Intelligent information extraction system based on multi-mode adaptive analysis

Abstract

The invention provides an intelligent information extraction system based on multi-modal adaptive analysis. The intelligent information extraction system comprises a multi-source data acquisition module, a legal semantic analysis module and an intelligent decision support module. According to the system, voice data is collected, noise reduction processing is carried out, legal semantic information is extracted, a dynamic legal knowledge graph is constructed, a mediation scheme is generated in combination with case characteristics, risks are evaluated, intelligent and adaptive optimization of a mediation strategy is achieved, and dispute handling efficiency and effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of cross-application of artificial intelligence and legal technology, and specifically to an intelligent information extraction system based on multimodal adaptive analysis. Background Art

[0002] Some existing dispute mediation platforms combine voice capture with text recording to preserve the parties' statements as searchable electronic archives. These platforms typically capture recordings via telephone or internet calls and subsequently rely on manual or general speech recognition models for text transcription. Given the potential for noise interference in live environments or call networks, some systems may still have room for improvement in speech clarity and subsequent recognition accuracy.

[0003] In the process of processing text and voice information, some technical solutions have attempted to use natural language processing methods based on rules or shallow semantics to extract key points and controversial issues from the content of the conversation. However, faced with highly professional legal terminology or emotional expressions of the parties, such solutions often find it difficult to fully identify and summarize the key information of the case. In addition, some systems will perform analysis based on the accumulated case library or general knowledge graph, but if the legal relationship involved in the case changes, or the relevant laws and regulations are updated frequently, the existing structured knowledge base cannot respond to this dynamic in a timely manner, resulting in the need for manual correction in the subsequent processing process.

[0004] When generating mediation plans, many dispute resolution systems rely on established rules or limited case templates to provide parties with feasible recommendations. Some plans offer a rough assessment of potential risks during this process, but these often rely on subjective judgment or a single metric, failing to account for the multidimensional factors inherent in different dispute types. Especially when cases are complex, with diverse relationships between parties or types of claims, generating mediation strategies based on a single set of parameters can be difficult to adapt to the actual situation, potentially impacting the intended effectiveness of the plan's implementation.

[0005] When it comes to feedback and adjustments during mediation execution, most current practices rely on manual backtracking to optimize parameters or refine processing strategies. Without sufficient data collection on the subsequent execution outcomes and influencing factors across various cases, the mediation system's ability to achieve sustainable optimization is relatively limited. This, to a certain extent, limits the existing system's adaptive capabilities, making it difficult to continuously improve its mediation efficiency and relevance based on local conditions. Summary of the Invention

[0006] In order to solve the above problems in the prior art, the present invention proposes an intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis, comprising:

[0007] Multi-source data collection module, used to collect dispute mediation voice data;

[0008] Legal semantic parsing module, used to extract legal information from mediation texts;

[0009] An intelligent decision support module is used to generate a mediation plan and perform risk assessment based on the legal information, and the intelligent decision support module includes:

[0010] A mediation solution generation engine for generating a set of candidate solutions containing mediation suggestions;

[0011] A risk identification unit to assess potential legal and executive risks of mediation proposals;

[0012] The dispute type adaptive unit is used to dynamically adjust the mediation solution generation parameters according to the case characteristics.

[0013] The multi-source data acquisition module includes:

[0014] A cloud call system is used to collect and mediate voice data through PSTN or VoIP channels and upload the voice data to a server;

[0015] An adaptive noise reduction unit, configured to pre-process the speech data and retain the original audio and the noise-reduced audio using a noise suppression algorithm based on a deep neural network;

[0016] The dialect recognition unit is used to label the dialect type of speech data.

[0017] The legal semantic analysis module includes:

[0018] a legal entity identification unit, configured to receive the mediation text transcribed by the speech recognition engine and identify legal terms, party identification information, and dispute elements in the text;

[0019] A dynamic legal knowledge graph construction unit is used to construct a knowledge graph structure based on the legal entity identification results. The knowledge graph structure includes nodes and edges. The nodes represent legal subjects, dispute objects and legal clauses, and the edges represent the legal application relationship or dispute affiliation between the nodes.

[0020] The mediation solution generation engine further includes:

[0021] Call the interface of the legal clause node and similar case node that matches the dispute elements identified in the current case;

[0022] A rules engine, used to generate mediation recommendations based on pre-set rules;

[0023] The case-based reasoning model is used to identify reference cases and generate mediation paths based on the similarity between the semantic features of the current case and historical cases.

[0024] The risk identification unit includes a risk scoring module for evaluating candidate mediation solutions in terms of difficulty of execution, legal compliance, and historical execution feedback, wherein the evaluation is based on a risk scoring matrix constructed based on a risk characteristic indicator system;

[0025] A risk ranking module is used to calculate the risk scoring matrix according to a set weighted scoring model to output risk ranking results of candidate mediation solutions.

[0026] The dispute type adaptive unit includes:

[0027] Case feature extraction module, used to extract case features including case type, relationship between parties, and strength of claim;

[0028] A parameter configuration module, configured to match the corresponding dispute type in a preset dispute case classification library based on the extracted case features, and adjust the parameter weight configuration in the mediation solution generation engine;

[0029] The parameter updating module is used to adaptively fine-tune the mediation parameters according to the mediation execution results and mediation effect feedback.

[0030] Feedback indicators include agreement signing rate, contract fulfillment rate, mediation cycle, and complaint rate. These indicators are normalized to form a mediation performance vector.

[0031]

[0032] in:

[0033] represents the mediation performance vector formed by the j-th mediation strategy in the t-th feedback cycle;

[0034] is the score of mediation strategy j under the i-th specific evaluation indicator dimension in the t-th feedback cycle; subscript j represents the mediation strategy number, which is used to distinguish different mediation proposals generated within the system; superscript (t) indicates that the current feedback cycle is the t-th feedback cycle, corresponding to the result record after a complete mediation execution; subscript i represents the i-th dimension indicator of mediation effect evaluation, totaling m.

[0035] The parameter update module calculates the update amount of the i-th parameter based on the following formula:

[0036]

[0037] Among them, η i is the learning rate, and are the historical mean and standard deviation of the i-th indicator in period t, ∈ is a constant, is the indicator confidence score, γ i is the confidence suppression coefficient.

[0038] The system forms a mediation parameter increment vector based on the update amount of each dimension

[0039]

[0040] And the parameter update is completed based on the following formula:

[0041]

[0042] in:

[0043] represents the current parameter vector of the j-th mediation strategy at the beginning of the t-th feedback cycle;

[0044] Represents the new parameter vector obtained after the feedback adjustment in this round;

[0045] The system adjusts the mediation strategy generation weights for subsequent similar cases based on the updated parameter vector.

[0046] Beneficial effects:

[0047] This invention significantly improves information extraction efficiency and strategy matching accuracy in dispute mediation by integrating multimodal voice acquisition, legal semantic analysis, and intelligent mediation decision-making. The system can adapt to different dispute types, dynamically optimize mediation parameter configuration, and intelligently generate and assess mediation plans. This reduces mediation costs, improves mediation success rates, and increases party satisfaction, demonstrating its practicality and potential for widespread adoption. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application, but do not constitute an improper limitation of the present invention. In the drawings:

[0049] Figure 1 A structural diagram of an intelligent extraction system for key information in dispute mediation based on multimodal adaptive analysis is shown. DETAILED DESCRIPTION

[0050] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0051] Example 1 Intelligent Extraction System for Key Information in Dispute Mediation

[0052] like Figure 1 As shown, this embodiment provides a system for intelligently extracting key information for dispute mediation based on multimodal adaptive analysis. This system includes a multi-source data acquisition module, a multimodal semantic parsing module, and an intelligent decision support module. These three modules work together to comprehensively extract, intelligently process, and generate strategies for key legal information in dispute mediation.

[0053] The multi-source data acquisition module collects voice data from the mediation process, ensuring the authenticity and integrity of the data source for subsequent analysis. This module includes a cloud call system and an adaptive noise reduction unit. The cloud call system records voice calls during the mediation process in real time via PSTN or VoIP lines. Audio data is collected simultaneously through local caching and cloud transmission. The adaptive noise reduction unit processes the collected voice signals, employing an ambient noise suppression algorithm based on a deep neural network to enhance speech clarity and ensure the accuracy of subsequent speech recognition and semantic extraction.

[0054] The Legal Semantic Parsing module transcribes the de-noised speech signal into text and performs legal semantic analysis on the text. This module comprises a Legal Entity Recognition unit and a Dynamic Legal Knowledge Graph Construction unit. The Legal Entity Recognition unit uses a BiLSTM-CRF model to annotate and classify legal terms, party identities, and dispute elements within the text. The Dynamic Legal Knowledge Graph Construction unit models the legal clauses, factual relationships, and evidence chains involved in the dispute based on the identified entities and relationships, forming a queryable and scalable legal knowledge structure to support the automated reasoning and generation of subsequent mediation solutions.

[0055] The intelligent decision support module generates a variety of possible mediation options based on semantic analysis results and evaluates the applicability and risks of each option. This module comprises a mediation solution generation engine, a risk identification unit, and a dispute type adaptation unit. Supported by a dynamic knowledge graph, the mediation solution generation engine combines case-based reasoning, a rule engine, and other mechanisms to output a set of mediation recommendations. The risk identification unit quantitatively analyzes the potential risks of each option based on factors such as enforcement difficulty, legal compliance, and historical data feedback. The dispute type adaptation unit dynamically adjusts the parameters for generating mediation solutions based on the characteristic information of specific cases. The case feature extraction module extracts multidimensional feature vectors, including case type, party relationship structure, and claim intensity level. The parameter configuration module matches the corresponding dispute type within a pre-set dispute case classification library based on these features and adjusts the parameter weights within the generation engine. The parameter update module continuously fine-tunes the parameters based on historical mediation results and enforcement feedback to enhance the system's adaptability to complex and diverse dispute scenarios.

[0056] Through the integration and coordination of the above modules, the present invention realizes a closed-loop processing mechanism for voice data collection, legal information extraction, dispute structure modeling and mediation strategy generation throughout the entire dispute mediation process, which helps to improve mediation efficiency and the professionalism and enforceability of processing results.

[0057] Example 2 Multi-source data acquisition module

[0058] The multi-source data acquisition module includes a cloud call system and an adaptive noise reduction unit. The cloud call system, deployed within the mediation work platform, supports dual-channel access via the Public Switched Telephone Network (PSTN) and Voice over Internet Protocol (VoIP). By connecting a call box to a terminal device, the system enables real-time recording and collection of mediation call content. The call box uses hardware interception to capture the audio signal, then uses C language to invoke computer permissions on the local device to obtain the recording stream, which is then uploaded to the system server via the HTTP protocol. Local caching and cloud uploads operate simultaneously to ensure data continuity and integrity despite network fluctuations.

[0059] After the recording data collection is completed, the adaptive noise reduction unit starts the audio preprocessing process. The present invention preferably adopts an environmental noise suppression algorithm based on a deep neural network (DNN), such as the Wave-U-Net architecture, to cope with possible background interference (such as environmental noise, overlapping human voices, etc.) at the mediation site. Under the condition of a 48kHz sampling rate, the algorithm can stably maintain a signal-to-noise ratio (SNR) of not less than 35dB, ensuring the clarity and stability of the voice signal. During the noise suppression process, the system retains both the original audio and the noise-reduced data to support the subsequent semantic processing module to compare and analyze specific modal data.

[0060] To accommodate the diverse dialects found in mediation scenarios across different regions, this module includes a pre-defined dialect feature annotation interface. During the acquisition process, the system uses a dialect classification model to identify the language of audio clips and performs feature modeling and annotation for specific regions (such as Wenzhou dialect and Cantonese). This helps the subsequent ASR engine improve recognition accuracy under dialect conditions. Verification results for this model demonstrate a dialect recognition accuracy rate of 98.7%, significantly reducing semantic recognition bias caused by language differences.

[0061] Through these technical means, the multi-source data acquisition module not only enables high-fidelity collection and processing of dispute mediation voice data, but also provides subsequent modules with a complete, high-quality, and semantically modelable source of original information, thereby enhancing the system's overall mediation support capabilities and information extraction accuracy. This module also supports redundant data storage, timestamp synchronization, and concurrent multi-channel recording, demonstrating excellent engineering feasibility and deployment adaptability.

[0062] Example 3 Legal Semantic Analysis Module

[0063] The function of the legal semantic analysis module is to transcribe the mediation voice data after noise reduction processing into text, and on this basis complete the structured analysis and knowledge modeling of legal semantics, providing semantic support for the subsequent generation of mediation plans and risk assessment.

[0064] The legal semantic parsing module in this embodiment consists of a legal entity recognition unit and a dynamic legal knowledge graph construction unit. The two work together to achieve the conversion from speech text to legal semantic structure. First, the speech-to-text process can be completed by the automatic speech recognition (ASR) engine in the system, and its output is input into the entity recognition unit model for further processing. The legal entity recognition unit uses a BiLSTM-CRF (bidirectional long short-term memory-conditional random field) model to sequence the input text to extract key legal terms, party identity information, dispute elements, and other content from the text.

[0065] The BiLSTM layer captures the contextual semantics of the text, enhancing the ability to identify keywords within the context through bidirectional semantic perception. The CRF layer searches for the globally optimal annotation path on the BiLSTM output, thereby improving the accuracy of overall named entity recognition. This model, trained on the dispute text corpus described in this example, achieved an F1 score of 0.92, enabling highly accurate legal information extraction in most civil mediation cases.

[0066] After entity recognition is completed, the system will send the extracted entity items to the dynamic legal knowledge graph construction unit for modeling processing. This unit converts the semantic connection between entities into a graph structure representation based on the preset legal relationship template and graph node type. The nodes in the graph include legal entities (such as plaintiffs, defendants, guarantors), dispute objects (such as loans, real estate), legal clauses (such as Article 680 of the Civil Code), etc. The edges represent the relationship types between nodes, such as "applicable clauses", "subjective disputes", "attached to evidence", etc. The system adopts an incremental update strategy, which can supplement the graph structure in real time according to new input cases to maintain its timeliness and scalability.

[0067] In practice, the system can load different subgraph templates based on case type to adapt to the legal logic structure of different fields, such as marriage and family, private lending, and property services. For example, in private lending cases, graph construction prioritizes identifying core elements such as "loan amount," "repayment period," "interest rate," and "guarantee liability," ensuring that the constructed graph is highly targeted and factually comprehensive. Once the graph is generated, the system can perform further rule-driven or neural network-driven reasoning to assist in identifying the focus of the dispute, building a chain of evidence, or determining applicable legal provisions.

[0068] In summary, this legal semantic parsing module realizes the structured extraction and semantic reconstruction of legal elements in mediation voice text by combining a high-precision legal entity recognition model with a flexible and scalable knowledge graph modeling mechanism. It can effectively support downstream modules in understanding legal disputes and generating mediation strategies, and significantly improve the legal semantic processing capabilities and intelligence level of the mediation system.

[0069] Example 4 Intelligent Decision Support Module

[0070] In this embodiment, the intelligent decision support module generates multiple candidate mediation solutions based on the output of the legal semantic analysis module. It then performs quantitative or semi-quantitative evaluations of each solution in terms of legal applicability, enforceability, and deviation from historical similar cases, thereby providing intelligent solution recommendations to mediators or support systems. This module primarily comprises a mediation solution generation engine, a risk identification unit, and a dispute type adaptation unit. These components form an organic linkage mechanism to jointly achieve the generation and dynamic optimization of mediation strategies.

[0071] With the support of the dynamic legal knowledge graph, the mediation solution generation engine first calls the legal clause nodes and typical case nodes that match the dispute elements identified in the current case to achieve semantic connection of the legal application relationship. Under the baseline legal framework, the engine can combine the rule engine and the case-based reasoning (CBR) model to generate multiple structured mediation suggestion sets. The rule engine part includes but is not limited to mediation priority judgment rules, amount and interest rate range recommendation rules, installment payment condition combination rules, etc.; the case reasoning part calculates the similarity between the semantic features of the current case and existing cases (such as vector space similarity or structured distance based on graph structure), identifies reference cases with deviations less than a preset threshold (such as 15%), and generates a transferable mediation path.

[0072] The risk identification unit is used to conduct a multi-dimensional risk analysis of the generated candidate mediation solutions. The evaluation content includes the difficulty of the solution's implementation, legal compliance, and success rate indicators in previous implementation feedback. This unit relies on the risk characteristic indicator system to establish a risk scoring matrix. The dimensions may include but are not limited to: whether it involves complex property division, whether there is reliance on third-party performance, the agreement fulfillment rate of historical similar solutions, and whether it violates the provisions of regulations such as the "Civil Procedure Law" and the "Mediation Law". In specific implementation, each solution can be assigned a set of quantitative risk scores, and the final risk ranking is output according to the set weighted scoring model.

[0073] The dispute type adaptive unit realizes context adaptation and continuous optimization in the mediation strategy generation process through parameter configuration and parameter update mechanism, enhancing the system's adaptability and intelligence level to complex and changeable dispute scenarios.

[0074] The dispute type adaptive unit includes a case feature extraction module, a parameter configuration module, and a parameter update module. The case feature extraction module extracts multidimensional case features including case type, party relationship structure, and appeal intensity level by calling the output results of the semantic analysis model. Case type can be represented as a discrete category variable, such as "private lending", "marriage and family", "property services", etc.; the party relationship structure is constructed by the equality of rights and obligations and the degree of economic dependence to construct a structural vector; the appeal intensity is normalized by factors such as semantic emotion recognition value and emotional expression word frequency to obtain an emotional index. The above variables are combined into a case feature vector F:

[0075] F={f1,f2,...f i ,f n}

[0076] Where: f n It represents the nth feature component in the case feature vector and is the last dimension (or the nth dimension) in the entire case feature set.

[0077] In the case feature extraction module, the system extracts several key features that affect the mediation strategy based on the semantic analysis of mediation speech. These features are sequentially combined into a vector F. Each f i is an independent characteristic variable, and f n is the last feature in this vector, which may be, for example, “whether the dispute involves cross-regional parties” or “the number of historical mediation failures”.

[0078] n: represents the total number of case feature dimensions, which is pre-defined by the system, for example, 10 dimensions, 12 dimensions, 20 dimensions, etc., depending on the complexity of the model design.

[0079] The parameter configuration module matches the case feature vector with the existing types in the system's preset dispute case classification library, selects the closest dispute type label, and loads the mediation parameter template corresponding to the type. This template is used to initialize the parameter vector θ in the mediation engine. j :

[0080] θ j =[w1,w2,...w i ,w m ]

[0081] Where: w m represents the mth parameter in the mediation strategy parameter vector and is the last weight item in the mediation parameter set. i They all control the priority of a certain optimization goal in the mediation engine, such as the relative attention paid to "shortest mediation time", "strongest agreement execution" or "highest party satisfaction". mIt is the last item and may represent optimization dimensions that will be added later, such as "mediation cost control weight" or "negotiation round limit".

[0082] m: represents the total number of dimensions of the mediation parameter vector, that is, the number of optimization targets, which is usually set when modeling the mediation strategy.

[0083] To enable dynamic adjustment and self-adaptation of the parameter configuration, the present invention further implements feedback-driven parameter optimization through a parameter update module. After each round of mediation, the system automatically records feedback indicators, including the completion rate, fulfillment rate, mediation cycle (time cost), complaint rate, etc. To support the feedback-driven mechanism of the parameter update module, after each round of mediation, the system collects and organizes the actual execution status to form a mediation performance vector, which is represented as follows:

[0084]

[0085] in:

[0086] represents the mediation performance vector formed by the j-th mediation strategy in the t-th feedback cycle, which is used to quantify the actual performance of the strategy in multiple evaluation dimensions;

[0087] is the score of mediation strategy j under the i-th specific evaluation indicator dimension in the t-th feedback cycle. This score can be normalized based on the original mediation data, and its value is usually normalized to the interval

[0088] [0,1], reflecting the relative advantages and disadvantages under this indicator;

[0089] Subscript j: represents the mediation strategy number, which is used to distinguish different mediation suggestions generated within the system;

[0090] Superscript (t): indicates that the current state is the tth feedback cycle, which corresponds to the result record after a complete mediation execution;

[0091] Subscript i: represents the i-th dimension indicator of mediation effect evaluation, totaling m, such as:

[0092] It can indicate the completion rate, that is, whether the proposed mediation plan is finally accepted and signed by both parties;

[0093] It can represent the fulfillment rate, which refers to whether the parties fulfill their obligations under the agreement after mediation;

[0094] It can represent the mediation cycle (Time Cost), which is the reverse indicator of the time taken from the system generating a solution to reaching an agreement (the shorter the cycle, the higher the score);

[0095] It can represent the complaint ratio, which is the degree of objection or dissatisfaction of the parties to the mediation results. It is usually a negative indicator and needs to be reversed during normalization.

[0096] Other dimensions (such as etc.) can be expanded to system satisfaction, mediator workload, post-intervention frequency, etc.

[0097] The vector It is the basis for achieving multi-dimensional evaluation and is used to compare with historical mean Standard deviation Combined with other statistics, the parameters w of the mediation strategy i Through this structure, the system can track the actual effect of the adjustment strategy dimension by dimension and drive the precise operation of the adaptive parameter fine-tuning mechanism.

[0098] To further implement dimension-by-dimension feedback learning and fine-tuning, the following adjustment parameter vector update formula is adopted:

[0099]

[0100] in:

[0101] represents the update amount of the i-th mediation parameter;

[0102] Score the i-th indicator in the current cycle;

[0103] are the historical mean and standard deviation of the dimension indicator respectively;

[0104] Score the system's confidence level for the indicator;

[0105] η i is the learning rate of this dimension, controlling the adjustment amplitude;

[0106] γ i is the confidence suppression coefficient;

[0107] ∈ is a small constant to prevent division by zero.

[0108] The system calculates the updated values of all parameter dimensions in sequence according to the above formula to obtain the mediation parameter increment vector:

[0109]

[0110] And accordingly complete the mediation parameter update:

[0111]

[0112] in:

[0113] represents the current parameter vector of the j-th mediation strategy at the beginning of the t-th feedback cycle, whose initial value comes from the mediation parameter template matching the dispute type;

[0114] The new parameter vector obtained after feedback adjustment in this round will be used to generate mediation solutions for similar cases in the future.

[0115] For example, if the system detects that in cases of "high intensity of claims + unstable family structure", the combination of "one-time performance + third-party guarantee" performs best in terms of fulfillment rate and satisfaction, then the relevant parameters in the mediation engine will automatically obtain higher weights in subsequent similar cases, thereby realizing the evolution and optimization of strategy priority.

[0116] In summary, this embodiment establishes a complete optimization loop for dispute mediation strategy generation through case-feature-driven initial parameter configuration and a dynamic parameter update mechanism using quantized feedback. This system exhibits significant intelligent adaptive capabilities and practical application value. This system can continuously improve mediation efficiency and increase enforcement success rates across a wide range of dispute scenarios, demonstrating strong engineering feasibility and innovation.

[0117] Through the above methods, the intelligent decision support module can not only automatically generate mediation suggestions based on fixed rules and knowledge graphs, but also realize adaptive adjustment of parameters through feature-driven in diverse case situations, thereby improving the pertinence, executability and overall mediation efficiency of the mediation plan, and meeting the actual needs for the generation of intelligent and personalized mediation strategies in real mediation scenarios.

[0118] The above description is only a preferred embodiment of the present invention. Therefore, any equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the scope of the patent application of the present invention.

Claims

1. An intelligent information extraction system based on multimodal adaptive analysis, characterized by: include: Multi-source data collection module, used to collect dispute mediation voice data; Legal semantic parsing module, used to extract legal information from mediation texts; An intelligent decision support module is used to generate a mediation plan and perform risk assessment based on the legal information, and the intelligent decision support module includes: A mediation solution generation engine for generating a set of candidate solutions containing mediation suggestions; A risk identification unit to assess potential legal and executive risks of mediation proposals; The dispute type adaptive unit is used to dynamically adjust the mediation solution generation parameters according to the case characteristics.

2. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 1, characterized in that: The multi-source data acquisition module includes: A cloud call system is used to collect and mediate voice data through PSTN or VoIP channels and upload the voice data to a server; An adaptive noise reduction unit, configured to pre-process the speech data and retain the original audio and the noise-reduced audio using a noise suppression algorithm based on a deep neural network; The dialect recognition unit is used to label the dialect type of speech data.

3. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 1, characterized in that: The legal semantic analysis module includes: a legal entity identification unit, configured to receive the mediation text transcribed by the speech recognition engine and identify legal terms, party identification information, and dispute elements in the text; A dynamic legal knowledge graph construction unit is used to construct a knowledge graph structure based on the legal entity identification results. The knowledge graph structure includes nodes and edges. The nodes represent legal subjects, dispute objects and legal clauses, and the edges represent the legal application relationship or dispute affiliation between the nodes.

4. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 1, characterized in that: The mediation solution generation engine further includes: Call the interface of the legal clause node and similar case node that matches the dispute elements identified in the current case; A rules engine, used to generate mediation recommendations based on pre-set rules; The case-based reasoning model is used to identify reference cases and generate mediation paths based on the similarity between the semantic features of the current case and historical cases.

5. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 1, characterized in that: The risk identification unit includes a risk scoring module for evaluating candidate mediation solutions in terms of difficulty of execution, legal compliance, and historical execution feedback, wherein the evaluation is based on a risk scoring matrix constructed based on a risk characteristic indicator system; A risk ranking module is used to calculate the risk scoring matrix according to a set weighted scoring model to output risk ranking results of candidate mediation solutions.

6. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 1, characterized in that: The dispute type adaptive unit includes: Case feature extraction module, used to extract case features including case type, relationship between parties, and strength of claim; A parameter configuration module, configured to match the corresponding dispute type in a preset dispute case classification library based on the extracted case features, and adjust the parameter weight configuration in the mediation solution generation engine; The parameter updating module is used to adaptively fine-tune the mediation parameters according to the mediation execution results and mediation effect feedback.

7. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 6, characterized in that: Feedback indicators include agreement signing rate, contract fulfillment rate, mediation cycle, and complaint rate. These indicators are normalized to form a mediation performance vector. in: represents the mediation performance vector formed by the j-th mediation strategy in the t-th feedback cycle; is the score of mediation strategy j under the i-th specific evaluation indicator dimension in the t-th feedback cycle; subscript j represents the mediation strategy number, which is used to distinguish different mediation proposals generated within the system; superscript (t) indicates that the current feedback cycle is the t-th feedback cycle, corresponding to the result record after a complete mediation execution; subscript i represents the i-th dimension indicator of mediation effect evaluation, totaling m.

8. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 7, characterized in that: The parameter update module calculates the update amount of the i-th parameter based on the following formula: Among them, η i is the learning rate, and are the historical mean and standard deviation of the i-th indicator in period t, ∈ is a constant, is the indicator confidence score, γ i is the confidence suppression coefficient.

9. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 8, characterized in that: The system forms a mediation parameter increment vector based on the update amount of each dimension And the parameter update is completed based on the following formula: in: represents the current parameter vector of the j-th mediation strategy at the beginning of the t-th feedback cycle; Represents the new parameter vector obtained after the feedback adjustment in this round.

10. The intelligent extraction system for key dispute mediation information based on multimodal adaptive analysis according to claim 9, characterized in that: The system adjusts the mediation strategy generation weights for subsequent similar cases based on the updated parameter vector.

Images