An AI-based multi-scene mediation intelligent decision system

Abstract

This invention discloses an AI-based intelligent decision-making system for multi-scenario mediation, relating to the fields of artificial intelligence and data processing technology. By converting original demands into high-dimensional semantic vectors and automatically clustering the vector space, this invention can identify highly semantically similar demand clusters from individual demands without pre-setting the number of categories. By calculating the cluster density and emotional intensity of each demand cluster, the system quantitatively evaluates the clusters, accurately classifying them into core rigid demand clusters, secondary flexible demand clusters, and waiverable demand clusters. This hierarchical mapping mechanism allows the system to go beyond the surface-level verbal expressions of the parties involved, identifying rigid needs, flexible demands, and strategic demands, laying a data foundation for subsequent quantitative decision-making. Combined with the Pareto optimality mechanism, this ensures the effectiveness of the final mediation solution and fully protects the interests of different groups.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and data processing technology, specifically to an AI-based multi-scenario mediation intelligent decision-making system. Background Technology

[0002] As socio-economic activities become increasingly complex, disputes have evolved from traditional two-party confrontations to multi-party games. In class action lawsuits across various scenarios, the diverse and intertwined demands and complex conflicts of interest among the parties make it easy for the mediation process to reach a stalemate. Traditional mediation methods rely on the mediator's personal experience and communication skills. However, when facing group disputes, mediators often struggle to consider all individual demands and cannot quantify the balance point of interests among the parties, resulting in a low success rate for mediation.

[0003] Most existing intelligent mediation systems or decision support tools are designed based on a "two-party confrontation" model, which can only adapt to the scenario of two parties and lacks the ability to model three or more parties. Therefore, they cannot handle the actual situation of diverse demands and complex conflicts of interest in group disputes.

[0004] In terms of the appeal handling mechanism, existing systems mostly adopt simple text classification or keyword matching, roughly categorizing individual appeals into a few categories, ignoring the emotional intensity and differences in the weight of interests behind the appeals, resulting in a lack of accurate basis for subsequent decisions; in terms of solution generation, existing technologies can only generate compromise solutions, which often only satisfy the interests of some people, but seriously damage the rights and interests of minority groups, and fail to maximize benefits.

[0005] In summary, existing technologies suffer from drawbacks in handling the interests of multiple parties, including crude identification of demands and lack of interest quantification. There is an urgent need for an intelligent decision-making system that can perform refined hierarchical clustering of demands from multiple parties, quantify and construct interest utility functions, and automatically generate a balanced mediation plan that takes into account the rights and interests of all parties, in order to overcome the technical bottlenecks in the mediation of group disputes. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI-based intelligent decision-making system for multi-scenario mediation. This system can convert original demands into high-dimensional semantic vectors and automatically cluster the vector space. Without pre-setting the number of categories, it can identify clusters of demands with high semantic similarity from individual demands. By calculating the cluster density and emotional intensity of each demand cluster, it quantitatively evaluates the demand clusters and accurately divides them into core rigid demand clusters, secondary flexible demand clusters, and abandonable demand clusters. This hierarchical mapping mechanism allows the system to go beyond the surface-level verbal expressions of the parties involved and identify rigid demands, flexible demands, and strategic demands, laying a data foundation for subsequent quantitative decision-making. Combined with the Pareto optimality mechanism, it ensures the effectiveness of the final mediation solution and fully protects the interests of different groups.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: an AI-based multi-scenario mediation intelligent decision-making system, comprising: The data acquisition and preprocessing module collects multimodal data generated by multiple parties during the mediation process, cleans and structures the multimodal data, generates an original request list containing multiple original requests for each party entity, and stores the original request list in the mediation database. The multi-party demand hierarchical clustering module retrieves the original demand list of all parties from the mediation database, converts each original demand into a demand semantic vector using a natural language processing model, performs cluster analysis on the demand semantic vectors using an unsupervised clustering algorithm to generate demand clusters, and divides the demand clusters into core rigid demand clusters, secondary flexible demand clusters, and waiverable demand clusters based on the cluster density and emotional intensity of the demand clusters. The utility function construction and game equilibrium module constructs a personalized utility function for each party based on the distribution of each party's demands in the core rigid demands cluster, secondary flexible demands cluster, and waiverable demands cluster. Based on the utility functions of all parties, it searches for a set of mediation solutions that satisfy the Pareto optimal condition. The group profile analysis and strategy generation module, based on the clustering results of the demand clusters and the Pareto optimal mediation solution set, identifies the core representatives and dissenting individuals in the group, generates dialogue strategies for the core representatives, and generates compensatory appeasement strategies for the dissenting individuals. The mediation database is used to store original claim data, claim vectors, clustering results, utility function parameters, historical mediation cases, and generated mediation strategies.

[0008] Furthermore, the acquisition and preprocessing module performs word segmentation, part-of-speech tagging, and dependency parsing on the acquired unstructured text data to extract the claim entities and claim modifiers; Perform sentiment analysis on the voice input data and extract voice sentiment feature values; The entity of the claim, the modifiers of the claim, and the voice emotion feature value are associated and fused to generate a structured claim triple. The claim triple includes the party identifier, the content of the claim, and the emotion intensity value. The claim triple is stored in the mediation database as the basic unit of the original claim list.

[0009] Furthermore, the multi-party claim hierarchical clustering module calls a pre-trained legal domain BERT model to convert each original claim into a claim semantic vector; Clustering algorithms are used to perform density clustering on the semantic vectors of the claims, identify high-density regions in the vector space as claim clusters, and mark discrete points that cannot be assigned to any cluster as noise claims. For each identified cluster of requests, the feature values ​​corresponding to all requests within the cluster are calculated. The mean of the emotional intensity values ​​is taken as the cluster emotional intensity, and the proportion of the number of requests contained in the cluster to the total number of requests is taken as the cluster density. Based on the calculated feature values, preset first and second cluster density thresholds, and first and second emotion thresholds, the appeal clusters are classified and determined, including: core rigid appeal clusters, secondary flexible appeal clusters, and abandonable appeal clusters.

[0010] Furthermore, when the multi-party demand hierarchical clustering module performs clustering analysis on the demand semantic vector using a clustering algorithm, the following steps are executed: Calculate the semantic vectors of any two claims and Cosine similarity between And convert cosine similarity into semantic distance. The semantic distance The value range is [0, 2], and the smaller the value, the closer the semantics of the two claims are; Preset neighborhood radius parameter and the minimum number of neighborhood points parameter Min For each claim semantic vector Statistics Centered on, with The number of other claim semantic vectors contained within a spherical region of radius [missing information]. ; based on With Min The semantic vectors of the appeals are divided into three categories: when Then Marked as a core point, it indicates that the demand is located at the core of a high-density demand area; when ,but Located at the core point Within its neighborhood, then Mark as boundary point; when The neighborhood that is neither a core point nor belongs to any core point will be These are marked as noise points and do not represent the population. The core point and its neighboring boundary points are aggregated into a claim cluster by density connectivity. Each claim cluster represents a claim category with high semantic similarity. All noise points do not belong to any claim cluster and are retained separately. After clustering is completed, for each identified cluster of claims Eigenvalue calculation is performed, wherein the eigenvalues ​​include cluster density and cluster sentiment intensity, and the cluster density... ,in, Indicates the first The total number of demands contained in each demand cluster. The total number of all requests, and the intensity of the cluster sentiment. ,in, Indicates the first The number of parties involved in each cluster of claims. Indicates the first The first of the clusters of demands The normalized values ​​of the emotional characteristics of the individual's speech.

[0011] Furthermore, the criteria for classification are as follows: Will satisfy cluster density and The cluster of demands is marked as the core rigid demands cluster, among which, To preset the density threshold of the first cluster, To preset the first emotional threshold; Will satisfy and The cluster of demands is labeled as a secondary flexible demand cluster, in which, To preset the second cluster density threshold, To preset a second emotional threshold; Will satisfy or The cluster of claims and all claims marked as noise points are marked as abandonable claims.

[0012] Furthermore, the utility function construction and game equilibrium module is applicable to each party. ,according to Based on the distribution of demands within the core rigid demands cluster, the secondary flexible demands cluster, and the abandonable demands cluster, construct a utility function. ,in, Represents the mediation scheme vector. , , These represent the specific degree of satisfaction sub-vectors for the core rigid demand cluster, the secondary flexible demand cluster, and the abandonable demand cluster, respectively. The subvector representing the degree of satisfaction of the core rigid demand cluster, its dimension This equals the number of specific demands contained within the core set of rigid demands, with each component representing a different number of demands. Indicates that for the first The degree to which each core, rigid demand is met is denoted by a value in the range [0, 1], where 0 represents no satisfaction and 1 represents full satisfaction. For the sub-vector representing the degree of satisfaction of the secondary flexible demand cluster, its dimension is... This equals the number of specific demands contained within the secondary flexible demands cluster, with each component... Indicates that for the first The specific degree to which each secondary flexible demand is met. Let the subvector represent the degree of satisfaction of the cluster of abandonable claims, and its dimension is... This equals the number of specific claims contained in the cluster of claims that can be abandoned, with each component representing... Indicates that for the first The specific degree to which each waiverable claim is met. , , These represent the sub-utility functions of the parties involved on the core rigid demands cluster, the secondary flexible demands cluster, and the waiverable demands cluster, respectively. Used to quantify the parties involved Degree of satisfaction of core rigid demands Satisfaction levels, for demands with continuous numerical characteristics, Using linear functions , For the parties involved For the The preference weights for each core demand are obtained by normalizing the emotional intensity value of the party concerned on that demand; for demands with Boolean characteristics or threshold characteristics, Using a step function, when The value is set to 1 when the threshold preset by the party concerned is reached, and 0 otherwise. Used to quantify the parties involved Degree of satisfaction of secondary flexible demands Satisfaction levels were uniformly expressed using a logarithmic function. , For the weighting coefficient of flexible appeal, Used to quantify the parties involved Degree of satisfaction of the cluster of waiveable demands Satisfaction levels were uniformly assessed using a linear function. , and For index number, , , These represent the respective preference weights of the parties for the three categories of claims, satisfying... , The assignment is based on: if the party concerned If a party has a demand within the core rigid demand cluster, its demand is calculated based on its proportion of the demand within that cluster and the average emotional intensity, ensuring that it occupies a dominant position in the utility function; if the party... Those who did not participate in the core set of rigid demands, and The assignment is based on: according to the parties involved The distribution of demands is allocated according to the proportion of demands in the secondary flexible demands cluster and the optional demands cluster.

[0013] Furthermore, the steps of constructing the utility function and searching the solution set of mediation schemes that satisfy the Pareto optimality condition by the game equilibrium module are as follows: Receive each party Constructed utility function ,in, , Given the total number of parties involved, the search problem for mediation solutions is treated as a multi-objective optimization problem, i.e., using the vector of mediation solutions... As decision variables, with Utility function To optimize the objective, within the feasible region Optimization is performed internally; Constructing the game strategy space based on the clustering results of demand clusters ,in, This is the core rigid demand strategy subspace, which contains different combinations of implementation methods for each demand item in the core rigid demand cluster. This is the secondary flexible appeal strategy subspace, which includes different combinations of concession ratios for each appeal item within the secondary flexible appeal cluster. The subspace for waiveable claims strategy includes combinations of different handling methods for each claim item in the waiveable claim cluster. Each element in the vector corresponds to a specific mediation scheme. And all elements satisfy the feasible region. constraint; A multi-objective genetic algorithm is used to iteratively solve the multi-objective optimization problem, including: Initialization: in policy space Generate in An initial population of candidate mediation schemes, each candidate scheme ; Non-dominated ranking: based on the utility function values ​​of all parties involved. Candidate solutions in the population are non-dominated and sorted into multiple Pareto front layers. The solutions in the first Pareto front layer satisfy the condition that there are no other feasible solutions. Makes all have And at least one Make ; Evolution and Termination: Offspring populations are generated through simulated binary crossover and polynomial mutation. After merging parent and offspring populations, environmental selection is performed. The process continues until a preset maximum number of iterations is reached. When the iteration ends, proceed. After the iteration terminates, all candidate mediation schemes located in the first Pareto front layer are output, forming the Pareto optimal mediation solution set. ,in, To determine the number of solutions in the solution set, this Each of the proposed solutions satisfies the following condition: there is no other feasible solution that is superior to the solution in the utility function of at least one party and not inferior to the solution in the utility function of all other parties.

[0014] Furthermore, the group profile analysis and strategy generation module constructs a demand-person binary structure based on the clustering results of the demand clusters. The nodes of the structure include demand cluster nodes and party node, and the edges indicate that the party is associated with the demand cluster. The betweenness centrality of each party node is calculated, and the top E parties with the highest betweenness centrality values ​​are identified as core representatives. The betweenness centrality is used to measure the importance of the parties as information transmission bridges in the graph. For each candidate solution in the Pareto optimal mediation solution set Calculate the utility value of each party under the candidate solutions. Its ideal utility value Utility loss value between ; Utility loss value Parties whose losses exceed a preset threshold and who are not core representatives are marked as dissenting individuals; For the core representative, a dialogue strategy template is retrieved from the mediation database and pushed to the mediator's terminal. For the dissenting individual, an alternative compensation scheme library is retrieved from the mediation database, based on the distribution of the dissenting individual's demands and utility loss value. Matching compensatory reassurance strategies and pushing them to the mediator's terminal.

[0015] Furthermore, the workflow of the system includes: S100. Collect the original request data of multiple parties through the data acquisition and preprocessing module, generate an original request list and store it in the mediation database; S200. The original demands are vectorized and clustered through the multi-party demand hierarchical clustering module, and the demands are divided into core rigid demand clusters, secondary flexible demand clusters and abandonable demand clusters, and the clustering results are stored in the mediation database. S300. The utility function construction and game equilibrium module constructs a utility function for each party based on the clustering results, and generates a Pareto optimal mediation solution set through a multi-objective optimization algorithm, which is then stored in the mediation database. S400. The group profile analysis and strategy generation module identifies core representatives and dissenting individuals based on clustering results and Pareto optimal solutions, and generates targeted dialogue strategies and compensatory appeasement strategies. S500 pushes the generated Pareto optimal mediation solution set and differentiated communication strategy to the mediator's terminal to assist the mediator in completing the final mediation work.

[0016] Compared with existing technologies, this AI-based multi-scenario mediation intelligent decision-making system has the following beneficial effects: I. This invention transforms the original demands into high-dimensional semantic vectors and automatically clusters the vector space. Without pre-setting the number of categories, it can identify demand clusters with high semantic similarity from individual demands. By calculating the cluster density and emotional intensity of each demand cluster, the demand clusters are quantitatively evaluated, and the demand clusters are accurately divided into core rigid demand clusters, secondary flexible demand clusters, and abandonable demand clusters. This hierarchical mapping mechanism enables the system to go beyond the surface verbal expressions of the parties involved and identify rigid demands, flexible demands, and strategic demands, laying a data foundation for subsequent quantitative decision-making. Combined with the Pareto optimality mechanism, it ensures the effectiveness of the final mediation solution and fully protects the interests of different groups.

[0017] Second, based on the results of hierarchical clustering of demands, this invention constructs a personalized utility function for each party involved. Through a weighted combination of core rigid demand sub-utility functions, secondary flexible demand sub-utility functions, and waiverable demand sub-utility functions, it quantifies the parties' satisfaction with the degree of fulfillment of different demand clusters. It also transforms emotional intensity values ​​into weight parameters in the utility function, achieving an objective mathematical expression of subjective demands. The problem of searching for mediation solutions is formalized into a multi-objective optimization problem. Using the utility functions of all parties as the optimization objective, it iterative optimization is performed in the strategy space to identify the Pareto front solution set, ensuring that the final output mediation solution is an equilibrium solution that maximizes benefits. This fully satisfies the rigid needs of the core group while ensuring the rationality of the negotiation space through the logarithmic utility design of flexible demands, achieving a balance and consideration of the rights and interests of all parties.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 A flowchart of an AI-based multi-scenario mediation intelligent decision-making system; Figure 2 A block diagram showing the modular components of an AI-based multi-scenario mediation intelligent decision-making system; Figure 3 This is a block diagram illustrating the working principle of the group profile analysis and strategy generation module in this embodiment of the invention. Detailed Implementation

[0021] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] To address the shortcomings of existing intelligent mediation systems in multi-party game scenarios, such as crude demand identification, lack of interest quantification, and insufficient scientific basis for generating mediation solutions, this invention provides an AI-based multi-scenario intelligent decision-making system for mediation. This system aims to achieve precise quantitative analysis and automatic generation of balanced mediation solutions for group disputes by constructing a fully intelligent process encompassing multimodal data collection, hierarchical clustering of demands, utility function modeling, game equilibrium solving, group profiling analysis, and strategy generation. It is applicable to complex mediation scenarios involving multiple parties, such as labor disputes, consumer rights protection, property disputes, and intellectual property disputes. Through AI technology, it transforms mediators' personal experience into calculable, optimizable, and interpretable intelligent decision support capabilities, breaking through the technical bottlenecks of traditional mediation.

[0023] like Figure 2 As shown, the present invention provides an AI-based intelligent decision-making system for multi-scenario mediation. This system includes: a data acquisition and preprocessing module, a multi-party demand hierarchical clustering module, a utility function construction and game equilibrium module, a group profile analysis and strategy generation module, and a mediation database, wherein: The data acquisition and preprocessing module collects multimodal data generated by multiple parties during the mediation process, cleans and structures the multimodal data, generates an original request list containing multiple original requests for each party entity, and stores the original request list in the mediation database. The multi-party demand hierarchical clustering module retrieves the original demand list of all parties from the mediation database, uses a natural language processing model to convert each original demand into a demand semantic vector, performs cluster analysis on the demand semantic vector through an unsupervised clustering algorithm to generate demand clusters, and divides the demand clusters into core rigid demand clusters, secondary flexible demand clusters, and waiverable demand clusters according to the cluster density and emotional intensity of the demand clusters. The utility function construction and game equilibrium module constructs a personalized utility function for each party based on the distribution of each party's demands in the core rigid demands cluster, secondary flexible demands cluster, and waiverable demands cluster. Based on the utility functions of all parties, it searches for a set of mediation solutions that satisfy the Pareto optimal conditions. The group profiling and strategy generation module, based on the clustering results of the demand clusters and the Pareto optimal mediation solution set, identifies the core representatives and dissenting individuals in the group, generates dialogue strategies for the core representatives, and generates compensatory appeasement strategies for the dissenting individuals. The mediation database is used to store original claim data, claim vectors, clustering results, utility function parameters, historical mediation cases, and generated mediation strategies.

[0024] like Figure 1 As shown, the workflow of this AI-based multi-scenario mediation intelligent decision-making system includes: S100. Collect the original demands data of multiple parties through the data collection and preprocessing module, generate a list of original demands and store it in the mediation database; S200. The original demands are vectorized and clustered through the multi-party demand hierarchical clustering module. The demands are divided into core rigid demand clusters, secondary flexible demand clusters and abandonable demand clusters. The clustering results are stored in the mediation database. S300: The utility function construction and game equilibrium module constructs a utility function for each party based on the clustering results, and generates a Pareto optimal mediation solution set through a multi-objective optimization algorithm, which is then stored in the mediation database. S400: Based on clustering results and Pareto optimal solutions, the group profiling and strategy generation module identifies core representatives and dissenting individuals, and generates targeted dialogue strategies and compensatory appeasement strategies. S500 pushes the generated Pareto optimal mediation solution set and differentiated communication strategy to the mediator's terminal to assist the mediator in completing the final mediation work.

[0025] In the specific implementation process, at the initial stage of mediation, the mediator guides the parties involved to state their demands through front-end devices. For text data, such as the demands statement entered by the parties through the mediation terminal and the text of scanned copies of submitted written materials after OCR recognition, the data acquisition and preprocessing module first performs word segmentation, part-of-speech tagging, and dependency parsing analysis. By calling Chinese natural language processing tools, the text is segmented into sentences, identifying components such as subject, predicate, and object. The module also extracts the demand entities using a predefined demand entity dictionary and demand modifiers to preliminarily determine the intensity of the demand. For voice input data, i.e., the real-time voice stream of the parties during their statements, the data acquisition and preprocessing module performs voice endpoint detection and then uses a pre-trained emotion recognition model to extract voice emotion feature values.

[0026] By associating the entity of the claim, the modifier of the claim, and the voice emotion feature value extracted from the same party and the same time window, a structured claim triple is generated. The claim triple constitutes the basic unit of the party's original claim list and is completely stored in the mediation database for subsequent modules to call, laying the data foundation for subsequent quantitative analysis.

[0027] The multi-party demand hierarchical clustering module retrieves the original demand list of all parties from the mediation database. It then calls a pre-trained BERT model to convert the text content of each original demand into a high-dimensional demand semantic vector. For a scenario containing G parties and N original demands, the module generates N demand semantic vectors. These demand semantic vectors are then clustered using the density-based clustering algorithm DBSCAN. The specific execution steps are as follows: Calculate the semantic vector of any two claims and Cosine similarity between And convert cosine similarity into semantic distance. The semantic distance The value range is [0, 2]. The smaller the value, the closer the semantics of the two claims are.

[0028] Preset neighborhood radius parameter and the minimum number of neighborhood points parameter Min For each claim semantic vector Statistics Centered on, with The number of other claim semantic vectors contained within a spherical region of radius [missing information]. .

[0029] Based on Ct(v) and The comparison results divide the semantic vectors of the claims into three categories: When Ct(v)≥ Then v will be marked as the core point, representing that the demand is located in the core position of the high-density demand area; When Ct(v) < However, if v is located within the ε neighborhood of one or more core points, then v is marked as a boundary point, representing that the claim is located at the edge of a high-density region. If v is neither a core point nor a neighborhood of any core point, then v is marked as a noise point, indicating that the semantic difference between this claim and other claims is large and it does not have the characteristics of a group.

[0030] By using density connectivity, core points and their neighboring boundary points are aggregated into a cluster of appeals. Each appeal cluster represents a category of appeals with high semantic similarity. All noise points do not belong to any appeal cluster and are retained separately.

[0031] After clustering is completed, for each identified cluster of claims Eigenvalue calculation is performed, wherein the eigenvalues ​​include cluster density and cluster sentiment intensity, and the cluster density... ,in, Indicates the first The total number of demands contained in each demand cluster. The total number of all requests, and the intensity of the cluster sentiment. ,in, Indicates the first The number of parties involved in each cluster of claims. Indicates the first The first of the clusters of demands The normalized values ​​of the emotional characteristics of the individual's speech.

[0032] Based on the calculated feature values ​​and the preset first cluster density threshold Second cluster density threshold First emotional threshold Second emotional threshold , , The demands are categorized and determined as follows: Will satisfy cluster density and The clusters of demands are marked as core rigid demand clusters. These clusters represent rigid interest demands with a large number of participants, strong emotional reactions, and a high degree of consistency in their demands.

[0033] Will satisfy and The groups of demands are marked as secondary flexible demands. These groups have a certain population base, but their emotions are relatively peaceful, and there is room for negotiation and compromise.

[0034] Will satisfy or The cluster of demands, as well as all demands marked as noise points, are marked as waiveable demands. These demands are either individual demands of a particular party or demands from a party who is emotionally calm and not insistent, and can be used as bargaining chips for exchange or waiver in mediation.

[0035] The original demands are hierarchically mapped into three clusters of demands with different weights and negotiating positions. The clustering results, including the specific demands contained in each demand cluster, the corresponding list of parties involved, and the emotional intensity of the cluster, are stored in the mediation database.

[0036] The utility function construction and game equilibrium module constructs a personalized utility function for each party based on clustering results, and searches for a set of mediation solutions that satisfy Pareto optimality conditions through a multi-objective optimization algorithm.

[0037] For each party involved First, we analyze the distribution of demands within the three categories of demands. In this embodiment, there are a total of One core rigid demand item (specific clauses extracted from the core rigid demand cluster). Secondary flexible requests For each waiverable claim, the mediation solution vector x is defined as follows: ,in , quantity , , ∈[0,1] respectively represent the first... The degree to which core rigid demands, secondary flexible demands, and demands that can be waived are met.

[0038] party Personalized utility function Constructed as a weighted sum: ,in, Represents the mediation scheme vector. , , These represent subvectors indicating the degree of satisfaction for the core rigid demand cluster, the secondary flexible demand cluster, and the optional demand cluster, respectively, where the weight coefficients are... , , These represent the respective preference weights of the parties for the three categories of claims, satisfying... , The assignment is based on: if the party concerned If a party has a demand within the core rigid demand cluster, its demand is calculated based on its proportion of the demand within that cluster and the average emotional intensity, ensuring that it occupies a dominant position in the utility function; if the party... Those who did not participate in the core set of rigid demands, and The assignment is based on: according to the parties involved The distribution of demands within the secondary flexible demands cluster and the optional demands cluster is allocated proportionally. , , These represent the sub-utility functions of the parties involved on the core rigid demands cluster, the secondary flexible demands cluster, and the waiverable demands cluster, respectively. Used to quantify the parties involved Degree of satisfaction of core rigid demands Satisfaction levels, for demands with continuous numerical characteristics, Using linear functions , For the parties involved For the The preference weights for each core demand are obtained by normalizing the emotional intensity value of the party concerned on that demand; for demands with Boolean characteristics or threshold characteristics, Using a step function, when The value is set to 1 when the threshold preset by the party concerned is reached, and 0 otherwise. Used to quantify the parties involved Degree of satisfaction of secondary flexible demands Satisfaction levels were uniformly expressed using a logarithmic function. , For the weighting coefficient of flexible appeal, Used to quantify the parties involved Degree of satisfaction of the cluster of waiveable demands Satisfaction levels were uniformly assessed using a linear function. , and This is the index number.

[0039] After constructing the utility functions for all G parties involved, the utility function construction and game equilibrium module formalizes the search problem for mediation solutions into a multi-objective optimization problem: maximizing... ,satisfy ∈ ,in It is the feasible domain, which consists of the physical and logical constraints of all the claims.

[0040] To efficiently solve this multi-objective optimization problem, in this embodiment, the utility function construction and game equilibrium module constructs the game strategy space. ,in, This is the core rigid demand strategy subspace, which contains different combinations of implementation methods for each demand item in the core rigid demand cluster. This is the secondary flexible appeal strategy subspace, which includes different combinations of concession ratios for each appeal item within the secondary flexible appeal cluster. The subspace for waiveable claims strategy includes combinations of different handling methods for each claim item in the waiveable claim cluster. Each element in the vector corresponds to a specific mediation scheme. And all elements satisfy the feasible region. constraint.

[0041] The utility function construction and game equilibrium module employs a multi-objective genetic algorithm for iterative solution. Specific steps include: Initialization: in policy space Generate in An initial population of candidate mediation schemes, each candidate scheme ; Non-dominated ranking: based on the utility function values ​​of all parties involved. Candidate solutions in the population are non-dominated and sorted into multiple Pareto front layers. The solutions in the first Pareto front layer satisfy the condition that there are no other feasible solutions. Makes all have And at least one Make ; Evolution and Termination: Offspring populations are generated through simulated binary crossover and polynomial mutation. After merging parent and offspring populations, environmental selection is performed. The process continues until a preset maximum number of iterations is reached. When the iteration ends, proceed. After the iteration terminates, all candidate mediation schemes located in the first Pareto front layer are output, forming the Pareto optimal mediation solution set. ,in, To determine the number of solutions in the solution set, this Each of the proposed solutions satisfies the following condition: there is no other feasible solution that is superior to the solution in the utility function of at least one party and not inferior to the solution in the utility function of all other parties.

[0042] The group profiling and strategy generation module constructs a bipartite graph of demands and individuals based on the clustering results of demand clusters. The graph's nodes include two types: demand cluster nodes and individual nodes. If an individual is associated with a demand cluster (i.e., the individual has made a demand belonging to that cluster), an edge is established between them. This bipartite graph visually represents the intersection of multiple demand clusters for an individual, such as... Figure 3 As shown, the working principle of this group profiling analysis and strategy generation module is as follows: Calculate the betweenness centrality of each party node. Betweenness centrality is used to measure the importance of a party as a bridge for information transmission in the graph, that is, how many shortest paths pass through the node. The top E parties with the highest betweenness centrality values ​​are identified as core representatives. For each candidate solution h in the Pareto optimal mediation solution set H, calculate the utility value for each party under that solution. Its ideal utility value Utility loss value between Ideal utility value It can be obtained through single-objective optimization, that is, the maximum value that can be achieved by maximizing the utility function of the individual alone, which is the utility loss value. This reflects the sacrifices that the parties involved had to make when accepting the compromise. Utility loss value Parties whose losses exceed a preset threshold and who are not core representatives are marked as dissenting individuals, and differentiated strategies are generated for different roles: For key representatives, dialogue strategy templates are retrieved from the mediation database; for dissenting individuals, alternative compensation schemes are retrieved from the mediation database. The generated Pareto optimal mediation solution set and the generated differentiated communication strategies are pushed to the mediator's work terminal through a visual interface. The mediator can view the specific terms of each solution, the estimated utility of each party, and communication suggestions for key figures.

[0043] In addition, after the mediator finally negotiates and determines a specific solution with all parties involved, the solution and its implementation results will be stored in the mediation database as a new historical mediation case for subsequent model iteration and optimization.

[0044] In summary, this invention, through the organic synergy of a data acquisition and preprocessing module, a multi-party demand hierarchical clustering module, a utility function construction and game equilibrium module, and a group profile analysis and strategy generation module, constructs an intelligent decision-making system covering the entire mediation process. This system can transform subjective and vague individual demands into objective and quantifiable mathematical expressions, automatically search for equilibrium solutions that take into account the rights and interests of all parties using AI technology, and generate practical communication strategies. This liberates mediators from tedious information processing and trial-and-error, significantly improving the success rate and efficiency of mediating complex group disputes, and achieving a leap from experience-driven to data- and intelligence-driven approaches.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI-based multi-scenario mediation intelligent decision-making system, characterized in that, The system includes: The data acquisition and preprocessing module collects multimodal data generated by multiple parties during the mediation process, cleans and structures the multimodal data, generates an original request list containing multiple original requests for each party entity, and stores the original request list in the mediation database. The multi-party demand hierarchical clustering module retrieves the original demand list of all parties from the mediation database, converts each original demand into a demand semantic vector using a natural language processing model, performs cluster analysis on the demand semantic vectors using an unsupervised clustering algorithm to generate demand clusters, and divides the demand clusters into core rigid demand clusters, secondary flexible demand clusters, and waiverable demand clusters based on the cluster density and emotional intensity of the demand clusters. The utility function construction and game equilibrium module constructs a personalized utility function for each party based on the distribution of each party's demands in the core rigid demands cluster, secondary flexible demands cluster, and waiverable demands cluster. Based on the utility functions of all parties, it searches for a set of mediation solutions that satisfy the Pareto optimal condition. The group profile analysis and strategy generation module, based on the clustering results of the demand clusters and the Pareto optimal mediation solution set, identifies the core representatives and dissenting individuals in the group, generates dialogue strategies for the core representatives, and generates compensatory appeasement strategies for the dissenting individuals. The mediation database is used to store original claim data, claim vectors, clustering results, utility function parameters, historical mediation cases, and generated mediation strategies.

2. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The acquisition and preprocessing module performs word segmentation, part-of-speech tagging, and dependency parsing on the acquired unstructured text data to extract the claim entities and claim modifiers. Perform sentiment analysis on the voice input data and extract voice sentiment feature values; The entity of the claim, the modifiers of the claim, and the voice emotion feature value are associated and fused to generate a structured claim triple. The claim triple includes the party identifier, the content of the claim, and the emotion intensity value. The claim triple is stored in the mediation database as the basic unit of the original claim list.

3. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The multi-party demand hierarchical clustering module calls a pre-trained BERT model in the legal domain to convert each original demand into a demand semantic vector. Clustering algorithms are used to perform density clustering on the semantic vectors of the claims, identify high-density regions in the vector space as claim clusters, and mark discrete points that cannot be assigned to any cluster as noise claims. For each identified cluster of requests, the feature values ​​corresponding to all requests within the cluster are calculated. The mean of the emotional intensity values ​​is taken as the cluster emotional intensity, and the proportion of the number of requests contained in the cluster to the total number of requests is taken as the cluster density. Based on the calculated feature values, preset first and second cluster density thresholds, and first and second emotion thresholds, the appeal clusters are classified and determined, including: core rigid appeal clusters, secondary flexible appeal clusters, and abandonable appeal clusters.

4. The AI-based multi-scenario mediation intelligent decision-making system according to claim 3, characterized in that, When the multi-party demand hierarchical clustering module performs clustering analysis on the demand semantic vectors using a clustering algorithm, the following steps are executed: Calculate the semantic vectors of any two claims and Cosine similarity between And convert cosine similarity into semantic distance. The semantic distance The value range is [0, 2]; Preset neighborhood radius parameter and the minimum number of neighborhood points parameter Min For each claim semantic vector Statistics Centered on, with The number of other claim semantic vectors contained within a spherical region of radius [missing information]. ; based on With Min The semantic vectors of the appeals are divided into three categories: when Then Marked as a core point, it indicates that the demand is located at the core of a high-density demand area; when ,but Located at the core point Within its neighborhood, then Mark as boundary point; when The neighborhood that is neither a core point nor belongs to any core point will be These are marked as noise points and do not represent the population. The core point and its neighboring boundary points are aggregated into a claim cluster by density connectivity. Each claim cluster represents a claim category with high semantic similarity. All noise points do not belong to any claim cluster and are retained separately. After clustering is completed, for each identified cluster of requests... Eigenvalue calculation is performed, wherein the eigenvalues ​​include cluster density and cluster sentiment intensity, and the cluster density... ,in, Indicates the first The total number of demands contained in each demand cluster. The total number of all requests, and the intensity of the cluster sentiment. ,in, Indicates the first The number of parties involved in each cluster of claims. Indicates the first The first of the clusters of demands The normalized values ​​of the emotional characteristics of the individual's speech.

5. The AI-based multi-scenario mediation intelligent decision-making system according to claim 3, characterized in that, The criteria for classification are as follows: Will satisfy cluster density and The cluster of demands is marked as the core rigid demands cluster, among which, To preset the first cluster density threshold, To preset the first emotional threshold; Will satisfy and The cluster of demands is labeled as a secondary flexible demand cluster, in which, To preset the second cluster density threshold, To preset a second emotional threshold; Will satisfy or The cluster of claims and all claims marked as noise points are marked as abandonable claims.

6. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The utility function construction and game equilibrium module is for each party. ,according to Based on the distribution of demands within the core rigid demands cluster, the secondary flexible demands cluster, and the abandonable demands cluster, construct a utility function. ,in, Represents the mediation scheme vector. , , These represent sub-vectors indicating the degree of satisfaction for the core rigid demands cluster, the secondary flexible demands cluster, and the abandonable demands cluster, respectively. , , These represent the sub-utility functions of the parties involved on the core rigid demands cluster, the secondary flexible demands cluster, and the waiverable demands cluster, respectively. Used to quantify the parties involved Degree of satisfaction of core rigid demands satisfaction , , These represent the respective preference weights of the parties for the three categories of claims, satisfying... .

7. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The steps for constructing the utility function and searching the solution set of mediation schemes that satisfy the Pareto optimality condition in the game equilibrium module are as follows: Receive each party Constructed utility function ,in, , Given the total number of parties involved, the search problem for mediation solutions is treated as a multi-objective optimization problem, i.e., using the vector of mediation solutions... As decision variables, with Utility function To optimize the objective, within the feasible region Optimization is performed internally; Constructing the game strategy space based on the clustering results of demand clusters ,in, For the core rigid demand strategy subspace, For secondary flexible demand strategy subspace, The policy space is a subspace for policies that can be abandoned. Each element in the vector corresponds to a specific mediation scheme. And all elements satisfy the feasible region. constraint; A multi-objective genetic algorithm is used to iteratively solve the multi-objective optimization problem, including: Initialization: in policy space Generate in An initial population of candidate mediation schemes, each candidate scheme ; Non-dominated ranking: based on the utility function values ​​of all parties involved. Candidate solutions in the population are non-dominated and sorted into multiple Pareto front layers. The solutions in the first Pareto front layer satisfy the condition that there are no other feasible solutions. Makes all have And at least one Make ; Evolution and Termination: Offspring populations are generated through simulated binary crossover and polynomial mutation. After merging parent and offspring populations, environmental selection is performed. The process continues until a preset maximum number of iterations is reached. When the iteration ends, proceed. After the iteration terminates, all candidate mediation schemes located in the first Pareto front layer are output, forming the Pareto optimal mediation solution set. ,in, The number of solutions in the solution set.

8. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The group profile analysis and strategy generation module constructs a demand-person binary structure based on the clustering results of demand clusters. The nodes of the structure include demand cluster nodes and party node, and the edges indicate that the party is associated with the demand cluster. Calculate the betweenness centrality of each party node, and the nodes with the highest betweenness centrality values ​​are the first to be included. One party was identified as the core representative; For each candidate solution in the Pareto optimal mediation solution set Calculate the utility value of each party under the candidate solutions. Its ideal utility value Utility loss value between ; Utility loss value Parties whose losses exceed a preset threshold and who are not core representatives are marked as dissenting individuals; For the core representative, a dialogue strategy template is retrieved from the mediation database and pushed to the mediator's terminal. For the dissenting individual, an alternative compensation scheme library is retrieved from the mediation database, based on the distribution of the dissenting individual's demands and utility loss value. Matching compensatory reassurance strategies and pushing them to the mediator's terminal.

9. The AI-based multi-scenario mediation intelligent decision-making system according to claim 1, characterized in that, The workflow of the system includes: S100. Collect the original request data of multiple parties through the data acquisition and preprocessing module, generate an original request list and store it in the mediation database; S200. The original demands are vectorized and clustered through the multi-party demand hierarchical clustering module, and the demands are divided into core rigid demand clusters, secondary flexible demand clusters and abandonable demand clusters, and the clustering results are stored in the mediation database. S300. The utility function construction and game equilibrium module constructs a utility function for each party based on the clustering results, and generates a Pareto optimal mediation solution set through a multi-objective optimization algorithm, which is then stored in the mediation database. S400. The group profile analysis and strategy generation module identifies core representatives and dissenting individuals based on clustering results and Pareto optimal solutions, and generates targeted dialogue strategies and compensatory appeasement strategies. S500 pushes the generated Pareto optimal mediation solution set and differentiated communication strategy to the mediator's terminal to assist the mediator in completing the final mediation work.

Images