Legal consulting service management method and system assisted by large language model

By obtaining and converting user data, building seed instruction databases and training auxiliary large language models, the problem of low accuracy of existing legal consulting services is solved, and more efficient legal knowledge matching and user adaptability are achieved.

CN118113853BActive Publication Date: 2024-11-15GUANGDONG ZHENGLU TECH GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410464903.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-17
Publication Date
2024-11-15
Estimated Expiration
2044-04-17

AI Technical Summary

Technical Problem

The existing legal consulting service methods are single, resulting in low accuracy of consulting services and low user adaptability.

Method used

By obtaining user-use Q&A data, user Q&A audio data and legal knowledge data, data conversion, generating structured text data sets, building seed instruction databases, dividing them into basic legal knowledge bases and stylized knowledge bases, and training auxiliary languages Model, match legal knowledge and output knowledge Q&A information.

Benefits of technology

It improves the accuracy and user adaptability of legal consulting services, and provides more accurate and personalized legal consulting services.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118113853B_ABST
    Figure CN118113853B_ABST
Patent Text Reader

Abstract

The present invention discloses a legal consulting service management method and system for assisting a large language model, and relates to the technical field of intelligent language text processing. The method comprises: obtaining legal consulting sample data, including user question and answer data, user question and answer audio data, and legal knowledge data, performing data conversion, and generating a structured text data set; constructing a seed instruction database according to the user question and answer data, and dividing the structured text data set into multiple basic legal knowledge bases and multiple stylized knowledge bases; training and obtaining an auxiliary large language model according to the seed instruction database, multiple basic legal knowledge bases, and multiple stylized knowledge bases, performing legal knowledge matching, and outputting knowledge question and answer information. The present invention solves the technical problem that the legal consulting service method in the prior art is single, resulting in low consulting service accuracy and low user adaptation, and achieves the technical effect of improving consulting service accuracy and user adaptation by optimizing and training an auxiliary large language model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent language text processing, and in particular to a legal consulting service management method and system that assists a large language model. Background Art

[0002] The intelligent response system for legal services can understand the user's query needs through machine learning and natural language processing technology. For example, it uses text mining technology to extract key information from a large amount of legal texts, uses semantic analysis technology to understand the intent of the query, and uses matching algorithms to find the most relevant legal provisions or cases to answer the user's questions. However, the consulting products currently on the market can only give rigid "standard answers" and "general drafts", and cannot provide consulting and legal documents based on the specific circumstances of each company. The consulting solutions are low in accuracy and user adaptability. Summary of the invention

[0003] The present application provides a legal consulting service management method and system assisted by a large language model, which is used to solve the technical problem that the legal consulting service method in the prior art is single, resulting in low consulting service accuracy and low user adaptability.

[0004] In a first aspect of the present application, a legal consulting service management method for an auxiliary large language model is provided, the method comprising: obtaining legal consulting sample data, the legal consulting sample data comprising user usage question and answer data, user question and answer audio data and legal knowledge data; performing data conversion on the user usage question and answer data, user question and answer audio data and legal knowledge data to generate a structured text data set; constructing a seed instruction database according to the user usage question and answer data; based on the seed instruction database, dividing the structured text data set into multiple basic legal knowledge bases and multiple stylized knowledge bases; training an auxiliary large language model according to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases; performing legal knowledge matching according to the auxiliary large language model in combination with real-time user consultation data, and outputting knowledge question and answer information.

[0005] According to a second aspect of the present application, a legal consulting service management system for assisting a large language model is provided, the system comprising: a consulting sample data acquisition module, the consulting sample data acquisition module is used to acquire legal consulting sample data, the legal consulting sample data includes user usage question and answer data, user question and answer audio data and legal knowledge data; a data conversion module, the data conversion module is used to perform data conversion on the user usage question and answer data, the user question and answer audio data and the legal knowledge data to generate a structured text data set; a seed instruction database construction module, the seed instruction database construction module is used to construct a seed instruction database based on the user usage question and answer data; a knowledge base partitioning module, the knowledge base partitioning module is used to partition the structured text data set into multiple basic legal knowledge bases and multiple stylized knowledge bases based on the seed instruction database; an auxiliary large language model training module, the auxiliary large language model training module is used to train an auxiliary large language model based on the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases; a knowledge question and answer information output module, the knowledge question and answer information output module is used to perform legal knowledge matching based on the auxiliary large language model and in combination with real-time user consulting data, and output knowledge question and answer information.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The legal consulting service management method of the auxiliary large language model provided in the present application relates to the technical field of intelligent language text processing. By acquiring user question and answer data, user question and answer audio data and legal knowledge data, data conversion is performed to generate a structured text data set, a seed instruction database is constructed according to the user question and answer data, the structured text data set is divided into multiple basic legal knowledge bases and multiple stylized knowledge bases, and an auxiliary large language model is trained to perform legal knowledge matching and output knowledge question and answer information. The technical problem that the legal consulting service method in the prior art is single, resulting in low consulting service accuracy and low user adaptation is solved, and the technical effect of improving consulting service accuracy and user adaptation by optimizing the training of the auxiliary large language model is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A flowchart of a method for managing a legal consulting service with a large language model provided in an embodiment of the present application;

[0010] Figure 2 A schematic diagram of a process for obtaining multiple basic legal knowledge bases and multiple stylized knowledge bases in the legal consulting service management method for assisting a large language model provided in an embodiment of the present application;

[0011] Figure 3 A schematic diagram of the structure of a legal consulting service management system that assists a large language model provided in an embodiment of the present application.

[0012] Explanation of the accompanying drawings: consultation sample data acquisition module 11, data conversion module 12, seed instruction database construction module 13, knowledge base division module 14, auxiliary large language model training module 15, knowledge question and answer information output module 16. DETAILED DESCRIPTION

[0013] The present application provides a legal consulting service management method assisted by a large language model, which is used to solve the technical problem that the legal consulting service method in the prior art is single, resulting in low consulting service accuracy and low user adaptability.

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices.

[0016] Embodiment 1

[0017] like Figure 1 As shown, the present application provides a legal consulting service management method that assists a large language model, the method comprising:

[0018] P10: Obtaining legal consultation sample data, wherein the legal consultation sample data includes user usage question and answer data, user question and answer audio data, and legal knowledge data;

[0019] Specifically, based on big data, sample data related to legal consultation is obtained, including user Q&A data, user Q&A audio data, and legal knowledge data. Among them, the user Q&A data refers to the legal consultation questions raised by users in text form and the corresponding answers, which may come from online legal consultation platforms, forums, social media, etc., and can reflect the user's real legal consultation needs and scenarios, as well as the common problems and concerns of user consultation. The user Q&A audio data refers to the legal consultation conducted by users in voice form in addition to text data, which can be converted into text form through voice recognition technology, further enriching the types and quantities of sample data, helping the model to better understand and process the user's voice input, and provide more convenient legal consultation services. The legal knowledge data is the core data for building legal consulting services, including professional legal knowledge such as legal provisions, cases, and interpretations, and is an important basis for answering user consultation questions.

[0020] P20: Convert the user Q&A data, user Q&A audio data and legal knowledge data to generate a structured text data set;

[0021] Furthermore, step P20 of the embodiment of the present application further includes:

[0022] P21: Extract key information based on the user Q&A data, user Q&A audio data, and legal knowledge data to obtain key consulting information;

[0023] P22: Construct a text data structure, wherein the text data structure includes a question-answer structure and a knowledge reference structure;

[0024] P23: Based on the key consulting information, the text data structure is filled with data and data category marking is performed to generate the structured text data set, wherein missing or uncertain information is marked as to be supplemented or in an uncertain state.

[0025] Optionally, the user Q&A data, user Q&A audio data and legal knowledge data are cleaned, converted to text and extracted to generate a structured text data set. Specifically, first, through natural language processing technology and text mining technology, key information related to legal consultation, such as question type, legal field, key legal concepts, etc., is extracted from the user Q&A data, user Q&A audio data and legal knowledge data as key consultation information.

[0026] Furthermore, a text data structure is constructed, which includes a question-and-answer structure and a knowledge reference structure, wherein the question-and-answer structure is used to organize users' questions and corresponding answers, and the knowledge reference structure is used to record the sources and basis of legal knowledge cited in the answers. The text data structure helps the model to better understand and reason about legal consulting issues and improve the accuracy and authority of the answers.

[0027] Furthermore, the key consulting information is used to fill the data of the corresponding structural parts of the text data structure, and the data is categorized for subsequent model training and reasoning. For missing or uncertain information, it is marked as to be supplemented or uncertain for further improvement and processing.

[0028] P30: constructing a seed instruction database based on the user's question and answer data;

[0029] Furthermore, step P30 of the embodiment of the present application further includes:

[0030] P31: Classify the user question and answer data, and label them according to the data categories to obtain a basic question and answer data set;

[0031] P32: extracting multiple core seed instructions based on the basic question-answering dataset;

[0032] P33: Associating and indexing the multiple core seed instructions with the answer data, and merging and removing duplicates of the multiple core seed instructions through similarity analysis, and obtaining a seed instruction database through structured storage.

[0033] It should be understood that the user question and answer data is used as the basic data to construct the seed instruction database. Specifically, the user question and answer data is classified, different types of questions and corresponding answer patterns are identified, and labeling is performed according to the data category, and each question and answer is assigned a corresponding category label to obtain a basic question and answer data set.

[0034] Furthermore, based on the basic question-and-answer dataset, multiple core seed instructions are further extracted. The core seed instructions are a simplification and summary of user questions, which can represent the core intention and needs of a type of question. By extracting the core seed instructions, complex and diverse user questions can be summarized into a limited set of instructions, simplifying the processing process of the model.

[0035] Furthermore, the multiple core seed instructions are associated with the corresponding answer data and indexed so that when the model receives similar questions, it can quickly find relevant answers. At the same time, in order to avoid redundancy and duplication of instructions, multiple core seed instructions are merged and deduplicated through similarity analysis to ensure the simplicity and efficiency of the seed instruction database. Finally, the core seed instructions that have been merged and deduplicated are organized into a seed instruction database through structured storage.

[0036] P40: Based on the seed instruction database, the structured text data set is divided into a plurality of basic legal knowledge bases and a plurality of stylized knowledge bases;

[0037] Further, such as Figure 2 As shown, step P40 of the embodiment of the present application also includes:

[0038] P41: traversing the seed instruction database, matching the multiple core seed instructions with the text in the structured text data set, and obtaining an instruction-text association set;

[0039] P42: Conduct content attribute analysis on the structured text data set, and classify the text according to the attribute analysis results to obtain a legal knowledge text set and a stylized knowledge text set;

[0040] P43: Based on the instruction-text association set, the legal knowledge text set and the stylized knowledge text set are matched and grouped to obtain multiple basic legal knowledge bases and multiple stylized knowledge bases corresponding to the multiple core seed instructions.

[0041] Specifically, referring to the seed instruction database, the structured text data set is divided into multiple basic legal knowledge bases and multiple stylized knowledge bases. The basic legal knowledge base contains general and basic legal knowledge, which is the core knowledge source for answering user questions. The stylized knowledge base includes legal knowledge in specific contexts or expressions, which helps the model generate answers that better meet user expectations.

[0042] Furthermore, each core seed instruction in the seed instruction database is traversed, matched with the text in the structured text data set, and the association between the instruction and the text is established to form an instruction-text association set. Furthermore, for the structured text data set, content attribute analysis is performed, including analysis and judgment of the subject, field, style, etc. of the text, and the text is divided into different categories to form a legal knowledge text set and a stylized knowledge text set. The legal knowledge text set mainly includes professional knowledge and interpretation related to the law, while the stylized knowledge text set focuses on legal applications and suggestions under different scenarios and needs.

[0043] Furthermore, based on the instruction-text association set, the legal knowledge text set and the stylized knowledge text set are matched and grouped. Exemplarily, each core seed instruction is matched with the relevant legal knowledge text and stylized knowledge text to form a corresponding basic legal knowledge base and stylized knowledge base, so that each core seed instruction has legal knowledge and stylized suggestions that are closely related to it, providing more accurate and personalized service support for the model.

[0044] P50: According to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases, an auxiliary large language model is trained;

[0045] Furthermore, step P50 of the embodiment of the present application also includes:

[0046] P51: associating and integrating the data in the seed instruction database, the basic legal knowledge base and the stylized knowledge base, and classifying and labeling them to form an associated data set;

[0047] P52: Determine multiple processing tasks based on the seed instruction database;

[0048] P53: Based on the multiple processing tasks, extract model training data according to the associated data set;

[0049] P54: Select a basic large language model, build an overall architecture of an auxiliary large language model, use the model training data for training, and obtain the auxiliary large language model.

[0050] It should be understood that the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases are used as training data to train the auxiliary large language model. Specifically, first, the data in the seed instruction database, the basic legal knowledge base and the stylized knowledge base are associated and integrated, including matching the core seed instructions with the relevant legal knowledge and stylized suggestions to form a complete associated data set. At the same time, the data is classified and labeled so that the model can clearly identify the category and purpose of each data.

[0051] Further, based on the seed instruction database, multiple processing tasks are determined, and the multiple processing tasks represent different types of legal consulting work that the model needs to complete, such as answering legal questions, explaining legal knowledge, providing legal advice, etc. Clarifying the processing tasks helps to extract training data and design the model structure in a targeted manner. Further, according to the multiple processing tasks, corresponding model training data is extracted from the associated data set, including user questions, legal knowledge, and stylized suggestions, etc., for training the performance of the model under different tasks.

[0052] Furthermore, a basic large language model suitable for processing natural language tasks, such as BERT, GPT, T5, etc., is selected, and an overall architecture is constructed according to the needs of the auxiliary large language model. The basic large language model is trained using the extracted model training data. Through continuous iteration and optimization, an auxiliary large language model is finally obtained that can accurately understand and answer legal consulting questions.

[0053] P60: Based on the auxiliary large language model, combined with the user's real-time consultation data, legal knowledge matching is performed and knowledge question and answer information is output.

[0054] Furthermore, step P60 of the embodiment of the present application also includes:

[0055] P61: The auxiliary large language model includes a demand identification module and a knowledge matching module;

[0056] P62: Based on the demand identification module, user demand identification is performed according to the user's real-time consultation data, including voice recognition and text recognition, to obtain demand identification results;

[0057] P63: Based on the demand identification results, legal knowledge matching is performed through the knowledge matching module to output knowledge question and answer information.

[0058] Specifically, after the auxiliary large language model is constructed, the auxiliary large language model is used to process the user's legal consultation data in real time, and accurate and timely legal knowledge matching and question-answering services are obtained by inputting the user's real-time consultation data into the auxiliary large language model.

[0059] It should be understood that the auxiliary large language model includes multiple modules, among which the key ones are the demand identification module and the knowledge matching module. The demand identification module is responsible for parsing the data input by the user and identifying the specific needs of the user for legal consultation, while the knowledge matching module is responsible for searching and matching relevant information in the existing legal knowledge base to provide accurate answers to the user.

[0060] When a user asks for legal advice, the demand recognition module will first process the real-time consultation data input by the user, including voice recognition of the user's voice input and converting it into text, or recognizing the consultation content of the text directly input by the user to obtain the demand recognition result. Through this step, the user's needs and intentions can be accurately understood.

[0061] Furthermore, after clarifying the user's needs, the needs are identified according to the knowledge matching module, matching and retrieval are performed in the basic legal knowledge base and the stylized knowledge base, the user's demand type and specific content are analyzed, and the relevant legal clauses, cases, interpretations, etc. are found. Combined with the information in the stylized knowledge base, personalized and professional legal knowledge question and answer information is obtained to provide users with timely and professional legal knowledge services.

[0062] Furthermore, the embodiment of the present application further includes step P70, and step P70 further includes:

[0063] P71: If the demand identification result is unrecognizable, the user's real-time consultation data is transferred to the manual service window, and the manual question and answer data is recorded;

[0064] P72: Based on the preset cycle, collect the manual question and answer data record set, perform data conversion, and generate new seed instructions and new structured text data sets;

[0065] P73: According to the newly added seed instruction and the newly added structured text data set, update training of the auxiliary large language model is performed to obtain an updated auxiliary large language model.

[0066] Optionally, in the process of real-time processing of user legal consultation data, the auxiliary large language model may encounter some consultation content that cannot be accurately identified. If the demand identification result is unrecognizable, it means that the current auxiliary large language model cannot accurately understand the user's consultation content or cannot find matching legal knowledge, then the user's real-time consultation data needs to be transferred to the manual service window, where professional legal personnel interact with users to provide legal consultation services. At the same time, the data of the entire question-and-answer process, including user questions, professional legal personnel's answers, and related explanations and discussions, are recorded for optimizing the model and expanding the knowledge base to further improve service quality.

[0067] Furthermore, according to the preset data collection cycle, such as one month, three months, etc., the specific time can be adaptively adjusted according to actual needs, and the manual question and answer data record set is collected. According to the above method, data conversion is performed to extract the user's questions and the answers of professional legal personnel respectively, to form new seed instructions and corresponding answer texts, and generate new seed instructions and new structured text data sets.

[0068] Furthermore, the auxiliary large language model is updated and trained using the newly added seed instructions and the newly added structured text data set, so that it can better understand and answer the user's legal consultation questions, and obtain a more complete and accurate updated auxiliary large language model. By updating and training the auxiliary large language model using the newly added data, the performance of the model can be continuously improved, so that it can better adapt to the ever-changing legal consultation needs. At the same time, it helps to maintain the timeliness and accuracy of the model and provide users with a better service experience.

[0069] In summary, the embodiments of the present application have at least the following technical effects:

[0070] This application obtains user question and answer data, user question and answer audio data and legal knowledge data, performs data conversion, generates a structured text data set, builds a seed instruction database based on the user question and answer data, divides the structured text data set into multiple basic legal knowledge bases and multiple stylized knowledge bases, trains to obtain an auxiliary large language model, performs legal knowledge matching, and outputs knowledge question and answer information.

[0071] The technical effect of improving the accuracy of consulting services and user adaptability by optimizing the training of assisted large language models has been achieved.

[0072] Embodiment 2

[0073] Based on the same inventive concept as the legal consulting service management method assisted by a large language model in the aforementioned embodiment, Figure 3 As shown, the present application provides a legal consulting service management system that assists a large language model. The system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0074] A consultation sample data acquisition module 11, wherein the consultation sample data acquisition module 11 is used to acquire legal consultation sample data, wherein the legal consultation sample data includes user question and answer data, user question and answer audio data, and legal knowledge data;

[0075] A data conversion module 12, the data conversion module 12 is used to convert the user usage question and answer data, the user question and answer audio data and the legal knowledge data to generate a structured text data set;

[0076] A seed instruction database construction module 13, wherein the seed instruction database construction module 13 is used to construct a seed instruction database according to the user's question and answer data;

[0077] A knowledge base partitioning module 14, wherein the knowledge base partitioning module 14 is used to partition the structured text data set into a plurality of basic legal knowledge bases and a plurality of stylized knowledge bases based on the seed instruction database;

[0078] An auxiliary large language model training module 15, the auxiliary large language model training module 15 is used to train and obtain an auxiliary large language model according to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases;

[0079] The knowledge question and answer information output module 16 is used to match legal knowledge based on the auxiliary large language model and the user's real-time consultation data, and output knowledge question and answer information.

[0080] Furthermore, the data conversion module 12 is further configured to perform the following steps:

[0081] Extract key information based on the user question and answer data, user question and answer audio data, and legal knowledge data to obtain key consulting information;

[0082] Constructing a text data structure, wherein the text data structure includes a question-answer structure and a knowledge reference structure;

[0083] According to the key consulting information, the text data structure is filled with data and data category marking is performed to generate the structured text data set, wherein missing or uncertain information is marked as to be supplemented or in an uncertain state.

[0084] Furthermore, the seed instruction database construction module 13 is also used to perform the following steps:

[0085] Classifying the user question and answer data and labeling them according to the data categories to obtain a basic question and answer data set;

[0086] Based on the basic question-answering dataset, extracting multiple core seed instructions;

[0087] The multiple core seed instructions are associated with the answer data and indexed, and after the multiple core seed instructions are merged and deduplicated through similarity analysis, a seed instruction database is obtained through structured storage.

[0088] Furthermore, the knowledge base partitioning module 14 is also used to perform the following steps:

[0089] Traversing the seed instruction database, matching the plurality of core seed instructions with texts in the structured text data set, and obtaining an instruction-text association set;

[0090] Conducting content attribute analysis on the structured text data set, and classifying the text according to the attribute analysis results to obtain a legal knowledge text set and a stylized knowledge text set;

[0091] Based on the instruction-text association set, the legal knowledge text set and the stylized knowledge text set are matched and grouped to obtain multiple basic legal knowledge bases and multiple stylized knowledge bases corresponding to the multiple core seed instructions.

[0092] Furthermore, the auxiliary large language model training module 15 is also used to perform the following steps:

[0093] Associating and integrating the data in the seed instruction database, the basic legal knowledge base, and the stylized knowledge base, and classifying and labeling them to form an associated data set;

[0094] Based on the seed instruction database, determining a plurality of processing tasks;

[0095] Based on the plurality of processing tasks, extracting model training data according to the associated data set;

[0096] A basic large language model is selected, an overall architecture of an auxiliary large language model is constructed, and the model training data is used for training to obtain the auxiliary large language model.

[0097] Furthermore, the knowledge question and answer information output module 16 is also used to perform the following steps:

[0098] The auxiliary large language model includes a demand identification module and a knowledge matching module;

[0099] Based on the demand recognition module, user demand recognition is performed according to the user's real-time consultation data, including voice recognition and text recognition, to obtain demand recognition results;

[0100] According to the demand identification results, legal knowledge matching is performed through the knowledge matching module to output knowledge question and answer information.

[0101] Furthermore, the system also includes:

[0102] A manual question and answer data recording module, wherein if the demand identification result is unrecognizable, the user's real-time consultation data is transferred to a manual service window and the manual question and answer data is recorded;

[0103] A new data conversion module is added, which is used to collect manual question and answer data record sets based on a preset period, and perform data conversion to generate new seed instructions and new structured text data sets;

[0104] A model update training module is used to perform update training of the auxiliary large language model according to the newly added seed instructions and the newly added structured text data set to obtain an updated auxiliary large language model.

[0105] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0106] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0107] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A legal consulting service management method assisted by a large language model, characterized in that: The method comprises: Acquire legal consultation sample data, wherein the legal consultation sample data includes user usage question and answer data, user question and answer audio data, and legal knowledge data; Convert the user question and answer data, user question and answer audio data, and legal knowledge data to generate a structured text data set; Building a seed instruction database based on the user's question and answer data; Based on the seed instruction database, dividing the structured text data set into a plurality of basic legal knowledge bases and a plurality of stylized knowledge bases; According to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases, training and obtaining an auxiliary large language model; According to the auxiliary large language model, combined with the user's real-time consultation data, legal knowledge matching is performed to output knowledge question and answer information; According to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases, training and obtaining an auxiliary large language model includes: Associating and integrating the data in the seed instruction database, the basic legal knowledge base, and the stylized knowledge base, and classifying and labeling them to form an associated data set; Based on the seed instruction database, determining a plurality of processing tasks; Based on the plurality of processing tasks, extracting model training data according to the associated data set; Select a basic large language model, build an overall architecture of an auxiliary large language model, and use the model training data for training to obtain the auxiliary large language model; According to the auxiliary large language model, combined with the user's real-time consultation data, legal knowledge matching is performed and knowledge question and answer information is output, including: The auxiliary large language model includes a demand identification module and a knowledge matching module; Based on the demand recognition module, user demand recognition is performed according to the user's real-time consultation data, including voice recognition and text recognition, to obtain demand recognition results; According to the demand identification results, legal knowledge matching is performed through the knowledge matching module to output knowledge question and answer information.

2. The method according to claim 1, characterized in that The user question and answer data, user question and answer audio data and legal knowledge data are converted to generate a structured text data set, including: Extract key information based on the user question and answer data, user question and answer audio data, and legal knowledge data to obtain key consulting information; Constructing a text data structure, wherein the text data structure includes a question-answer structure and a knowledge reference structure; According to the key consulting information, the text data structure is filled with data and data category marking is performed to generate the structured text data set, wherein missing or uncertain information is marked as to be supplemented or in an uncertain state.

3. The method according to claim 1, characterized in that A seed instruction database is constructed according to the user's question and answer data, including: Classifying the user question and answer data and labeling them according to the data categories to obtain a basic question and answer data set; Based on the basic question-answering dataset, extracting multiple core seed instructions; The multiple core seed instructions are associated with the answer data and indexed, and after the multiple core seed instructions are merged and deduplicated through similarity analysis, a seed instruction database is obtained through structured storage.

4. The method according to claim 3, characterized in that Based on the seed instruction database, the structured text data set is divided into a plurality of basic legal knowledge bases and a plurality of stylized knowledge bases, including: Traversing the seed instruction database, matching the plurality of core seed instructions with texts in the structured text data set, and obtaining an instruction-text association set; Conducting content attribute analysis on the structured text data set, and classifying the text according to the attribute analysis results to obtain a legal knowledge text set and a stylized knowledge text set; Based on the instruction-text association set, the legal knowledge text set and the stylized knowledge text set are matched and grouped to obtain multiple basic legal knowledge bases and multiple stylized knowledge bases corresponding to the multiple core seed instructions.

5. The method according to claim 4, characterized in that The method further comprises: If the demand identification result is unrecognizable, the user's real-time consultation data is transferred to the manual service window, and the manual question and answer data is recorded; Based on the preset cycle, collect the manual question and answer data record set, perform data conversion, and generate new seed instructions and new structured text data sets; According to the newly added seed instruction and the newly added structured text data set, the auxiliary large language model is updated and trained to obtain an updated auxiliary large language model.

6. A legal consulting service management system that assists large language models, characterized in that: The system is used to execute the method according to any one of claims 1 to 5, and the system comprises: A consultation sample data acquisition module, which is used to acquire legal consultation sample data, wherein the legal consultation sample data includes user question and answer data, user question and answer audio data, and legal knowledge data; A data conversion module, wherein the data conversion module is used to perform data conversion on the user question and answer data, the user question and answer audio data, and the legal knowledge data to generate a structured text data set; A seed instruction database construction module, wherein the seed instruction database construction module is used to construct a seed instruction database according to the user's question and answer data; A knowledge base partitioning module, the knowledge base partitioning module is used to partition the structured text data set into a plurality of basic legal knowledge bases and a plurality of stylized knowledge bases based on the seed instruction database; an auxiliary large language model training module, the auxiliary large language model training module being used to train and obtain an auxiliary large language model according to the seed instruction database, the multiple basic legal knowledge bases and the multiple stylized knowledge bases; The knowledge question and answer information output module is used to match legal knowledge based on the auxiliary large language model and the user's real-time consultation data, and output the knowledge question and answer information.