Method and system for automatically generating personalized mediation protocol based on mediation case characteristics

Abstract

The invention provides a method and system for automatically generating a personalized mediation protocol based on mediation case characteristics, and relates to the technical field of intelligent laws, and the method comprises the steps: obtaining historical mediation case information; performing structured processing according to the historical mediation case information to obtain a historical mediation case feature data set; processing according to the historical mediation case feature data set to obtain a mediation protocol template library; obtaining to-be-mediated case information and carrying out feature extraction processing to obtain input feature data; performing pattern recognition and fuzzy reasoning processing according to the input feature data to obtain a fuzzy reasoning result; performing matching processing and screening based on a fuzzy reasoning result to obtain a preliminary mediation protocol; and performing optimization processing on the preliminary mediation protocol according to the input feature data to obtain a personalized mediation protocol. According to the method, the mediation protocol template library is dynamically constructed and optimized, and the mediation protocol compiling time is greatly shortened by combining the automatic matching and personalized generation method of the mediation case characteristics.

Description

Technical Field

[0001] The present invention relates to the field of intelligent legal technology, and in particular to a method and system for automatically generating a personalized mediation agreement based on mediation case characteristics. Background Art

[0002] With the continuous development of social economy and the popularization of legal awareness, the number of social disputes has shown an increasing trend year by year. Mediation, as a non-litigation dispute resolution method, has attracted more and more attention and attention from all walks of life due to its fast, efficient, economical and flexible characteristics. In particular, in a large number of civil and commercial disputes, family conflicts, labor disputes and other diverse social issues, mediation has gradually become an important solution path, which has played a positive role in maintaining social harmony and stability. However, in traditional mediation practice, the preparation of mediation agreements mainly relies on the personal experience and manual writing of mediators, which has many limitations.

[0003] First, the process of manually writing mediation agreements is cumbersome and time-consuming, which not only increases the workload of mediators, but also may lead to inefficiency and make it difficult to meet the growing demand for dispute mediation. Secondly, due to human negligence or lack of legal expertise, the terms of the mediation agreement may be imprecise or incomplete, and may even lead to execution disputes, further damaging the rights and interests of the parties. In addition, manually written mediation agreements are usually based on template ideas, which make it difficult to accurately capture the specific characteristics of the case and the personalized needs of the parties, resulting in a lack of pertinence in the content of the agreement, thereby affecting the success rate of mediation and the effectiveness of the implementation of the agreement. Faced with rapidly changing social needs and legal environment, the traditional manual writing method is obviously unable to meet the complex and changing mediation scenarios. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for automatically generating a personalized mediation agreement based on mediation case characteristics to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a method for automatically generating a personalized mediation agreement based on mediation case characteristics, comprising:

[0006] Obtaining historical mediation case information, including case type, dispute amount, party behavior pattern, case background, mediation needs, and mediation results;

[0007] Performing structured processing on the historical mediation case information, and converting various types of information in the case into a standardized data format, thereby obtaining a historical mediation case feature data set;

[0008] Constructing a semantic relationship network between case features according to the historical mediation case feature data set, and processing the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library;

[0009] Acquire information of cases to be mediated, and perform feature extraction processing on the information of cases to be mediated to obtain input feature data;

[0010] Performing pattern recognition and fuzzy reasoning processing based on the input feature data, predicting the behavioral responses of the parties in the case through a preset sociological model, and using a fuzzy reasoning algorithm to reason and analyze the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results;

[0011] Perform matching processing based on the fuzzy reasoning result, and obtain a preliminary mediation agreement by screening from the mediation agreement template library;

[0012] The preliminary mediation agreement is optimized according to the input feature data, and the mediation agreement text is automatically generated by utilizing the generator part of the generative adversarial network model. The discriminator part evaluates the legality and rationality of the generated agreement clauses to obtain a personalized mediation agreement.

[0013] In a second aspect, the present application also provides a system for automatically generating a personalized mediation agreement based on the characteristics of the mediation case, including:

[0014] An acquisition module is used to acquire historical mediation case information, wherein the historical mediation case information includes case type, dispute amount, party behavior pattern, case background, mediation demand and mediation result;

[0015] A conversion module, used to perform structured processing based on the historical mediation case information, and obtain a historical mediation case feature data set by converting various types of information in the case into a standardized data format;

[0016] A construction module, used to construct a semantic relationship network between case features based on the historical mediation case feature data set, and process the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library;

[0017] An input module, used to obtain information of cases to be mediated, and to perform feature extraction processing on the information of cases to be mediated to obtain input feature data;

[0018] An analysis module is used to perform pattern recognition and fuzzy reasoning processing based on the input feature data, predict the behavioral responses of the parties in the case through a preset sociological model, and use a fuzzy reasoning algorithm to perform reasoning and analysis on the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results;

[0019] A matching module, performing matching processing based on the fuzzy reasoning result, and obtaining a preliminary mediation agreement by screening from the mediation agreement template library;

[0020] The output module is used to optimize the preliminary mediation agreement according to the input feature data, automatically generate the mediation agreement text by utilizing the generator part of the generative adversarial network model, and the discriminator part evaluates the legality and rationality of the generated agreement clauses to obtain a personalized mediation agreement.

[0021] The beneficial effects of the present invention are:

[0022] The present invention significantly reduces the time required to write mediation agreements by dynamically constructing and optimizing a mediation agreement template library, and by automatically matching and personalizing the generation method based on the characteristics of mediation cases. By combining natural language processing and machine learning technologies, the present invention can intelligently generate mediation agreements with rigorous clauses, clear logic, and that meet the needs of the parties, thus avoiding the problem of inaccuracy caused by negligence in traditional manual writing. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a method flow for automatically generating a personalized mediation agreement based on mediation case characteristics in an embodiment of the present invention;

[0025] Figure 2 Schematic diagram of the system structure for automatically generating a personalized mediation agreement based on mediation case characteristics in an embodiment of the present invention

[0026] Figure 3 Schematic diagram of the structure of a device for automatically generating a personalized mediation agreement based on mediation case characteristics as described in an embodiment of the present invention.

[0027] Markings in the figure: 800, a device for automatically generating a personalized mediation agreement based on the characteristics of the mediation case; 801, a processor; 802, a memory; 803, a multimedia component; 804, an I / O interface; 805, a communication component; 901, an acquisition module; 902, a conversion module; 903, a construction module; 904, an input module; 905, an analysis module; 906, a matching module; 907, an output module;. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0030] Embodiment 1:

[0031] This embodiment provides a method for automatically generating a personalized mediation agreement based on mediation case characteristics.

[0032] See also Figure 1 , the figure shows that the method includes steps S100 to S700.

[0033] Step S100, obtaining historical mediation case information, the historical mediation case information including case type, dispute amount, party behavior pattern, case background, mediation demand and mediation result;

[0034] Understandably, the focus of this step is to obtain case information comprehensively and accurately to ensure the integrity and authenticity of the data. Each type of information has a unique role: case type is used for classification, dispute amount provides numerical basis, party behavior patterns reveal the dynamic characteristics of the case, case background and mediation needs determine the specific characteristics and goals of the case, and mediation results provide feedback for subsequent model optimization and dynamic updating of templates. The collection of these data is the basis of automated processing and directly affects the reliability and efficiency of the entire system.

[0035] Step S200: Performing structured processing on historical mediation case information, converting various types of information in the case into a standardized data format, and obtaining a historical mediation case feature data set;

[0036] It should be noted that in the implementation process, different types of data are first classified and processed: for text information (such as case background, mediation needs, etc.), natural language processing technology is used to extract key information; for numerical information (such as dispute amount), regular expressions or rule matching are used to achieve numerical processing; for category information (such as case type, mediation results), classification or coding technology is applied for standardized representation. Then, redundant or erroneous information is removed through data cleaning to ensure data consistency.

[0037] Step S300: construct a semantic relationship network between case features based on the historical mediation case feature data set, and process the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library;

[0038] It should be noted that this step uses the construction and analysis of semantic relationship networks to effectively integrate the correlation information between case features and generate a mediation agreement template library with high adaptability and accuracy. This not only improves the intelligence level of template generation, but also ensures that the template library can respond to changes in laws and regulations and the diversification of case needs in a timely manner through a dynamic learning mechanism.

[0039] Step S400, obtaining information of cases to be mediated, and performing feature extraction processing on the information of cases to be mediated to obtain input feature data;

[0040] It can be understood that this step structures and quantifies the information of the new case to be mediated and extracts the core features that are helpful for matching and generating agreements.

[0041] Step S500: Perform pattern recognition and fuzzy reasoning processing based on the input feature data, predict the behavioral responses of the parties in the case through a preset sociological model, and use a fuzzy reasoning algorithm to reason and analyze the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results;

[0042] It should be noted that this step, through the comprehensive use of pattern recognition, sociological models and fuzzy reasoning, can deeply analyze the complex characteristics of the case, transform unstructured or fuzzy information into clear reasoning results, provide a scientific basis for the precise matching and generation of protocols, and significantly improve the system's ability to handle complex cases.

[0043] Step S600: Perform matching processing based on the fuzzy reasoning result, and obtain a preliminary mediation agreement from the mediation agreement template library;

[0044] It can be understood that by using the matching mechanism of fuzzy reasoning results and template features, this step can quickly and efficiently screen out the most suitable mediation agreement template, and dynamically adjust the terms in accordance with the personalized needs of the case, which significantly improves the accuracy and pertinence of agreement generation, while reducing the workload of manual participation.

[0045] Step S700: Optimize the preliminary mediation agreement according to the input feature data, automatically generate the mediation agreement text by using the generator part of the generative adversarial network model, and evaluate the legality and rationality of the generated agreement clauses by the discriminator part to obtain a personalized mediation agreement.

[0046] It can be understood that this step realizes the dynamic optimization and adaptive generation of the mediation agreement through the generative adversarial network, which not only meets the personalized needs of the case, but also ensures the legality and rationality of the agreement text.

[0047] Further, step S200 includes step S210 to step S250.

[0048] Step S210: Perform standardization processing according to the case type, extract the case type from the case description through text analysis and classification algorithm, and convert the case category into numerical category data to obtain standardized case type data;

[0049] Preferably, this step adopts TF-IDF feature extraction combined with support vector machine (SVM) classification method to achieve case type standardization. Specifically, improvements are made to the specific characteristics of the case text to improve the accuracy and adaptability of the classification. The mediation case text has significant legal terms and structured patterns, such as high-frequency words such as "contract dispute" and "bilateral agreement", which may be given lower weights in conventional algorithms. Therefore, by constructing a custom stop word list, filtering out high-frequency but low-discrimination words such as "bilateral" and "agreement", and introducing industry-specific vocabulary lists, giving higher weights to specific case types (such as "liquidated damages" in "contract dispute" or "compensation amount" in "property dispute"), the classification model can significantly improve its ability to identify key features. In addition, by combining numerical features such as the disputed amount and the behavior patterns of the parties with text features to form a multimodal feature vector, the classifier's ability to distinguish cases with similar texts but different features is further enhanced. In view of the phenomenon that case categories may overlap (such as "contract disputes" and "property disputes"), a multi-label classification framework is adopted to map case feature vectors to multiple categories at the same time, solve the problem of blurred category boundaries, and model the correlation between multiple labels to improve the classification effect of complex cases. In model training, the generalization ability and overfitting risk of the classifier are balanced by adjusting the regularization parameters of the support vector machine, and a weight adjustment strategy is introduced for the problem of category imbalance to ensure sufficient recognition ability for small category cases. These improvements not only solve the problems of high-frequency interference of legal terms and blurred cross-category boundaries, but also significantly improve the robustness, accuracy and practical application value of the mediation case classification model.

[0050] Step S220: numerically process the dispute amount, extract the dispute amount information in the case through regular expressions and convert it into a numerical variable to obtain standardized dispute amount data;

[0051] It should be noted that for the text data in mediation cases, since the disputed amount often exists in an unstructured form, it is difficult for traditional methods to directly use this information for numerical processing. Therefore, this step realizes the automatic extraction and standardization of the amount through regular expression matching and conversion rules. Specifically, first use regular expressions to identify the amount information in the text. The regular expression design needs to adapt to the characteristics of the case text and can flexibly match multiple formats, such as: digital format (for example, "5 million yuan"), by matching continuous numbers and combining unit vocabulary. Text description (for example, "one thousand US dollars"), by combining natural language processing technology, the text amount is converted into digital form. With currency symbols (for example, "$1,000"), by expanding the matching rules for multiple currency symbols that may be involved in international cases. Secondly, after extracting the amount text, the unit is standardized. For example, "ten thousand yuan" is converted into the basic unit "yuan", and "US dollars" or other currencies are uniformly converted into a predefined base currency (such as RMB). This process requires calling an exchange rate conversion tool or a rule-based unit mapping table for processing. Finally, the cleaned and converted amount information is encoded into a numerical variable, and the original amount and the converted numerical representation are retained to form standardized dispute amount data. These data are not only used for subsequent classification and matching, but can also directly participate in the quantitative analysis of the focus of the dispute, improving the algorithm's sensitivity and processing capabilities for economic factors. This step greatly reduces the need for manual intervention by automatically extracting and standardizing the dispute amount data, while ensuring the accuracy and consistency of the amount information. In mediation cases, the standardized dispute amount data provides accurate numerical input for the model, significantly improving the system's ability to model the economic characteristics of the case and the quality of subsequent mediation agreement clause generation.

[0052] Step S230: sentiment analysis and classification processing is performed based on the behavior patterns of the parties involved. The BERT algorithm is used to analyze and classify the sentiment tendency of the behavior descriptions of the parties involved in the case, and labeling is performed in combination with the mediation results to obtain structured behavior pattern data;

[0053] In this step, sentiment analysis and classification techniques are used to mine the sentiment tendencies and behavior types of the descriptions of the parties' behaviors in the case text (such as "actively cooperating with mediation" and "refusing to compromise"). Specifically, the case description text is first semantically encoded through the pre-trained BERT (Bidirectional Encoder Representation Transformer) model to generate a high-dimensional semantic vector representation. This process can fully capture context-dependent information and is particularly suitable for complex behavioral expressions in mediation case texts. For example, for "Party A is resolute and does not accept the other party's conditions", BERT can capture the semantic relationship between "resolute" and "accept" and the context, and extract the parties' resistance to mediation. Secondly, a classification model (such as a fully connected neural network based on deep learning) is used to perform sentiment tendency analysis and behavior classification on the semantic vector. Sentiment tendency analysis distinguishes the emotional polarity of the parties (positive, neutral or negative), and behavior classification divides the parties' behavior patterns into cooperative, confrontational, neutral and other categories. Combined with emotion-related modifiers and verbs in the case text (such as "anger" and "accept"), the model can further improve the accuracy of classification. Finally, the sentiment analysis and behavior classification results are combined with the mediation results for labeling. By statistically analyzing the mediation results in historical case data, a correlation model between behavior patterns and mediation success rates is constructed to generate corresponding labels for each behavior pattern. For example, "cooperative" may be strongly correlated with "mediation success", while "confrontational" is related to "mediation failure". These labeled behavior pattern data are stored in a structured form to support subsequent fuzzy reasoning and protocol generation.

[0054] Step S240: extract features based on the case background and mediation needs, extract the dispute causes and appeal points in the case background, and use the TF-IDF algorithm to analyze the keywords and sentiment distribution of the case description to obtain structured mediation demand data;

[0055] In practical applications, the case description text is first preprocessed, including word segmentation, stop word removal, and part-of-speech tagging. Then, for the cause of the dispute and the point of appeal, descriptive sentence fragments are extracted by constructing key sentence extraction rules (such as sentences containing words such as "dispute focus", "core appeal" or "main issue"). For example, "Party A believes that Party B has not fulfilled its delivery obligations in accordance with the contract" can extract the focus of the dispute as "failure to fulfill delivery obligations in accordance with the contract". Next, the TF-IDF algorithm is used to calculate the weights of the keywords in the case description to capture the core semantic information in the case. Finally, combined with sentiment analysis technology, the extracted key sentences and keywords are classified into sentiment tendencies to determine the attitudes of the parties in the case towards specific appeals (such as positive, negative or neutral). For example, "Party B firmly denies responsibility" indicates a negative attitude towards the dispute point, while "Party A is willing to compromise" shows a positive tendency. Through the above steps, structured mediation demand data is generated, which contains the cause of the dispute, the point of appeal, and the sentiment tendency. These data provide a clear case semantic and sentiment basis for the subsequent generation of mediation agreements.

[0056] Step S250: Integrate the standardized case type data, the standardized dispute amount data, the structured behavior pattern data and the structured mediation demand data to obtain a historical mediation case feature data set.

[0057] During the integration process, each feature data needs to be aligned to a unified dimension. To this end, feature standardization or normalization technology can be used to adjust numerical data (such as disputed amounts) to the same scale range. For example, the minimum-maximum normalization is used to normalize the amount value to the [0,1] interval to avoid weight imbalances caused by large numerical differences between different features. For categorical and textual features (such as case types and mediation needs), they are combined with numerical data into a complete feature vector through feature splicing. The final historical mediation case feature data set is stored in the form of a high-dimensional vector, in which the feature vector of each case contains complete multi-dimensional information such as case type, disputed amount, behavior pattern, and mediation needs, providing comprehensive data support for subsequent model training, feature association analysis, and protocol generation.

[0058] Further, step S300 includes step S310 to step S340.

[0059] Step S310: perform feature association based on the historical mediation case feature data set, construct a co-occurrence matrix, and apply a word embedding algorithm to vectorize the case features, convert non-numerical data into numerical vectors, and obtain a semantic vector representation of the case features;

[0060] Specifically, we first construct a co-occurrence matrix based on the historical mediation case feature dataset. Assume that there is a set of case feature datasets, and set the co-occurrence matrix C to represent the co-occurrence relationship between features. There are N case features (f1, f2, …, f N ), then the co-occurrence matrix C is an N×N matrix, where each element C ij Represents feature f i and feature f j The co-occurrence frequency in the case data set. The calculation formula is:

[0061]

[0062] Where M is the number of cases; k is the case number; D k represents the kth case; f i represents the i-th feature; f j represents the jth feature; C ij Represents feature f i and feature f j In case D k The co-occurrence frequency in i ,f j ,D k ) represents the indicator function. If in case D k In the feature f i and f j If they appear at the same time, then I(f i ,f j ,D k )=1, otherwise it is 0.

[0063] Next, we apply a word embedding algorithm to vectorize the co-occurrence matrix, with the goal of learning a low-dimensional vector representation of each case feature. This makes features with high co-occurrence frequency have a closer distance in the vector space. Assume that feature f i The word embedding vector is represented as v i , the co-occurrence matrix c is combined with the word embedding algorithm for training in the following way:

[0064]

[0065] in, Represents feature f i The word embedding vector of p(C ij |v i ,v j )) represents the feature f i and feature f j The co-occurrence probability in the co-occurrence matrix C is modeled using the softmax function:

[0066]

[0067] Among them, V represents the word embedding matrix of all features, which is the parameter that needs to be optimized during training; v k Represents feature f k The word embedding vector of .

[0068] Through training, each non-numerical feature is mapped to the semantic vector space to generate a semantic vector representation of the feature. This vectorization process not only captures the co-occurrence relationship between features, but also learns the potential semantic association of features through word embedding technology. This semantic vector representation provides high-quality input for subsequent feature clustering and model training.

[0069] Step S320: performing clustering processing according to the semantic vector representation, and obtaining at least two case feature clusters by aggregating case features with similar semantics;

[0070] It can be understood that this step aggregates the scattered feature vectors into feature clusters with semantic consistency through clustering. This process reduces the complexity of feature dimensions and reveals the semantic structure between case features, laying an important foundation for the subsequent construction of semantic relationship networks and the design of mediation agreement templates.

[0071] Step S330: Perform network construction processing based on all case feature clusters, establish semantic connection relationships between features through a graph neural network, and obtain a semantic relationship network, where each node in the semantic relationship network represents a case feature, and each edge represents association information between features;

[0072] Specifically, first, the nodes in the semantic relationship network represent the case characteristics. For example, each node can correspond to specific case characteristics such as "contract breach" and "rent arrears". These nodes come from the feature clusters in the clustering results, and each feature in the feature cluster has similar semantic information. Then, by calculating the similarity within and between feature clusters, the edge weights between nodes are determined. The edge weights can be calculated using semantic similarity (such as cosine similarity) or a weighted method based on co-occurrence relationships. The directionality and weight of the edge indicate the strength of the association between features. For example, a high weight may indicate that two features frequently co-occur in the case description.

[0073] Then, the semantic relationship network is processed using a graph neural network. The graph neural network updates the feature representation of each node to reflect the semantic information of its neighboring nodes through a message passing mechanism. The node update formula is:

[0074]

[0075] in, represents the feature representation of node u after the t+1th iteration; represents the neighbor set of node u; d u represents the number of neighbors of node u; W t represents the weight matrix of the tth layer; σ represents the activation function, such as the ReLU function; v represents the neighbor node of node u.

[0076] Through multiple layers of iteration, node features gradually integrate the information of their neighboring nodes to achieve a deep expression of the semantic association between features. The resulting semantic relationship network can not only represent the direct connection between case features, but also capture their implicit high-order semantic relationships.

[0077] Step S340: Generate templates based on the semantic relationship network, use convolutional neural network to process the semantic relationship network of case features, automatically identify key clauses and template structures in the mediation agreement through training models, generate standardized mediation agreement templates suitable for different case types, and obtain the final mediation agreement template library.

[0078] Specifically, the feature node and edge representations in the semantic relationship network are converted into the tensor input form of graph data. The feature vector of each node contains semantic features, while the edge weight represents the semantic association between nodes. The input data is processed by the graph convolution layer, and the graph convolution operation can capture the local semantic structure between nodes and their neighbors. Through multiple layers of graph convolution, the feature representation of each node in the semantic relationship network is further fused and optimized, and finally the global semantic relationship of key features is captured. Next, the key terms of the mediation agreement template are identified through the feature extraction capability of the convolutional neural network. The convolutional neural network uses the processed feature graph in the network input to automatically generate template suggestions containing the content, order and structural relationship of the terms. For example, different case types may require specific terms (such as compensation terms, behavioral agreements), which are identified and extracted through weight learning of the convolutional neural network layer. Finally, the generated templates are normalized to form a standardized agreement template library. The template library generates a dynamic structured framework for different types of cases, which can be quickly matched and applied according to the characteristics of the case.

[0079] Further, step S400 includes step S410 to step S430.

[0080] Step S410: Perform text processing and feature extraction on the case information to be mediated based on a preset natural language processing model to obtain semantic feature data of the case;

[0081] Specifically, the case text information is first preprocessed, including operations such as word segmentation, stop word removal, word form restoration, and unified formatting. Preprocessing ensures the consistency and efficiency of text data, laying the foundation for subsequent feature extraction. Then, the preprocessed text is deeply semantically parsed through the preset natural language processing model. Commonly used models such as the BERT model based on the Transformer architecture generate dynamic word vectors for each word through a context-dependent mechanism. This method can capture the implicit semantic relationships in the case description, and through these vectorized semantic features, the key semantic information in the case text can be fully represented. Subsequently, the model uses named entity recognition (NER) to extract specific case entities. These entities are further mapped to fixed feature labels, such as "contract dispute" and "compensation amount", which adds structured information to text analysis. Finally, the extracted semantic features are subjected to dimensionality reduction processing to reduce redundant information and optimize feature representation to generate semantic feature data. The data is stored in the form of high-dimensional vectors, which can comprehensively reflect the text features and semantic connotations of the case.

[0082] Step S420, performing standardization processing according to the quantitative features in the case information to be mediated to obtain numerical feature data;

[0083] It is understandable that by normalizing the units and scales of quantitative features, the impact caused by dimensional differences between features is effectively reduced. The standardized data not only improves the stability of model processing, but also enhances the expressiveness of features in analysis and prediction, providing a solid foundation for the system to handle cases of different scales and complexities.

[0084] Step S430: Integrate the semantic feature data and the numerical feature data to obtain integrated feature data, and use a principal component analysis algorithm to perform dimensionality reduction processing on the integrated feature data to obtain input feature data.

[0085] Further, step S500 includes step S510 to step S530.

[0086] Step S510: Perform pattern recognition processing according to the input feature data, use a deep neural network to train the input features after dimensionality reduction, extract potential patterns in the data to obtain pattern recognition results;

[0087] It should be noted that semantic feature data is usually a high-dimensional vector that represents the semantic content in the case text; numerical feature data is a low-dimensional numerical variable, such as the disputed amount and case time. During the integration process, these two types of features need to be normalized to maintain the balance of feature importance, such as mapping semantic features and numerical features to the same scale interval (such as [0,1]). The integrated feature representation is a high-dimensional joint feature vector, which contains both the textual semantic information of the case and the numerical feature information. Next, the principal component analysis (PCA) algorithm is used to reduce the dimensionality of the integrated feature data. The goal of principal component analysis is to project high-dimensional data into a space with lower dimensions but retain as much original information as possible through linear transformation. Specifically, principal component analysis calculates the covariance matrix of the feature data, decomposes its eigenvalues ​​and eigenvectors, and selects the first few principal components with a higher cumulative explained variance ratio as the new feature representation. This not only reduces the redundancy of the features, but also reduces the computational complexity of the model. The input feature data after dimensionality reduction is a set of low-dimensional vectors, which not only retains the core information of the case characteristics, but also removes noise and redundant features, providing efficient input for subsequent model processing.

[0088] Step S520: Perform behavior prediction processing based on the pattern recognition results, perform time series modeling through long short-term memory networks, analyze the changes in the attitudes and willingness to cooperate of the parties in the case over time, and predict future behavior trends to obtain behavior prediction results;

[0089] Specifically, first, the pattern recognition results provide key features of the case, such as the parties' emotional tendencies (positive, neutral, or negative), behavior types (cooperative or confrontational), and the evolution of the disputed points of the case. These features are represented in the form of time series to form input data. For example, the description of "Party A was more cooperative in the early stage of mediation, but disputes arose over compensation issues" can be quantified as a time series: [cooperation, cooperation, confrontation]. Then, the long short-term memory network is used to model these time series data. The long short-term memory network is an extension of the recurrent neural network (RNN) that is good at processing time series data and can capture long-distance dependencies between features. For input features, such as the current behavioral features of the parties, the long short-term memory network selectively updates and retains feature memory through a gating mechanism (input gate, forget gate, and output gate) to generate hidden states for each time step. This mechanism allows the model to infer the future trend of the parties' behavior based on historical features. During the training process, the long short-term memory network minimizes the gap between the predicted behavior and the actual behavior through the loss function, and gradually learns the dynamic change pattern of case features. For example, through time step prediction, we can obtain the behavioral trend that "when the dispute over the amount of compensation continues to expand, the parties' confrontational emotions increase."

[0090] Step S530: Perform fuzzy reasoning on the behavior prediction results based on preset fuzzy rules, and obtain reasoning results based on the dispute focus and behavior patterns of the parties in the case.

[0091] It can be understood that this step converts these input variables into fuzzy sets based on the variable inputs provided by the behavior prediction results (such as cooperation probability, confrontation probability), combined with the dispute focus of the case (such as "compensation amount" or "performance method") and the behavior patterns of the parties (such as "cooperative" or "confrontational"). Subsequently, the input variables are reasoned according to the preset fuzzy rule base (for example, "if the behavior of the parties is confrontational and the dispute focus is the amount of compensation, the mediation difficulty is high"). Fuzzy reasoning includes the steps of fuzzification, rule evaluation, reasoning calculation and defuzzification. The activation strength of the rules is calculated through fuzzy sets and the reasoning output is generated. Finally, the fuzzy results are interpreted as clear mediation strategy recommendations or mediation difficulty assessments. This step flexibly handles the uncertainty and complex characteristics of the case through fuzzy reasoning, and draws key conclusions such as the priority of the focus of the dispute and the possibility of successful mediation.

[0092] Further, step S600 includes step S610 to step S640.

[0093] Step S610: weighting the case features according to the fuzzy reasoning results, by weighting various fuzzy features in the case and combining the reasoning confidence corresponding to the fuzzy features to obtain a weighted case feature vector;

[0094] This step enhances the accuracy of case analysis by weighting case features in combination with fuzzy reasoning results. The weighted feature vector not only fully reflects the semantic features of the case and the importance of the dispute focus, but also effectively improves the system's ability to analyze complex case information.

[0095] Step S620: perform similarity calculation based on the weighted case feature vector, and obtain a similarity score result by calculating the similarity between the case feature vector and the template feature vector. The similarity score result includes the matching degree between each mediation agreement template and the current case.

[0096] It should be noted that this step quantifies the semantic consistency between the case and the template by calculating the similarity of the feature vector, providing an intuitive and operational basis for template selection. This process combines the priority of weighted features to ensure the dominant role of key features in matching, while effectively improving the accuracy and pertinence of the system in template recommendation.

[0097] Step S630: sorting the templates in the mediation agreement template library according to the similarity scoring results, and selecting the optimal mediation agreement template based on the sorting results;

[0098] Specifically, first, similarity scoring results are received, which indicate the degree of adaptation of the current case to each template in the template library. The scoring results are presented in numerical form, and the higher the score, the closer the template is to the characteristics of the case. The templates are sorted from high to low according to the scores to generate an ordered list of templates. Then, the template with the highest matching degree is selected according to the sorting results. Furthermore, additional constraints are combined in the screening process, such as the legal applicability, geographical characteristics or historical use of the template. If the case involves multiple dispute points, the system can also integrate the similarities of multiple feature dimensions and use a weighted scoring method to comprehensively sort the templates to ensure that the selected templates can fully adapt to the needs of the case. Finally, the highest-ranked template is selected as the optimal mediation agreement template for the current case for use in subsequent links. This template not only makes full use of the correlation of historical data, but also can dynamically adapt to the personalized characteristics of the case.

[0099] Step S640: Generate an agreement based on the optimal mediation agreement template, adjust the terms in the template based on the specific needs of the case, and generate a preliminary mediation agreement.

[0100] It should be noted that this step first loads the selected optimal template, which contains the general mediation agreement clauses and framework structure. The optimal template provides the basic framework of the agreement, but it still needs to be further adjusted according to the actual needs of the case. For example, the template may have preset content for compensation clauses, performance clauses or confidentiality clauses, which need to be refined or modified according to the dispute focus of the case or the wishes of the parties. Then the specific content of the template clauses is automatically adjusted according to the characteristics of the case (such as the dispute amount, the behavior pattern of the parties and the mediation needs). For example, for cases with large dispute amounts, the compensation clauses in the template may require a more detailed installment payment plan; while for cases where the parties tend to cooperate, flexibility agreements can be added to the performance clauses. This dynamic adjustment process can be achieved through a rule engine or a data-driven algorithm to map the case characteristics to the modification logic of the template clauses. Finally, a preliminary mediation agreement is generated based on the adjusted template clauses. The agreement text retains legal rigor while fully reflecting the specific needs of the case and the intentions of the parties.

[0101] Further, step S700 includes step S710 to step S730.

[0102] Step S710: performing content generation processing according to the input feature data, capturing semantic information of case features based on a preset generative adversarial network, and personalizing the terms of the preliminary agreement to generate a draft mediation agreement;

[0103] First, input feature data (such as case type, dispute amount, party behavior patterns, and mediation needs) are used as input to the generative network. After preprocessing, these features are mapped into semantic vector representations to guide the generation process. The generative adversarial network consists of two main parts: the generator and the discriminator. The generator receives the input features and generates the terms through a deep neural network, including compensation terms, performance terms, etc., to ensure that the generated agreement content meets the basic characteristics and needs of the case.

[0104] Next, the discriminator evaluates the clauses output by the generator to determine whether they are true and reasonable. The discriminator receives the generated draft agreement and compares it with the actual historical agreement clauses. By learning to distinguish the semantic and logical differences between the two, it continuously adjusts the output of the generator. During the adversarial training process of generation and discrimination, the generator gradually improves the quality and authenticity of the generated clauses, and the discriminator helps to exclude clauses that do not meet actual needs. In this process, personalized adjustments are mainly reflected in semantic consistency, flexibility, and contextual relevance. Semantic consistency means that the generator ensures that the content of the clauses is semantically consistent with the characteristics of the case based on the input features. For example, more rigorous compensation clauses are generated for cases with larger dispute amounts. Flexibility means that the generated clauses can automatically adjust the wording according to the behavior patterns of the parties (such as cooperative or adversarial) to adapt to the wishes of the parties. Contextual relevance is reflected in the dynamic addition or adjustment of special clauses according to the case type, such as the need to add a confidentiality agreement in intellectual property cases. Finally, the generated draft mediation agreement contains a complete clause structure, and the content has been optimized to match the case characteristics, providing a high-quality foundation for subsequent legality and rationality evaluation.

[0105] Step S720: Evaluate the draft mediation agreement, use the discriminant network of the generative adversarial network to judge the legality and rationality of the generated agreement clauses, check the agreement clauses based on the preset legal rules, industry standards and historical cases, judge whether the agreement meets the legality requirements and whether it is reasonable, and obtain the legality-rationality evaluation result;

[0106] It can be understood that in this step, the discriminant network in the adversarial network is first trained as an evaluator, receiving the content of the terms of the draft mediation agreement and comparing it with the legal agreement terms in historical cases in terms of semantics and structure. The output value of the discriminant network represents the probability that the agreement terms conform to the historical agreement model, which is used as a preliminary basis for judging the legality and rationality of the agreement. For example, if the language structure of the terms is consistent with the historically successfully executed agreement, it is given a higher credibility score. Then, the agreement terms are checked by the preset legal rules and industry standards. This process includes matching the compliance of key elements in the terms (such as the upper limit of the compensation amount, the performance period, the dispute resolution method, etc.) with the relevant legal provisions. For example, for compensation terms with a large amount of dispute, the system will check whether it contains mandatory enforcement clauses and whether it complies with the compensation scope prescribed by the law. At the same time, industry standards can provide field-specific supplementary rules, such as whether the labor dispute agreement contains appropriate overtime wage compensation clauses. Finally, combined with the analysis of historical cases, the rationality of the agreement terms is judged. This includes a balance check on the wording of the terms, such as whether it is too biased towards the interests of one party, whether there are potential enforcement difficulties, etc. Reasonableness assessment not only focuses on legal compliance, but also considers fairness and operability in actual implementation.

[0107] Step S730: Feedback adjustment is performed according to the legality-rationality evaluation results, the generation strategy of the generative adversarial network is adjusted through the reinforcement learning algorithm, and the final personalized mediation agreement is output.

[0108] In practical applications, the output of the legality-reasonableness evaluation result is first received, which contains the feedback score of the agreement clause in terms of legal compliance and practical rationality. The reinforcement learning algorithm uses this evaluation result as a reward signal to guide the optimization process of the generative network. Specifically, the generative network is regarded as an "agent" in reinforcement learning, and its goal is to generate an agreement clause that receives a higher reward in the discriminant network and rule evaluation. The reward function is designed based on the legality and rationality scores of the clauses. For example, clauses with high legality weights receive positive rewards; clauses with legal conflicts or logical loopholes are punished.

[0109] The generative network then updates the generation strategy using a policy gradient algorithm or a Q-learning algorithm. The generative network adjusts the parameters of its generative model through iterative training to gradually improve the quality of the generated content. For example, through training, the characteristics of clauses in specific types of cases are learned, such as the legal limits on the amount of compensation in a labor dispute agreement, or the general wording of the performance clauses in a contract dispute agreement. After multiple rounds of feedback optimization, a fully adjusted and more personalized mediation agreement text can be generated. The final output mediation agreement not only fully meets the legal requirements of the case, but also accurately reflects the unique needs of the case and the wishes of the parties.

[0110] Embodiment 2:

[0111] like Figure 2 As shown, this embodiment provides a system for automatically generating a personalized mediation agreement based on mediation case characteristics, and the system includes:

[0112] The acquisition module 901 is used to acquire historical mediation case information, which includes case type, dispute amount, party behavior pattern, case background, mediation demand and mediation result;

[0113] The conversion module 902 is used to perform structured processing based on the historical mediation case information, and obtain a historical mediation case feature data set by converting various types of information in the case into a standardized data format;

[0114] A construction module 903 is used to construct a semantic relationship network between case features based on the historical mediation case feature data set, and process the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library;

[0115] An input module 904 is used to obtain information about cases to be mediated, and to extract features from the information about cases to be mediated to obtain input feature data;

[0116] The analysis module 905 is used to perform pattern recognition and fuzzy reasoning processing based on the input feature data, predict the behavioral responses of the parties in the case through a preset sociological model, and use a fuzzy reasoning algorithm to perform reasoning and analysis on the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results;

[0117] A matching module 906 performs matching processing based on the fuzzy reasoning result, and obtains a preliminary mediation agreement by screening from the mediation agreement template library;

[0118] The output module 907 is used to optimize the preliminary mediation agreement according to the input feature data, automatically generate the mediation agreement text by utilizing the generator part of the generative adversarial network model, and evaluate the legality and rationality of the generated agreement clauses by the discriminator part to obtain a personalized mediation agreement.

[0119] In a specific embodiment disclosed in the present invention, the conversion module 902 includes:

[0120] The first processing unit is used to perform standardized processing according to the case type, extract the case type from the case description through text analysis and classification algorithm and convert the case category into numerical category data to obtain standardized case type data;

[0121] The second processing unit is used to perform numerical processing according to the dispute amount, extract the dispute amount information in the case through regular expressions and convert it into a numerical variable to obtain standardized dispute amount data;

[0122] The first analysis unit is used to perform sentiment analysis and classification based on the behavior patterns of the parties. The BERT algorithm is used to analyze and classify the sentiment tendency of the behavior descriptions of the parties in the case, and the mediation results are combined for labeling to obtain structured behavior pattern data;

[0123] The first extraction unit is used to extract features based on the case background and mediation needs. By extracting the causes of disputes and appeals in the case background and using the TF-IDF algorithm to analyze the keywords and sentiment distribution of the case description, structured mediation demand data is obtained;

[0124] The first integration unit is used to integrate standardized case type data, standardized dispute amount data, structured behavior pattern data and structured mediation demand data to obtain a historical mediation case feature data set.

[0125] In a specific implementation of the present invention, the construction module 903 includes:

[0126] The first association unit is used to perform feature association based on the historical mediation case feature data set, construct a co-occurrence matrix, and apply a word embedding algorithm to vectorize the case features, thereby converting non-numerical data into numerical vectors to obtain a semantic vector representation of the case features;

[0127] A first clustering unit is used to perform clustering processing according to the semantic vector representation, and obtain at least two case feature clusters by aggregating case features with similar semantics;

[0128] The first construction unit is used to perform network construction processing based on all case feature clusters, establish semantic connection relationships between features through a graph neural network, and obtain a semantic relationship network. Each node in the semantic relationship network represents a case feature, and each edge represents association information between features;

[0129] The first generation unit is used to generate templates based on the semantic relationship network, use a convolutional neural network to process the semantic relationship network of case features, automatically identify key clauses and template structures in the mediation agreement through a training model, generate standardized mediation agreement templates suitable for different case types, and obtain a final mediation agreement template library.

[0130] Embodiment 3:

[0131] Corresponding to the above method embodiment, this embodiment also provides a device for automatically generating a personalized mediation agreement based on mediation case characteristics. The device for automatically generating a personalized mediation agreement based on mediation case characteristics described below and the method for automatically generating a personalized mediation agreement based on mediation case characteristics described above can refer to each other.

[0132] Figure 3 FIG. 8 is a block diagram of a device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics according to an exemplary embodiment. Figure 3 As shown, the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics may include: a processor 801, a memory 802. The device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0133] The processor 801 is used to control the overall operation of the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics, so as to complete all or part of the steps in the method for automatically generating a personalized mediation agreement based on mediation case characteristics. The memory 802 is used to store various types of data to support the operation of the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics, such as instructions for any application or method operating on the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, which is used to receive external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the device 800 that automatically generates a personalized mediation agreement based on the characteristics of the mediation case and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0134] In an exemplary embodiment, the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned method of automatically generating a personalized mediation agreement based on mediation case characteristics.

[0135] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the method for automatically generating a personalized mediation agreement based on mediation case characteristics are implemented. For example, the computer-readable storage medium may be the memory 802 including the program instructions, and the program instructions may be executed by the processor 801 of the device 800 for automatically generating a personalized mediation agreement based on mediation case characteristics to complete the method for automatically generating a personalized mediation agreement based on mediation case characteristics.

[0136] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for automatically generating a personalized mediation agreement based on mediation case characteristics, characterized in that: include: Obtaining historical mediation case information, including case type, dispute amount, party behavior pattern, case background, mediation needs, and mediation results; Performing structured processing on the historical mediation case information, and converting various types of information in the case into a standardized data format, thereby obtaining a historical mediation case feature data set; Constructing a semantic relationship network between case features according to the historical mediation case feature data set, and processing the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library; Acquire information of cases to be mediated, and perform feature extraction processing on the information of cases to be mediated to obtain input feature data; Performing pattern recognition and fuzzy reasoning processing based on the input feature data, predicting the behavioral responses of the parties in the case through a preset sociological model, and using a fuzzy reasoning algorithm to reason and analyze the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results; Perform matching processing based on the fuzzy reasoning result, and obtain a preliminary mediation agreement by screening from the mediation agreement template library; The preliminary mediation agreement is optimized according to the input feature data, and the mediation agreement text is automatically generated by utilizing the generator part of the generative adversarial network model. The discriminator part evaluates the legality and rationality of the generated agreement clauses to obtain a personalized mediation agreement.

2. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: According to the structured processing of the historical mediation case information, various types of information in the case are converted into a standardized data format to obtain a historical mediation case feature data set, including: Performing standardization processing according to the case types, extracting case types from case descriptions through text analysis and classification algorithms and converting case categories into numerical category data to obtain standardized case type data; Numerical processing is performed based on the dispute amount. The dispute amount information in the case is extracted through regular expressions and converted into numerical variables to obtain standardized dispute amount data; Perform sentiment analysis and classification based on the behavior patterns of the parties involved, use the BERT algorithm to analyze and classify the sentiment tendency of the behavior descriptions of the parties involved in the case, and label them in combination with the mediation results to obtain structured behavior pattern data; Feature extraction is performed based on the case background and mediation needs. By extracting the causes of dispute and appeal points in the case background and using the TF-IDF algorithm to analyze the keywords and sentiment distribution of the case description, structured mediation demand data is obtained; The standardized case type data, the standardized dispute amount data, the structured behavior pattern data and the structured mediation demand data are integrated to obtain a historical mediation case feature data set.

3. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: A semantic relationship network between case features is constructed based on the historical mediation case feature data set, and the semantic relationship network is processed using a machine learning algorithm to obtain a mediation agreement template library, including: Based on the historical mediation case feature data set, feature association is performed, a co-occurrence matrix is ​​constructed, and the case features are vectorized by applying a word embedding algorithm, which converts non-numerical data into numerical vectors to obtain a semantic vector representation of the case features. Performing clustering processing according to the semantic vector representation, and obtaining at least two case feature clusters by aggregating case features with similar semantics; Performing network construction processing based on all the case feature clusters, establishing semantic connection relationships between features through a graph neural network, and obtaining a semantic relationship network, wherein each node in the semantic relationship network represents a case feature, and each edge represents association information between features; Templates are generated based on the semantic relationship network, and the semantic relationship network of case features is processed using a convolutional neural network. The key clauses and template structure in the mediation agreement are automatically identified through a training model, and standardized mediation agreement templates suitable for different case types are generated to obtain a final mediation agreement template library.

4. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: Acquire information of cases to be mediated, and perform feature extraction processing on the information of cases to be mediated to obtain input feature data, including: Based on a preset natural language processing model, text processing and feature extraction are performed on the case information to be mediated to obtain semantic feature data of the case; Performing standardization processing on the quantitative features in the case information to be mediated to obtain numerical feature data; The semantic feature data and the numerical feature data are integrated to obtain integrated feature data, and a principal component analysis algorithm is used to perform dimensionality reduction processing on the integrated feature data to obtain input feature data.

5. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: Pattern recognition and fuzzy reasoning are performed based on the input feature data, and the behavioral responses of the parties in the case are predicted through a preset sociological model. The fuzzy reasoning algorithm is used to reason and analyze the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results, including: Performing pattern recognition processing on the input feature data, using a deep neural network to train the input features after dimensionality reduction, extracting potential patterns in the data to obtain pattern recognition results; Performing behavior prediction processing based on the pattern recognition results, performing time series modeling through long short-term memory networks, analyzing the changes in attitudes and willingness to cooperate of the parties in the case over time, and predicting future behavior trends to obtain behavior prediction results; The behavior prediction results are subjected to fuzzy reasoning based on preset fuzzy rules, and the reasoning results are obtained by combining the dispute focus and the behavior patterns of the parties in the case.

6. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: Matching is performed based on the fuzzy reasoning result, and a preliminary mediation agreement is obtained by screening the mediation agreement template library, including: Performing weighted processing on case features according to the fuzzy reasoning results, by weighting various fuzzy features in the case and combining the reasoning confidence corresponding to the fuzzy features, a weighted case feature vector is obtained; Performing similarity calculation based on the weighted case feature vector, and obtaining a similarity score result by calculating the similarity between the case feature vector and the template feature vector, wherein the similarity score result includes the matching degree between each mediation agreement template and the current case; Sorting the templates in the mediation agreement template library according to the similarity scoring results, and screening out the optimal mediation agreement template based on the sorting results; The agreement is generated according to the optimal mediation agreement template, and the clauses in the template are adjusted based on the specific needs of the case to generate a preliminary mediation agreement.

7. The method for automatically generating a personalized mediation agreement based on mediation case characteristics according to claim 1 is characterized in that: The preliminary mediation agreement is optimized according to the input feature data, and the mediation agreement text is automatically generated by using the generator part of the generative adversarial network model. The discriminator part evaluates the legality and rationality of the generated agreement clauses to obtain a personalized mediation agreement, including: Performing content generation processing according to the input feature data, capturing semantic information of case features based on a preset generative adversarial network, and individually adjusting the terms of the preliminary agreement to generate a draft mediation agreement; An evaluation is performed based on the draft mediation agreement, and the legality and rationality of the generated agreement clauses are judged by using the discriminant network of the generative adversarial network. The agreement clauses are checked based on preset legal rules, industry standards and historical cases to determine whether the agreement meets the legality requirements and whether it is reasonable, thereby obtaining a legality-rationality evaluation result; Feedback adjustment is performed according to the legality-rationality evaluation results, the generation strategy of the generative adversarial network is adjusted through a reinforcement learning algorithm, and the final personalized mediation agreement is output.

8. A system for automatically generating personalized mediation agreements based on mediation case characteristics, characterized in that: include: An acquisition module is used to acquire historical mediation case information, wherein the historical mediation case information includes case type, dispute amount, party behavior pattern, case background, mediation demand and mediation result; A conversion module, used to perform structured processing based on the historical mediation case information, and obtain a historical mediation case feature data set by converting various types of information in the case into a standardized data format; A construction module, used to construct a semantic relationship network between case features based on the historical mediation case feature data set, and process the semantic relationship network using a machine learning algorithm to obtain a mediation agreement template library; An input module, used to obtain information of cases to be mediated, and to perform feature extraction processing on the information of cases to be mediated to obtain input feature data; An analysis module is used to perform pattern recognition and fuzzy reasoning processing based on the input feature data, predict the behavioral responses of the parties in the case through a preset sociological model, and use a fuzzy reasoning algorithm to perform reasoning and analysis on the dispute focus, behavioral patterns and emotional characteristics of the case to obtain fuzzy reasoning results; A matching module, performing matching processing based on the fuzzy reasoning result, and obtaining a preliminary mediation agreement by screening from the mediation agreement template library; The output module is used to optimize the preliminary mediation agreement according to the input feature data, automatically generate the mediation agreement text by utilizing the generator part of the generative adversarial network model, and the discriminator part evaluates the legality and rationality of the generated agreement clauses to obtain a personalized mediation agreement.

9. The system for automatically generating personalized mediation agreements based on mediation case characteristics according to claim 8 is characterized in that: The conversion module comprises: A first processing unit is used to perform standardized processing according to the case type, extract the case type from the case description through text analysis and classification algorithm and convert the case category into numerical category data to obtain standardized case type data; The second processing unit is used to perform numerical processing according to the dispute amount, extract the dispute amount information in the case through regular expressions and convert it into a numerical variable to obtain standardized dispute amount data; The first analysis unit is used to perform sentiment analysis and classification processing based on the behavior patterns of the parties, analyze and classify the sentiment tendency of the descriptions of the behavior of the parties in the case through the BERT algorithm, and perform labeling processing in combination with the mediation results to obtain structured behavior pattern data; The first extraction unit is used to extract features according to the case background and mediation needs, extract the dispute causes and appeal points in the case background, and use the TF-IDF algorithm to analyze the keywords and sentiment distribution of the case description to obtain structured mediation demand data; The first integration unit is used to integrate the standardized case type data, the standardized dispute amount data, the structured behavior pattern data and the structured mediation demand data to obtain a historical mediation case feature data set.

10. The system for automatically generating personalized mediation agreements based on mediation case characteristics according to claim 8, characterized in that: The building blocks include: The first association unit is used to perform feature association based on the historical mediation case feature data set, construct a co-occurrence matrix, and apply a word embedding algorithm to vectorize the case features, thereby converting non-numerical data into numerical vectors to obtain a semantic vector representation of the case features; A first clustering unit, configured to perform clustering processing according to the semantic vector representation, and obtain at least two case feature clusters by aggregating case features with similar semantics; A first construction unit is used to perform network construction processing according to all the case feature clusters, establish semantic connection relationships between features through a graph neural network, and obtain a semantic relationship network, wherein each node in the semantic relationship network represents a case feature, and each edge represents association information between features; The first generation unit is used to generate templates according to the semantic relationship network, use a convolutional neural network to process the semantic relationship network of case features, automatically identify key clauses and template structures in the mediation agreement through a training model, generate standardized mediation agreement templates suitable for different case types, and obtain a final mediation agreement template library.

Images