Large model-based simulation dialogue partner training system and method

Abstract

The invention relates to a simulation dialogue partner training system based on a large model. The simulation dialogue partner training system comprises a product knowledge base module, a question and answer library construction and management module, a pair training simulation module, a stage control module, a supernatural voice dialogue module, a verbal skill prompt module, a real-time evaluation module and a dialogue summary report module. The intelligent partner training system for customer communication training has high simulation dialogue ability, flexible configuration ability and a perfect evaluation feedback mechanism, can effectively replace a traditional scripted training or real person simulation partner training mode, and improves the communication actual combat ability of front-line personnel such as sales personnel and customer service personnel. According to the invention, low-efficiency links in a traditional training mode are replaced by an intelligent means, the training quality, the standardization degree and the reproducibility are improved, an enterprise is finally assisted to create a more professional and efficient customer front-line team, and the system can be widely applied to scenes such as enterprise training, new employee on-duty training and product popularization preparation.

Description

Technical Field

[0001] The present invention belongs to the technical field of training systems, and in particular relates to a large model-based simulated dialogue training system and method. Background Art

[0002] In the financial services industry, frontline sales staff (such as account managers, sales consultants, and customer service specialists) frequently communicate with customers in their daily work, explaining products, handling objections, and closing deals. These interactions require employees to demonstrate strong communication skills, understand products, grasp customer psychology, and adaptability. Consequently, companies invest significant resources in simulation exercises and sales pitch training during talent development and business training. To improve training efficiency, the following customer simulation methods are currently widely used: Scripted Q&A training: Trainees answer standard questions based on textbooks or question banks. The content is fixed, the process is linear, and it lacks interactivity, making it difficult to form practical memories. Live-action role-playing simulations: Experienced staff or instructors simulate customer conversations with trainees. While flexible, they rely heavily on manpower, limit training scale, and have a low degree of standardization. Scenario-based video / recorded courses: These impart knowledge through contextualized content, but lack real-time interaction and feedback, leading to passive learning and limited training effectiveness.

[0003] While existing technologies and practical solutions for customer simulation training have improved training effectiveness to a certain extent, they still suffer from the following key flaws, which severely restrict their applicability in large-scale, high-quality training scenarios: Rigid interactions and an inability to adapt to complex conversations: Systems based on fixed scripts or pre-set question banks are unable to generate customer questions in real time, making it difficult to dynamically adjust the "customer's" behavior and expression based on the trainer's responses. This makes the conversations lack authentic interactivity and fails to simulate the natural customer behavior during real-world communication, such as emotional shifts, thought processes, and repeated questions.

[0004] Single role and lack of personalized customer simulation capabilities: Most existing systems can only simulate basic consultations from "ordinary customers". It is difficult to distinguish and build multiple typical customer profiles (such as price-sensitive, professional, skeptical, and aggressive types), and cannot meet the needs of coping strategy training for different customer types.

[0005] Lack of standardized evaluation mechanism makes training results unquantifiable: Traditional real-life simulations rely heavily on the instructor's subjective judgment, making it difficult to objectively record and evaluate trainees' performance in listening and understanding, speech application, rhythm control, and emotional guidance. This results in the inability to standardize, archive, reuse, or improve training results.

[0006] High cost, low efficiency, and difficulty in updating and maintaining: The real-person training model is costly, has a long cycle, and is difficult to organize, making it difficult to adapt to the needs of high-frequency training or large-scale deployment. At the same time, after changes in the content of the speech and product scenarios, the old training resources need to be rebuilt in large quantities, which has high maintenance costs.

[0007] Therefore, companies urgently need a flexible, scalable, and clear-feedback customer simulation training system to reproduce customer conversation scenarios at low cost and high quality, thereby improving employees' product understanding, adaptability, and customer communication skills. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to design a simulated dialogue training system and method based on a large model, replace the inefficient links in the traditional training method with intelligent means, improve the training quality, standardization and replicability, and ultimately help enterprises build a more professional and efficient customer front-line team, thereby solving existing technical problems.

[0009] To solve the above technical problems, the present invention provides a large-scale model-based realistic dialogue training system, which specifically includes the following modules: The product knowledge base module is used to store and parse structured or unstructured product documents, business manuals and FAQs to form searchable product knowledge data.

[0010] The question and answer library construction and management module is linked with the product knowledge base module to construct and manage a set of standard question and answer pairs. The set of standard question and answer pairs can be automatically generated or manually entered based on product knowledge data, and supports label classification and multi-level association relationship construction according to the depth of the question and the stage of the conversation.

[0011] The sparring simulation module is used to generate a virtual customer role based on configuration parameters, including the virtual customer's identity, personality style, communication scenario, and knowledge point targets to be trained. The sparring simulation module has the ability to model and memorize the conversation context and generate logically coherent virtual customer responses based on the student's responses.

[0012] The stage control module is used to divide the conversation process between virtual customers and trainees into five stages: conversation initiation, demand exploration, objection handling, product matching and recommendation, and conversation conclusion. It can automatically identify the current stage of the conversation and adjust the virtual customer's behavioral strategy and response style based on the stage characteristics.

[0013] The real-time evaluation module is used to evaluate each round of students' responses during the conversation. The evaluation dimensions include professionalism, emotional management, communication expression and content completeness, and generate real-time feedback information.

[0014] Furthermore, the question-answer library construction and management module also includes: an automatic generation unit, configured to call a large model to automatically generate question-answer pairs from the product knowledge data in the product knowledge base module; The classification unit is used to label and classify the generated question-answer pairs according to "question depth" or "discourse stage", and to build a multi-level question association relationship from shallow questions to in-depth questions.

[0015] Furthermore, it also includes a supernatural voice dialogue module, which trains a voice model based on real customer conversation recordings to simulate the speaking speed, intonation, tone and emotional expression of different virtual customers, and supports role-based voice style output, including young female style, serious middle-aged male style and style with local accent.

[0016] Furthermore, it also includes a speech prompt module, which is linked with the product knowledge base module, the question and answer library construction and management module and the stage control module, and is used to generate speech suggestions that are suitable for the scenario in real time according to the current dialogue stage, dialogue context and the student's response content. The speech suggestions include an introduction to the product's advantages, wording for dealing with objections and examples of guided questions.

[0017] Furthermore, it also includes a dialogue summary report module, which is linked with the real-time evaluation module and the stage control module to generate a structured training report containing an overall score, key weaknesses extraction and improvement suggestions based on the real-time evaluation results and the completion status of each stage after each sparring session.

[0018] The present invention also provides a large-scale model-based simulated dialogue training method, which specifically includes the following steps: Step S1: Construct a product knowledge base and question-and-answer database: parse product documents, business manuals, and frequently asked questions to form product knowledge data, generate or enter standard question-and-answer pairs based on the product knowledge data, and label and classify the standard question-and-answer pairs according to the depth of the questions and the stage of the conversation, and construct multi-level association relationships.

[0019] Step S2: Configure virtual customer parameters: set the virtual customer's identity, personality style, communication scenario, and knowledge point targets to be trained, and generate a virtual customer role with specific behavioral characteristics.

[0020] Step S3: Conduct multiple rounds of dynamic dialogue simulation: Based on the product knowledge base, question and answer library and virtual customer parameters, the trainee conducts multiple rounds of dialogue with the virtual customer, and the virtual customer generates a logically coherent response based on the dialogue context and the trainee's response content.

[0021] Step S4: Conversation stage control: Divide the conversation process into five stages: conversation start, demand mining, objection handling, product matching and recommendation, and conversation closing. Automatically identify the current conversation stage and adjust the virtual customer's behavior strategy and response style.

[0022] Step S5: Real-time evaluation and report generation: Evaluate each round of students' responses from the dimensions of professionalism, emotional management, communication expression and content completeness and generate real-time feedback. After the training is completed, a structured training report is generated based on the evaluation results and the completion status of each stage.

[0023] Furthermore, in step S1, "generating standard question-answer pairs" includes: calling a large model to automatically generate question-answer pairs from product knowledge data, or manually entering preset question-answer pairs; the "labeling classification" includes marking the depth of questions according to "straightforward questions / in-depth questions", or marking the stage of conversation according to "guidance / objection / dealing".

[0024] Furthermore, in step S2, "personality style" includes hesitant, impulsive, skeptical, and rational; "communication scenario" includes online consultation, telephone sales, and offline visits; and the behavioral characteristics of the virtual customer role are dynamically adjusted according to the configuration parameters.

[0025] Furthermore, it also includes a supernatural voice interaction step: a voice model trained based on real customer conversation recordings simulates the virtual customer's speaking speed, intonation, tone and emotional expression, outputs a role-based voice that matches the virtual customer's identity and personality, and realizes multi-round conversation simulation in voice form.

[0026] Furthermore, step S3 also includes the following steps: Based on the current dialogue stage, dialogue context and the student's response content, adaptive information is retrieved from the product knowledge base and question-and-answer library to generate and push suggestions for product introductions, responses to objections or guiding questions, to assist students in optimizing their responses.

[0027] Beneficial effects of the present invention: The large-scale model-based realistic dialogue training system and method of the present invention have significant advantages in terms of system design, functional modules, and user experience. Its advantages can be summarized as follows: (1) Highly realistic customer simulation and natural and coherent interaction: The system of the present invention supports the construction of multiple customer portraits (including personality, identity, industry and other dimensions), and has the ability to remember context and perceive the dialogue stage. It can dynamically adjust customer behavior and language style according to the content of the user dialogue, significantly improving the authenticity and immersion of the training, far exceeding the traditional script-based or static question-and-answer training system.

[0028] (2) Supporting knowledge-oriented training with flexible content coverage: The system supports uploading any product documents, configuring knowledge points, and generating question-answer trees through the product knowledge base and question-answer library construction mechanism, and generates personalized training tasks based on this, realizing a complete chain of training from product memory to actual combat expression, avoiding the problems of traditional methods such as single content and difficulty in updating.

[0029] (3) Configurable training objectives and controllable processes: The present invention innovatively introduces a dialogue stage control mechanism and a task-driven model, which divides the customer dialogue process into standard stages and imposes behavioral constraints, so that the training process has clear goals and a sense of rhythm, which is closer to actual business scenarios, and overcomes the problems of "unclear processes and inaccurate feedback" in the existing technology.

[0030] (4) Supports supernatural voice simulation to enhance the actual combat experience: Different from traditional TTS technology, this system adopts corpus training method to provide a variety of anthropomorphic voice styles to simulate the tone, personality and expression habits of different customers. It is especially suitable for voice call scenarios and improves the response training effect of voice-related personnel.

[0031] (5) Guided speech prompts to help improve expression skills: The system intelligently recommends scenario-appropriate speech suggestions based on the dialogue stage and context, helping users improve speech logic, word selection skills and product association capabilities, and achieve "learning while practicing" during the training process, greatly shortening the learning curve.

[0032] (6) Real-time scoring and dialogue reporting, quantifiable and replayable training: This invention provides a two-level evaluation mechanism of sentence-by-sentence scoring and training reporting, which evaluates user performance from multiple dimensions such as professionalism, emotional control, expression logic, and content completeness, and generates an overall report and optimization suggestions, solving the problem of subjective feedback and inconsistent standards in traditional real-person training. (7) Highly automated, low-cost, and easy to deploy: The system supports low-code configuration and flexible content updates, and can quickly adapt to different industries, products, and training objectives. It does not require continuous reliance on instructors and organizational costs, significantly reducing training organization and implementation costs, and has good scalability and commercial value. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The specific embodiments of the present invention will be further explained below with reference to the accompanying drawings.

[0034] Figure 1 This is a system block diagram of the large model-based simulated dialogue training system of the present invention.

[0035] Figure 2 This is a flow chart of the large model-based simulated dialogue training method of the present invention.

[0036] Figure 3This is a flowchart of the steps for building a product knowledge base and a question-and-answer base in the large-model-based simulated dialogue training method of the present invention. DETAILED DESCRIPTION Example 1

[0037] Combine Figure 1 This embodiment provides a large-scale model-based realistic dialogue training system, which specifically includes the following modules: The product knowledge base module is used to store and parse structured or unstructured product documents, business manuals and FAQs to form searchable product knowledge data.

[0038] The question and answer library construction and management module is linked with the product knowledge base module to construct and manage a set of standard question and answer pairs. The set of standard question and answer pairs can be automatically generated or manually entered based on product knowledge data, and supports label classification and multi-level association relationship construction according to the depth of the question and the stage of the conversation.

[0039] In this embodiment, preferably, the question-answer library construction and management module further includes: an automatic generation unit, configured to call a large model to automatically generate question-answer pairs from the product knowledge data in the product knowledge base module; The classification unit is used to label and classify the generated question-answer pairs according to "question depth" or "discourse stage", and to build a multi-level question association relationship from shallow questions to in-depth questions.

[0040] The large-model-based simulated dialogue training system of this embodiment uses a training task generation mechanism based on a knowledge base and a question-answer tree to parse product documents into a structured knowledge base, and uses a combination of question-answer generation and manual maintenance to construct standard question-answer pairs. It also supports labeling by question depth, speech stage, etc., and constructs a multi-level question tree structure, and uses the above content for training task generation and dynamic loading of training content.

[0041] The sparring simulation module is used to generate a virtual customer role based on configuration parameters, including the virtual customer's identity, personality style, communication scenario, and knowledge point targets to be trained. The sparring simulation module has the ability to model and memorize the conversation context and generate logically coherent virtual customer responses based on the student's responses.

[0042] Specifically, in this embodiment, the customer's identity, personality traits, questioning style, communication style, and behavioral changes are dynamically displayed during the simulation process through parameter configuration, and support is provided for automatic generation or selection of different types of customer dialogue paths, thereby realizing a multi-dimensional role modeling and configuration mechanism for customer behavior simulation.

[0043] The stage control module is used to divide the conversation process between virtual customers and trainees into five stages: conversation initiation, demand exploration, objection handling, product matching and recommendation, and conversation conclusion. It can automatically identify the current stage of the conversation and adjust the virtual customer's behavioral strategy and response style based on the stage characteristics.

[0044] Specifically, in this embodiment, by introducing a stage-controlled dialogue flow guidance mechanism, the customer communication process is divided into five stages. The system has the ability to automatically identify the current dialogue stage and call different customer behavior models and speech strategies in different stages, thereby achieving dialogue rhythm control that is closer to the real process.

[0045] The real-time evaluation module is used to evaluate each round of students' responses during the conversation. The evaluation dimensions include professionalism, emotional management, communication expression and content completeness, and generate real-time feedback information.

[0046] Specifically, in this embodiment, based on the intelligent evaluation system with multi-dimensional scoring indicators, the system automatically evaluates the performance of each user's reply in terms of professionalism, expressiveness, emotional control, content completeness, etc., supports the generation of sentence-by-sentence feedback and overall training reports, and evaluates training effects and ability shortcomings.

[0047] Preferably, this embodiment also includes a supernatural voice dialogue module, which trains a voice model based on real customer conversation recordings, and is used to simulate the speaking speed, intonation, tone and emotional expression of different virtual customers, and supports role-based voice style output, wherein the role-based voice styles include young female style, serious middle-aged male style and style with local accent.

[0048] Specifically, in this embodiment, a supernatural voice interaction technology that combines voice simulation and emotional expression is adopted, and multi-style voice samples are used to train a role-based voice model, supporting the output of natural voice features of different personalities and identities, simulating the customer's tone, rhythm, emotion and other expressions, and achieving a more realistic "voice training".

[0049] Preferably, this embodiment also includes a speech prompt module, which is linked with the product knowledge base module, the question and answer library construction and management module and the stage control module to generate speech suggestions suitable for the scenario in real time according to the current dialogue stage, dialogue context and the student's response content. The speech suggestions include an introduction to the product's advantages, wording for dealing with objections and examples of guided questions.

[0050] Specifically, in this embodiment, a real-time speech prompt and context-enhanced interactive assistance mechanism is used to give real-time speech recommendations based on the current dialogue stage, context, user expression, and knowledge base content to assist trainers in responding to customer questions. This module is linked with the knowledge base to improve the quality of real-time communication.

[0051] Preferably, this embodiment further includes a dialogue summary report module, which is linked to the real-time evaluation module and the stage control module to generate a structured training report containing an overall score, key weaknesses extraction and improvement suggestions based on the real-time evaluation results and the completion status of each stage after each sparring session.

[0052] Specifically, in this embodiment, a task completion-oriented comprehensive report generation mechanism is adopted. After each training session, the system evaluates the task achievement based on the overall task settings; combined with the scores of each stage and user performance, a personalized report is generated, including scores, questions, and suggestions.

[0053] This embodiment of the large-model-based realistic conversation training system focuses on customer training objectives, employing structured task design and modular system construction for the large language model's capabilities to achieve a highly simulated, interactive, controllable, and evaluable customer conversation simulation environment. Through the collaborative work of these multiple functional modules, a highly immersive, highly realistic, and highly responsive intelligent customer training system is created, truly achieving the goal of "using AI to play the role of customer to help employees practice products, strategies, and responses," significantly improving training efficiency, quality, and replicability. Example 2

[0054] Combine Figure 2 and Figure 3 This embodiment provides a method for training realistic dialogue based on a large model, which specifically includes the following steps: Step S1: Construct a product knowledge base and question-and-answer database: parse product documents, business manuals, and frequently asked questions to form product knowledge data, generate or enter standard question-and-answer pairs based on the product knowledge data, and label and classify the standard question-and-answer pairs according to the depth of the questions and the stage of the conversation, and construct multi-level association relationships.

[0055] Preferably, in step S1 of this embodiment, "generating standard question-answer pairs" includes: calling a large model to automatically generate question-answer pairs from product knowledge data, or manually entering preset question-answer pairs; the "labeling classification" includes marking the depth of questions according to "straightforward questions / in-depth questions", or marking the stage of conversation according to "guidance / objection / dealing".

[0056] Step S2: Configure virtual customer parameters: set the virtual customer's identity, personality style, communication scenario, and knowledge point targets to be trained, and generate a virtual customer role with specific behavioral characteristics.

[0057] In this embodiment, preferably, in step S2, "personality style" includes hesitant, impulsive, skeptical, and rational; "communication scenario" includes online consultation, telephone sales, and offline visits; and the behavioral characteristics of the virtual customer role are dynamically adjusted according to the configuration parameters.

[0058] Step S3: Conduct multiple rounds of dynamic dialogue simulation: Based on the product knowledge base, question and answer library and virtual customer parameters, the trainee conducts multiple rounds of dialogue with the virtual customer, and the virtual customer generates a logically coherent response based on the dialogue context and the trainee's response content.

[0059] In this embodiment, preferably, step S3 further includes a speech prompting step: Based on the current dialogue stage, dialogue context and the student's response content, adaptive information is retrieved from the product knowledge base and question-and-answer library to generate and push suggestions for product introductions, responses to objections or guiding questions, to assist students in optimizing their responses.

[0060] Step S4: Conversation stage control: Divide the conversation process into five stages: conversation start, demand mining, objection handling, product matching and recommendation, and conversation closing. Automatically identify the current conversation stage and adjust the virtual customer's behavior strategy and response style.

[0061] Step S5: Real-time evaluation and report generation: Evaluate each round of students' responses from the dimensions of professionalism, emotional management, communication expression and content completeness and generate real-time feedback. After the training is completed, a structured training report is generated based on the evaluation results and the completion status of each stage.

[0062] Preferably, this embodiment also includes a supernatural voice interaction step: a voice model trained based on real customer conversation recordings simulates the virtual customer's speaking speed, intonation, tone and emotional expression, outputs a role-based voice that matches the virtual customer's identity and personality, and realizes multi-round conversation simulation in voice form.

[0063] The large-model-based realistic dialogue training method of this embodiment achieves the following technical goals: (1) Build a flexible and diverse customer role simulation mechanism: It can generate different types of "virtual customer" roles based on set parameters, covering a variety of behavioral characteristics, personalities, problem concerns and communication styles, thereby supporting a training experience that is closer to real business scenarios.

[0064] (2) Supporting multi-round dynamic dialogue simulation with continuous context: With complete dialogue context modeling and memory capabilities, it can accurately understand the content of students' answers during training and make logically coherent and semantically natural customer responses, thus achieving immersive training.

[0065] (3) The task objectives and evaluation mechanism are introduced to achieve quantitative evaluation of training effects: each training session can be configured with specific task objectives (such as completing introductions, handling objections, facilitating order placement, etc.), and the training process can be automatically recorded in a structured manner, the performance of trainees in key links can be evaluated, and personalized feedback and improvement suggestions can be generated.

[0066] (4) Reduce training costs and improve efficiency and scalability: Multiple types of customer scenarios can be built on demand, supporting rapid adaptation to different product knowledge bases and business goals, achieving standardization, automation, and reusability of training resources, and significantly reducing the manpower and time costs of organizational training.

[0067] The large-model-based simulated dialogue training method of this embodiment replaces the inefficient links in traditional training methods with intelligent means, improves the quality, standardization and replicability of training, and has the characteristics of high interactivity, diversified scenario support, reusable training process and perfect evaluation mechanism. It can truly meet the practical needs in modern training scenarios and ultimately help enterprises build more professional and efficient customer front-line teams; it has highly realistic dialogue capabilities, flexible configuration capabilities and perfect evaluation and feedback mechanisms, and can effectively replace traditional scripted training or real-person simulation training methods, improve the communication and practical ability of front-line personnel such as sales and customer service, and can be widely used in scenarios such as corporate training, new employee on-the-job training, and product promotion preparation.

[0068] In the above description, many specific details are set forth in order to fully understand the present invention. However, the above description is only a preferred embodiment of the present invention. The present invention can be implemented in many other ways different from those described herein, so the present invention is not limited to the specific implementation disclosed above. At the same time, any person skilled in the art can make many possible changes and modifications to the technical solution of the present invention using the methods and technical contents disclosed above without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A large-scale model-based realistic dialogue training system, characterized by: Specifically, it includes the following modules: The product knowledge base module is used to store and parse structured or unstructured product documents, business manuals and FAQs to form searchable product knowledge data; A question-and-answer library construction and management module, linked to the product knowledge base module, is used to construct and manage a set of standard question-and-answer pairs. The set of standard question-and-answer pairs can be automatically generated or manually entered based on product knowledge data, and supports labeling and classification based on question depth and conversation stage, as well as the construction of multi-level association relationships; A training simulation module is used to generate a virtual customer character based on configuration parameters, including the virtual customer's identity, personality style, communication scenarios, and knowledge points to be trained. The training simulation module has the ability to model and memorize conversation context and generate logically coherent virtual customer responses based on the student's responses. The stage control module is used to divide the conversation process between virtual customers and trainees into five stages: conversation initiation, demand exploration, objection handling, product matching and recommendation, and conversation conclusion. It can also automatically identify the current stage of the conversation and adjust the virtual customer's behavioral strategy and response style based on the stage characteristics; The real-time evaluation module is used to evaluate each round of students' responses during the conversation. The evaluation dimensions include professionalism, emotional management, communication expression and content completeness, and generate real-time feedback information.

2. The large model-based simulated dialogue training system according to claim 1, characterized in that: The question-answer database construction and management module also includes: an automatic generation unit, configured to call a large model to automatically generate question-answer pairs from the product knowledge data in the product knowledge base module; The classification unit is used to label and classify the generated question-answer pairs according to "question depth" or "discourse stage" and to build a multi-level question association relationship from shallow questions to in-depth questions.

3. The large model-based realistic dialogue training system according to claim 1, characterized in that: It also includes a supernatural voice dialogue module, which trains a voice model based on real customer conversation recordings to simulate the speaking speed, intonation, tone and emotional expression of different virtual customers, and supports role-based voice style output, including young female style, serious middle-aged male style and style with local accent.

4. The large model-based realistic dialogue training system according to claim 1, characterized in that: It also includes a speech prompt module, which is linked to the product knowledge base module, the question and answer library construction and management module and the stage control module, and is used to generate speech suggestions that are suitable for the scenario in real time according to the current dialogue stage, dialogue context and the student's response content. The speech suggestions include an introduction to the product's advantages, terms for dealing with objections and examples of guided questions.

5. The large model-based realistic dialogue training system according to claim 1, characterized in that: It also includes a dialogue summary report module, which is linked to the real-time evaluation module and the stage control module to generate a structured training report containing an overall score, key weaknesses extraction and improvement suggestions based on the real-time evaluation results and the completion status of each stage after each sparring session.

6. A method for training realistic dialogue based on a large model, characterized by: The following steps are involved: Step S1: Build a product knowledge base and question-answer database: Analyze product documents, business manuals, and FAQs to form product knowledge data, generate or enter standard question-answer pairs based on the product knowledge data, and label and classify the standard question-answer pairs according to the depth of the questions and the stage of conversation, and build multi-level association relationships; Step S2: Configure virtual customer parameters: Set the virtual customer's identity, personality style, communication scenarios, and knowledge points to be trained, and generate a virtual customer role with specific behavioral characteristics; Step S3: Conducting multiple rounds of dynamic dialogue simulation: Based on the product knowledge base, question and answer database, and virtual customer parameters, the trainee conducts multiple rounds of dialogue with the virtual customer, and the virtual customer generates a logically coherent response based on the dialogue context and the trainee's answers; Step S4: Conversation stage control: Divide the conversation process into five stages: conversation initiation, demand discovery, objection handling, product matching and recommendation, and conversation conclusion. Automatically identify the current conversation stage and adjust the virtual customer's behavior strategy and response style. Step S5: Real-time evaluation and report generation: Evaluate each round of students' responses from the dimensions of professionalism, emotional management, communication expression and content completeness and generate real-time feedback. After the training is completed, a structured training report is generated based on the evaluation results and the completion status of each stage.

7. The large model-based simulated dialogue training method according to claim 6, characterized in that: In step S1, "generating standard question-answer pairs" includes: calling a large model to automatically generate question-answer pairs from product knowledge data, or manually entering preset question-answer pairs; the "labeling classification" includes marking the depth of questions according to "straightforward questions / in-depth questions", or marking the stage of conversation according to "guidance / objection / dealing".

8. The large model-based simulated dialogue training method according to claim 6, characterized in that: In step S2, "personality style" includes hesitant, impulsive, skeptical, and rational; "communication scenario" includes online consultation, telephone sales, and offline visits; and the behavioral characteristics of the virtual customer role are dynamically adjusted according to the configuration parameters.

9. The large model-based simulated dialogue training method according to claim 6, characterized in that: It also includes supernatural voice interaction steps: a voice model trained based on real customer conversation recordings simulates the virtual customer's speaking speed, intonation, tone and emotional expression, outputs a role-based voice that matches the virtual customer's identity and personality, and realizes multi-round conversation simulation in voice form.

10. The large model-based simulated dialogue training method according to claim 6, characterized in that: Step S3 also includes the following steps: Based on the current dialogue stage, dialogue context and the student's response content, adaptive information is retrieved from the product knowledge base and question-and-answer library to generate and push suggestions for product introductions, responses to objections or guiding questions, to assist students in optimizing their responses.

Images