace smoothing is applied.

[0075] Where ε is the smoothing parameter and K is the total number of outcome states; for technical parameter types of new cases lacking historical data, the prior estimate of the conditional probability is obtained through expert evaluation and dynamically adjusted in conjunction with the Bayesian update formula.

[0076] Calculate the probability difference data ΔP = P(R) j |A i ≥θ i )-P(R j |A i <θ i ), which serves as a numerical output of the strength index of technical causal relationships.

[0077] like Figure 4 As shown, a semantic vectorization algorithm is used to label legal cause-related behavioral elements as a set of legal behavior vectors LA = {LA1, LA2, ..., LA...}. m Legal outcome elements are tagged as a set of legal outcome vectors LR = {LR1, LR2, ..., LR...} nLegal condition elements are tagged as a set of legal condition vectors LC = {LC1, LC2, ..., LC...} k};

[0078] Extracting legal parameters L from LA i Legal parameter L i This includes data on subjective fault, objective behavior, damage results, and causal relationships. Subjective fault includes the degree of intent and negligence; objective behavior includes the illegality of the act; damage result parameters include quantitative indicators of damage, such as disability level, amount of property loss, and duration of business loss. These parameters are obtained through named entity recognition and numerical extraction. Causal relationship parameters assess the temporal and spatial correlation. Temporal correlation is assessed by calculating the time interval between the behavior and the result; the shorter the interval, the higher the correlation. Spatial correlation analyzes the relationship between the location where the behavior occurred and the location where the damage occurred.

[0079] According to legal parameter L i By using a pre-set database of judicial precedents, the corresponding judicial determination standard S is retrieved. i Construct legal parameters L i - Judicial Determination Standard S i The mapping set Ψ = {(L1,S1),(L2,S2),......,(L n ,S n )};

[0080] For each legal action vector LA i and legal outcome vector LR j Calculate vector similarity Sim(LA) i ,LR j ) = cos(LA i ,LR j Based on the principle of adequate causation, calculate the legal condition vector LC. k The weight w of the similarity k w k =Sim(LC k ,LA i )×Sim(LC k ,LR j According to the influence weight w k and judicial determination standards S i Calculate legal liability α i α i =Σ(S i ×w k ); Calculate the strength index of legal causal relationships

[0081]

[0082] Based on the magnitude of the probability change ΔP, a threshold classification algorithm is used to divide the causal relationship strength index of technical categories into three levels: strong correlation, moderate correlation, and weak correlation. When ΔP ≥ 0.7, it is determined to be a strong correlation; when 0.3 ≤ ΔP < 0.7, it is determined to be a moderate correlation; and when ΔP < 0.3, it is determined to be a weak correlation. Based on the technical parameter V... i The system retrieves the parameter threshold θi from the technical specification library based on the type of parameter and obtains the corresponding technical standard clause number; it then calculates the technical parameter V. i Relative to the parameter threshold θ i Degree of deviation: Deviation Construct a technical responsibility determination matrix M, where rows represent the strength level of causal relationship and columns represent the deviation range; find the technical responsibility level based on M[strength level][deviation range); generate a technical responsibility determination based on the output of the technical responsibility determination matrix M: when there is a strong correlation and δ i When the percentage is greater than 30%, the main technical responsibility is output; among them, the relevant and δ i If the percentage is greater than 20%, output "minor technical responsibility"; otherwise, output "light technical responsibility".

[0083] Technical responsibility level, deviation δ i Violated technical standard clause number, technical parameter V i and parameter threshold θ i The technical assessment results, encapsulated in JSON format, will be used as the primary assessment result.

[0084] When β∈[0.8, 1], a direct causal relationship label is set; when β∈[0.4, 0.8), an indirect causal relationship label is set; when β∈[0, 0.4), a no-causal relationship label is set; from the legal parameter L i Extract objective behavioral feature vectors and subjective fault feature vectors to construct a legal behavior feature matrix F = [objective behavioral features; subjective fault features]. Based on the legal behavior feature matrix F, calculate the vector similarity and retrieve the top k most relevant legal provisions from the legal provision database to obtain the set of legal provision numbers and the relevance score. Based on legal responsibility α... i And the causal relationship type label, calculate the proportion of responsibility R: when the causal relationship type is direct, R = α i When the causal relationship is indirect, R = α i×β; When the causal relationship type is none, R = 0; Based on the numerical range of the liability-bearing ratio R: when R ∈ [0.8, 1], it is mapped to full compensation liability; when R ∈ [0.5, 0.8), it is mapped to primary compensation liability; when R ∈ [0.2, 0.5), it is mapped to secondary compensation liability; when R ∈ (0, 0.2), it is mapped to supplementary compensation liability; The causal relationship type label, the legal provision number set, the liability-bearing ratio R, and the compensation liability type are encoded into structured data to generate a legal determination result, which serves as the second determination result.

[0085] like Figure 5 As shown, based on the technical responsibility level and deviation δ in the first determination result... i Calculate the technical responsibility score T1: Quantify and assign values ​​to the technical responsibility levels: major technical responsibility is assigned 1.0, minor technical responsibility is assigned 0.6, and slight technical responsibility is assigned 0.3; Calculate the deviation influence coefficient: when δ i When ≤10%, the influence coefficient is 0.8; when 10% <δ i When ≤30%, the influence coefficient is 1.0; when δ i When the percentage is greater than 30%, the influence coefficient is 1.2; Technical responsibility score T1 = Technical responsibility level assignment × Deviation influence coefficient × Probability change ΔP; Based on the liability-bearing ratio R and compensation liability type in the second determination result, calculate the legal responsibility score T2: Quantify the compensation liability type: full compensation liability is assigned 1.0, primary compensation liability is assigned 0.7, secondary compensation liability is assigned 0.4, and supplementary compensation liability is assigned 0.2; Calculate the legal causation correction coefficient: when β≥0.8, the correction coefficient is 1.0; when 0.4≤β<0.8, the correction coefficient is 0.8; when β<0.4, the correction coefficient is 0.5; Legal responsibility score T2 = Compensation liability type assignment × Liability-bearing ratio R × Legal causation correction coefficient;

[0086] Calculate the overall responsibility coefficient Among them, 0.4 and 0.6 are the preset weights for technical and legal factors, respectively; based on the comprehensive liability coefficient C and the amount in dispute M, the range of compensation liability is calculated: lower limit amount

[0087] =M×C×0.8; Upper limit amount =M×C×1.2; When the calculated result exceeds the disputed amount M, the upper limit amount is set to M; Based on the preset mediation report template, the first determination result, the second determination result, the technical responsibility score T1, the legal responsibility score T2, the comprehensive responsibility coefficient C, and the range of compensation liability amount are filled in in a structured manner to generate a mediation report containing four modules: technical analysis, legal analysis, responsibility determination, and compensation recommendation.

[0088] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for online mediation, characterized in that, include: S1, receiving case data through an online system, the case data including basic information of the parties, disputed facts and relevant evidence; S2, performs natural language processing and feature extraction on the case data to obtain a set of disputed elements; S3, through causal relationship analysis, makes a conditional relationship judgment on the set of disputed elements to obtain a first determination result and a second determination result. The first determination result represents the facts determined by technical means, and the second determination result represents the facts determined by law. S4. Generate a mediation report based on the first and second determination results; S5, conduct online mediation based on the mediation report; S2 performs natural language processing and feature extraction on the case data to obtain a set of disputed elements, including: Multimodal recognition and preprocessing of case data yields structured data containing text, image, and tabular data; Based on structured data, entities containing names of people, places, times, amounts, and events are extracted using named entity recognition algorithms, and the original evidence sources of each entity are marked. Based on structured data, semantic analysis is used to identify semantic information in disputed facts, including words related to the subject, object, causal relationship, and responsibility. Based on a pre-built knowledge graph of civil and commercial disputes, the extracted entities and semantic information are associated and mapped. Based on causal relationship vocabulary, the extracted entities are classified into causal elements according to time series, resulting in causal behavioral elements, conditional elements, and outcome elements; Based on the association mapping between entities and semantic information, the causal behavioral elements, conditional elements, and result elements are labeled as technical and legal categories, respectively. The marked causal, conditional, and outcome elements are used as a set of disputed elements. S3, through causal analysis, assesses the conditional relationships of the set of disputed elements to obtain a first determination and a second determination. The first determination represents the facts determined technically, and the second determination represents the facts determined legally, including: Extract causal, conditional, and consequential elements, categorized as technical and legal, from the set of disputed elements to form technical and legal causal chains; Based on the technological causal relationship chain, the conditional relationship is verified using an inference algorithm to obtain a technological causal relationship strength index; Based on the legal causal relationship chain, a similarity matching algorithm is used to verify the conditional relationship and obtain a legal causal relationship strength index; Based on the technical causal relationship strength index and the preset technical specification library, a matching algorithm is used to retrieve technical standards and obtain the technical identification result, which is used as the first identification result. Based on the legal causal strength index and a pre-set legal provisions database, a legal retrieval is performed using a matching algorithm to obtain a legal determination result, which serves as the second determination result.

2. The online mediation method according to claim 1, characterized in that: Based on the technological causal relationship chain, inference algorithms are used to verify the conditional relationships, resulting in a technological causal relationship strength index, including: Mark technical-related behavioral elements as nodes. Technical result elements are marked as nodes. Technical condition elements are marked as nodes. Constructing a causal directed graph of technology ; Identify technical parameters from technical cause-related behavioral elements. The technical parameters include medical technical parameters, engineering technical parameters, product quality parameters, or operating procedure parameters; Retrieve technical parameters from a pre-defined technical specification library. Corresponding parameter threshold Construct a set of technical parameters-threshold mappings ; Calculate the technical parameters separately Below the parameter threshold Conditional probability at time and exceeding the parameter threshold Conditional probability at time ; Calculate the change in probability , as an indicator of the strength of causal relationships in the technical category.

3. The online mediation method according to claim 1, characterized in that: Based on the legal causal relationship chain, a similarity matching algorithm is used to verify the conditional relationship, resulting in a legal causal relationship strength index, including: Using a semantic vectorization algorithm, legal cause-related behavioral elements are labeled as a set of legal behavior vectors. Legal outcome elements are tagged as a set of legal outcome vectors. Legal condition elements are tagged as a set of legal condition vectors. ; from Extracting legal parameters Legal parameters This includes data on subjective fault, objective conduct, damage results, and causal relationships; subjective fault includes the degree of intent and the degree of negligence; objective conduct includes the illegality of the conduct; damage results include the degree and scope of damage; and causal relationships include temporal and spatial correlations. According to legal parameters By using a pre-set database of judicial precedents, the corresponding judicial determination standards can be retrieved. Constructing legal parameters - Judicial Determination Standards Mapping set ; For each legal action vector and legal outcome vector Calculate vector similarity ; Calculate the legal condition vector based on the principle of proximate causation. Weight of similarity , ; Based on the influence weight and judicial determination standards Calculate legal liability , ; Calculating the strength index of legal causation .

4. The online mediation method according to claim 3, characterized in that: Based on the technical causal relationship strength index and a pre-set technical specification library, a matching algorithm is used to retrieve technical standards and obtain technical identification results, which serve as the first identification result, including: Based on the probability change Based on the magnitude of the correlation, technical causal relationship strength indicators are divided into three levels: strong correlation, moderate correlation, and weak correlation. According to technical parameters and parameter threshold Calculate the deviation ; Based on the degree of causality and deviation Construct a technical responsibility determination matrix M; Based on matrix M, a technical assessment result is generated and used as the first assessment result.

5. The online mediation method according to claim 3, characterized in that: Based on the legal causal strength index and a pre-set legal provisions database, a legal search is performed using a matching algorithm to obtain a legal determination result, which serves as the second determination result, including: Based on the legal causal strength index β, the legal causal strength index is divided into three levels: direct causal relationship, indirect causal relationship, and no causal relationship; From legal parameters Extract objective behavioral feature vectors and subjective fault feature vectors, and construct a legal behavior feature matrix F = [objective behavioral features; subjective fault features]; Based on the legal behavior feature matrix F, the top k legal provisions with the highest relevance are retrieved from the legal provisions database by calculating vector similarity, thus obtaining the set of legal provision numbers and relevance scores. According to legal liability And the causal relationship type label, calculate the proportion of responsibility R: when the causal relationship type is direct, When the causal relationship is indirect. When the causal relationship type is none, R=0; The type of liability for compensation is determined based on the liability-sharing ratio R, and a legal determination result is generated as the second determination result.

6. The online mediation method according to claim 5, characterized in that: S4. Based on the first and second determinations, generate a mediation report, including: Based on the technical responsibility level and deviation in the first assessment result Calculate the technical responsibility score ; The legal liability score is calculated based on the liability ratio R and the type of compensation liability in the second determination result. ; according to and The amount in dispute in the case is used to calculate the scope of the liability for compensation. A mediation report is generated based on the scope of the amount of compensation liability and the preset mediation report template.

7. A system for online mediation, used to implement the method according to any one of claims 1 to 6, characterized in that, include: The data receiving module receives case data through the online system; The feature extraction module performs natural language processing and feature extraction on case data to obtain a set of disputed elements. The causality module determines the causal relationship of the set of disputed elements to obtain the first and second determination results, including a technical analysis unit and a legal analysis unit. The structure generation module generates a mediation report based on the first and second determination results and a pre-set legal provisions library. The online mediation module allows for online mediation based on the mediation report.

8. The online mediation system according to claim 7, characterized in that: The technical analysis unit reasoned and verified the disputed elements, matched technical standards with the technical specification library, and generated a technical assessment result, which served as the first assessment result. The legal analysis unit verifies the similarity of disputed elements and matches them with a pre-set legal provisions database to generate a legal determination result, which serves as the second determination result.

Images