Search

CN-122019723-A - AI (automatic identification) mediation method based on data large model

CN122019723ACN 122019723 ACN122019723 ACN 122019723ACN-122019723-A

Abstract

The application discloses an AI (analog interface) mediation method based on a data large model, which adopts the AI mediation method, wherein an interaction system is used for receiving multi-modal information of a principal and feeding back the multi-modal dialogue information output by a mediation engine system to the principal so as to complete multi-modal dialogue interaction between an AI mediator and the principal, the mediation engine system is used for initiating one or more rounds of AI mediation of one or more AI mediators, generating the multi-modal dialogue information by combining a multi-modal mediation template of the mediation large model system and the dialogue context in the interaction system and sending the multi-modal dialogue information to the interaction system, and the mediation large model system is used for generating the multi-modal mediation template according to multi-modal mediation data and multi-dimensional intelligent labeling results of the AI mediator and sending the multi-modal mediation template to the mediation engine system. By the scheme of the application, the adjustment efficiency can be improved, and the adjustment quality can be enhanced.

Inventors

  • LIN PEIXING
  • CHEN YICHENG

Assignees

  • 深圳市法自然信息科技有限公司

Dates

Publication Date
20260512
Application Date
20240913

Claims (10)

  1. 1. An AI mediation method based on a data large model is characterized in that an AI mediation system is adopted, and the AI mediation system comprises an interaction system, a mediation engine system and a mediation large model system: The interaction system comprises an interaction subsystem, a digital person subsystem and a manual mediation subsystem, wherein the interaction subsystem is used for responding to the multi-mode information of a principal so as to drive an AI mediator to conduct multi-mode dialogue interaction operation with the principal; The mediation engine system comprises a mediation start subsystem, a mediation conversation subsystem, an intelligent labeling subsystem and an intelligent monitoring subsystem, wherein the mediation start subsystem is used for mediating a case importing system and selecting a target AI mediator to transmit AI mediation parameters to the mediation conversation subsystem, the mediation conversation subsystem is used for generating multi-mode conversation information according to the AI mediation parameters of the mediation start subsystem, a multi-mode mediation template of a mediation library subsystem in a mediation big model system, multi-dimensional intelligent labeling of the intelligent labeling subsystem and context content of the mediation conversation, and transmitting the multi-mode conversation information to an interaction subsystem in the interaction system, and the intelligent labeling subsystem is used for receiving text elements, audio text elements and video text elements of the interaction system and generating corresponding data labels to respectively transmit the multi-dimensional intelligent labeling results to the mediation conversation subsystem and the mediation big model system; The large-model mediation system comprises a mediation library subsystem, a mediation training subsystem, a mediation modeling subsystem and a third party resource library subsystem, wherein the mediation library subsystem is used for constructing core mediation data according to the output of one or more subsystems, the mediation training subsystem is used for optimizing a multi-mode mediation template by utilizing an AI strategy feature set and a target machine learning model, the mediation modeling subsystem is used for integrating a target algorithm through a multi-core algorithm structure and generating a multi-mode mediation algorithm model for the mediation training subsystem to call, and the third party resource library subsystem is used for acquiring the target database and the target large model; The method comprises the steps that cases to be mediated are imported into an AI mediation system through an API or a file, the AI mediation system conducts case analysis, invokes an AI mediation robot trained by a mediation big model, the AI mediation robot mediates with a principal by using ICT, AR/VR, big data and blockchain technologies according to the content of a mediation application, the AI mediation robot contacts the principal through telephone, short message, video, AR/VR and digital, the AI mediation robot conducts portrait on the principal, analyzes communication intention of the principal, conducts continuous mediation communication with the principal, conducts interactive and overlapped communication negotiation, guides the principal to mediate, achieves a mediation protocol, mediates and mediates legal documents of a judicial confirmation application, fulfills legal obligations agreed in the mediation protocol, monitors operation conditions of the AI mediation staff in real time, discovers business problems in the mediation process in real time, adjusts the mediation strategy and maintains stable operation of the mediation robot.
  2. 2. The AI mediation method of claim 1, wherein the step of executing the interaction subsystem includes: Collecting and extracting the audio information of the principal, converting continuous analog audio signals into discrete digital signals, framing, extracting MFCC characteristics, searching an optimal probability path, identifying phonemes by an acoustic model, identifying words and sentences by a vocabulary model and a language model, identifying and sorting the audio signals of the principal in the process of mediation by a mediator and a mediation assistant into small-section audio and corresponding text element information, and transmitting the small-section audio and the corresponding text element information to an AI intelligent labeling subsystem; The method comprises the steps of collecting and extracting video information of a principal, detecting the position and the outline of the face by using a computer vision technology, identifying key points on the face by using a deep learning model, identifying micro-expressions and actions by using a machine learning model, a convolutional neural network and a cyclic neural network to form video element slices, and transmitting identified emotion forming video text element information to an AI intelligent labeling subsystem by using local video element slices such as micro-action videos of eyes, ears, mouths, noses, eyebrows, hairs, wen, cheeks and facial muscles and whole video element slices such as action videos of heads, shoulders, necks, hands, elbows, arms, trunk, crotch and limbs; Analyzing characters, punctuations and expression symbols sent by the principal in the instant message, identifying network expression, analyzing context, finishing text elements forming the principal, and transmitting the text elements of the principal formed by the identification and finishing to an AI intelligent labeling subsystem; The method comprises the steps of receiving multi-modal dialogue of an AI mediation dialogue subsystem, calling an AI digital person subsystem to complete one-time mediation dialogue with a principal, receiving multi-modal dialogue of the AI mediation dialogue subsystem, when the principal does not wish to mediate or the principal definitely needs manual mediation, ending the AI mediation, transferring an AI mediation record to the AI manual mediation subsystem, transferring the AI mediation record to the manual mediation subsystem, forming multi-modal mediation data of the principal according to original characters, voice and video records of the principal, and identifying the text elements, audio elements and video elements after the arrangement, and transmitting the multi-modal mediation data to an AI mediation library subsystem of an AI mediation large model system for multi-modal storage, wherein the text elements comprise keywords recognized by law and law, income, work, clothing and accommodation, consumption, family and social activities, keywords of active, pessimistic and depressed mental states, about keywords with about willingness, tussimistic and refusal, emotion keywords of sadness, delightness, cruiness, anti-sense, disfiguring, sense, audio words of delightness and corresponding words of voice and profuse.
  3. 3. The AI mediation method of claim 1, wherein the AI digital person subsystem receives multimodal dialogue information of the AI interactive subsystem, and implements the digital moderator's praise propaganda, legal consultation, interactive communication with the party, directs the party to have a mediation intent, and mediates using speech synthesis, speech recognition, semantic understanding, image processing, machine translation, avatar-driven AI core techniques.
  4. 4. The AI mediation method of claim 1, wherein the AI mediation promoter system automatically analyzes cases, automatically and/or manually selects AI mediation robots corresponding to the mediating members with high mediation success rate according to different cases and account information of the parties, and sets AI mediation parameters, such as the number of mediation robots, the starting time, the termination conditions, and the AI mediation promoter system starts AI mediation to transmit the AI mediation parameters to the AI mediation dialogue subsystem.
  5. 5. The AI-mediation method of claim 1, wherein the mediation dialog subsystem generates in real time a dialog content with an AI mediator and a party, receives AI-mediation parameters of an AI-mediation initiating subsystem, invokes a multimodal mediation template of the mediator in an AI-mediation library subsystem, generates a multimodal dialog to communicate with the party, feeds back to the AI-interaction subsystem, receives a multidimensional intelligent annotation of the AI-intelligent annotation subsystem, generates a multimodal dialog to communicate with the party in accordance with the multidimensional intelligent annotation of the AI-intelligent annotation subsystem, and the multimodal mediation template of the mediator of the AI-mediation library subsystem, combines the contextual content of the current mediation dialog, and feeds back to the AI-interaction subsystem.
  6. 6. The AI-mediation method of claim 1, wherein the intelligent labeling subsystem performs multi-modal intelligent labeling on the principal, the AI-intelligent labeling subsystem receives a message text element, an audio text element and a video text element of the AI-interaction subsystem, combines and analyzes the message text element, the audio text element and the video text element of the principal, performs multi-dimensional intelligent labeling on the principal when keywords are matched in the elements, and performs multi-modal storage by marking the multi-dimensional intelligent labeling on the principal with preset data labels, wherein the multi-dimensional intelligent labeling comprises principal original files, audio and video information, analyzing and collating corresponding message/audio/video text elements and corresponding data labels, such as relationships between the elements and the labels, feeding back the multi-dimensional intelligent labeling of the principal to the AI-mediation conversation subsystem, and transmitting the multi-modal intelligent labeling and analysis results of the principal to the AI-mediation library subsystem.
  7. 7. The AI mediation method of claim 1, wherein the intelligent monitoring subsystem mediates multi-modal mediation supervision data of a mediation library subsystem in a mediation large model system and obtains multi-modal mediation dialogue data of a human mediator in real time, identifies whether mediation behavior of the human mediator meets mediation process specifications and prompts the mediator to improve in real time, generates a mediation analysis report of the human mediator, and combines the mediation supervision process data to form multi-modal mediation supervision data for transmission to the mediation library subsystem in the mediation large model system.
  8. 8. The AI tuning method of claim 7, wherein the AI tuning library subsystem checks tuning results during the monitoring tuning process, and takes multi-modal tuning supervision data such as tuning process specification data, tuning compliance data, tuning supervision cases from the AI tuning library subsystem, obtains multi-modal tuning dialogue data of the human tuning agent in real time, collates text, voice and video elements, and tuning of language and expression actions in real time, analyzes whether the human tuning agent follows tuning process specifications indicating the identity of the tuning agent, verifies the identity of the party, clarifies tuning thing, listens to the sound of the party, analyzes whether the human tuning agent is tuning compliance, if the sense word is hit, whether the tuning intention of the tuning agent is maintained, if the sense is in compliance with the tuning intention of the party, if the sense is out of control, analyzes that the tuning behavior of the human tuning agent is out of compliance with tuning process specification, prompts the tuning agent to improve in real time, analyzes the fact that the tuning of the human tuning agent is out of compliance, makes monitoring actions in real time, such as detecting the content of the word library, records and reminds the human tuning agent, detects the condition of the excessive dialogue, analyzes whether the human tuning agent is in compliance with respect to follow the tuning action, if the sense of the tuning agent is in response to sense of the word, checks whether the sense of the tuning agent is in fact that the sense of the tuning agent is in compliance with respect to be in compliance with the sense of the tuning, and the sense of the tuning agent is required to be immediately, and the tuning state is generated, and the tuning system is output to the tuning system is in order to make an improved to the tuning data.
  9. 9. The AI mediation method of claim 1, wherein the mediation library subsystem obtains information of different cases and related laws and regulations, case information from a third party repository, and receives and stores multi-modal mediation data of disputes of different cases from all mediators manually mediated by the mediators from the manual mediation subsystem: the method comprises the steps of receiving and storing multi-mode mediation data of a party acquired, analyzed and arranged in the AI mediation process from an AI interaction subsystem, receiving and storing multi-mode smart labels and analysis results of the party acquired and analyzed by a mediator in one mediation from an AI smart label subsystem, analyzing and arranging multi-mode mediation process data of a dispute in which the mediator has achieved mediation and has not achieved mediation, including mediation dialect with high mediation success rate of different cases, mediation skills, multi-mode mediation data of mediation dialogue, and outputting the multi-mode mediation data to an AI mediation training subsystem, receiving multi-mode mediation templates of the trained mediator from the AI mediation training subsystem, wherein the multi-mode mediation templates comprise mediation experiences, mediation dialect, skills and mediation cases of the mediator in various disputes, successful cases of a third party, litigation cases, thinking logic and logic of the mediation, and mediation intention of the mediator on the current person, and general mediation, expression and action scoring and action data of the mediation, and expression of the mediation training subsystem.
  10. 10. The AI mediation method as in claim 1, wherein the mediation training subsystem invokes the multi-modal mediation algorithm model generated and issued by the AI mediation modeling subsystem in mediating different cases by the mediator in the multi-modal mediation data of the dispute, wherein the AI mediation training subsystem invokes, debugs and runs the AI mediation modeling subsystem multi-modal mediation algorithm model based on the existing multi-modal mediation templates and multi-modal mediation data of the appointed mediator, constructs a complete set of model training tools such as entity recognition model, intention classification model, emotion classification model, ASR related model and TTS related model, constructs AI policy feature set, machine learning model engine and self-iteration, continuously updates the AI policy engine, outputs AI policies, and continuously iteratively optimizes the multi-modal mediation templates of the mediator by using the AI policy engine and model training tool.

Description

AI (automatic identification) mediation method based on data large model The application provides a divisional application which is proposed by an application application of an AI (analog input) regulation system based on a data large model aiming at application number CN202411286603.2 and the application creation name, and the application date of the main application is 2024.09.13. Technical Field The application relates to the field of artificial intelligence, in particular to an AI (automatic identification) mediation method based on a data large model. Background Case mediation plays an important role in modern judicial systems, and aims to solve disputes through non-litigation means, reduce court burden and improve judicial efficiency. With the continuous development and complexity of society, the number of cases is dramatically increased, and the traditional litigation mode is time-consuming, labor-consuming and high in cost. The case mediation provides an efficient and economical alternative scheme, can quickly resolve disputes, reduces the time and economic burden of the parties, and promotes social harmony. The existing case mediation system technology is mainly based on experience and judgment of a manual mediator. Moderators are typically professionals with legal knowledge and regulatory skills who promote agreement between parties by communicating with them, analyzing the case, and presenting solutions. This approach is effective to some extent, but has obvious problems that firstly, the quality and ability of the moderator are uneven, and the quality of the moderator is difficult to ensure. Secondly, manual reconciliation is inefficient, and particularly in cases of large cases, reconcilers are difficult to deal with. Finally, the manual reconciliation process is susceptible to subjective factors, which may lead to an unfair result. Meanwhile, various disputes are large in stock, fast in growth, insufficient in judicial resources, and the problems of 'reminding difficulty, difficulty in collection and difficulty in litigation' are faced, for example, due to high time cost, the participation of a mediation organization and a group of mediating members represented by lawyers is insufficient, the efficiency is low, and the problems of more cases, less people and incapacitation of a basic-level national court are solved. At present, based on artificial intelligence technology of commercial large models such as judicial large models, ancient large models and the like, the method has been well utilized in multiple fields, and has driven the rapid development of industry. The field of mediation is also barren land at present. With the rapid development of the Internet, the social contradiction disputes in the digital era show the characteristic development trend of how fast and wide and difficult. The method has the following problems that the mediation is not standardized and the industry standard is not formed. The number of moderators is limited and experience is limited, resulting in insufficient expertise. The dispute case adjustment base is large, the growth speed is high, and the dispute case adjustment base cannot be effectively solved in time. The most basic intelligent outbound call has failed to meet mediation needs. Therefore, a technical solution is needed to improve the reconciliation efficiency and enhance the reconciliation quality. Disclosure of Invention In order to solve the defects of the prior art, the embodiment of the application provides an AI (automatic identification) mediation method based on a data large model. The application solves the problems of insufficient quantity of current modulators, long culture period and the like. The embodiment of the application provides an AI (advanced technology attachment) mediation method based on a data large model, which adopts an AI mediation system, wherein the AI mediation system comprises an interaction system, a mediation engine system and a mediation large model system: The interaction system comprises an interaction subsystem, a digital person subsystem and a manual mediation subsystem, wherein the interaction subsystem is used for responding to the multi-mode information of a principal so as to drive an AI mediator to conduct multi-mode dialogue interaction operation with the principal; The mediation engine system comprises a mediation start subsystem, a mediation conversation subsystem, an intelligent labeling subsystem and an intelligent monitoring subsystem, wherein the mediation start subsystem is used for mediating a case importing system and selecting a target AI mediator to transmit AI mediation parameters to the mediation conversation subsystem, the mediation conversation subsystem is used for generating multi-mode conversation information according to the AI mediation parameters of the mediation start subsystem, a multi-mode mediation template of a mediation library subsystem in a mediation big model system, multi-dimen