Intelligent drilling system and method based on cooperative work of multiple agents, medium, device and program product

Abstract

The invention provides an intelligent drilling system and method based on cooperative work of multiple agents, a medium, a device and a program product, different types of virtual customers are created through a large model, multiple agents cooperate to form a perfect training system, the quality control standard problem caused by traditional manual drilling is solved, resources are released, efficiency is improved, and the training effect is improved. And the training data can be stored for tracing and analysis. Meanwhile, the problems that first-generation intelligent drilling is difficult to open flexibly, the course configuration cost is high, and the period is long are solved. According to the intelligent drilling, a virtual customer is generated through a large language model in a one-key mode, and a vivid and challenging dialogue scene in insurance marketing is created; adjusting the session theme at any time according to the context of the session and the emotion and demand of the customer; the virtual clients with characters and attitudes actively ask and inversely ask, so that the natural fluency and actual combat experience of the dialogue are improved. Meanwhile, the verbal skill assistant provides real-time support and suggestions in the dialogue process, and helps students to better understand and solve the demands of clients. And after the drilling is finished, the intelligent drilling provides analysis of a customer portrait solution thought, carries out comprehensive evaluation on the whole dialogue record, and provides feedback and improvement suggestions. In the process, students are helped to convert knowledge into ability.

Description

Technical Field

[0001] The present application relates to the field of large model technology, and in particular to an intelligent rehearsal system, method, medium, device and program product based on the collaborative work of multiple intelligent agents. Background Art

[0002] Under the traditional insurance training model, agents need to have one-on-one offline simulated interviews with supervisors to achieve the training goal of clearance drills. With the standardization of training effects, training efficiency, and retention of drill data, online drills are also being explored in the training field. Since the first generation of intelligent drills are based on tasks and FAQs, there are also various problems, such as the pre-written clearance process is not flexible, the training content is memorized scripts and is not practical, the course update burden is heavy, and the time is long. Therefore, it is still difficult for users to improve the efficiency and effectiveness of training. Trainee use: How to achieve data recording and retention of the drill process for subsequent data analysis and traceability. Therefore, the current insurance industry training model has at least the following technical problems:

[0003] Traditional offline drills:

[0004] (1) The issue of quality control standards for drills: How to ensure consistency in course passing standards among different supervisors to improve training effectiveness.

[0005] (2) Exercise efficiency issues: How to reduce the time invested by supervisors in the training process and increase the frequency of agents’ exercises.

[0006] (3) Exercise data retention issue: How to record and retain the data during the exercise process for subsequent data analysis and traceability.

[0007] First generation smart drill

[0008] (1) Problems with drill effectiveness: Pre-setting interview questions and processes makes the experience rigid and inflexible, making it difficult for trainees to exercise their initiative and creativity.

[0009] (2) The problem of drill output efficiency: The course output is slow, and a lot of setup work is required, including pre-setting all questions and answers and processes, formulating detailed and extensive review standards, and a large amount of tuning work. Summary of the invention

[0010] In view of the shortcomings of the prior art mentioned above, the purpose of this application is to provide a method for solving technical problems in the existing insurance training model in terms of exercise quality control, exercise efficiency, exercise data retention, exercise effect, and exercise output efficiency.

[0011] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present application provides an intelligent rehearsal system based on the collaborative work of multiple agents, including: a customer portrait generation agent, which is used to construct structured data instructions based on insurance industry customer data to guide the large model to generate customer portraits of virtual insurance customers with different personality characteristics; a custom portrait inspection and polishing agent, which is used to inspect the custom customer portraits generated in response to user operations, and polish the customer portraits that lack necessary structured information; a situation introduction generation agent, which is used to generate dialogue scenarios for the insurance industry based on customer portraits; a training dialogue generation agent, which is used to output insurance business dialogue content based on simulated customer portraits and dialogue scenarios as input; a speech assistant agent, which is used to provide corresponding reply suggestions in the generated insurance industry dialogue scenarios based on the questions raised by the virtual customer, combined with the customer portrait, scenario description and the context of the dialogue; an intelligent ending judgment agent, which is used to dynamically generate corresponding dialogue endings based on the context of the dialogue, customer emotional characteristics and demand characteristics, and the current interaction status.

[0012] In some embodiments of the first aspect of the present application, the intelligent drill system further includes any one or two of the following agents: a customer analysis strategy agent, which is used to perform tracing and analysis based on the customer portrait, scenario description, and dialogue record after the dialogue is over, and analyze the characteristics of the virtual customer portrait through comprehensive learning and reasoning, and provide prediction problems and solutions. A drill evaluation agent, which is used to generate evaluation information on the training drill based on the dialogue record after the dialogue is over.

[0013] In some embodiments of the first aspect of the present application, the method for constructing the customer portrait generation intelligence includes: collecting customer data in the insurance industry and constructing structured data instructions, and using the structured data instructions to preprocess and extract features of the collected customer data; the structured data instructions are used to convert the original data into a structured format suitable for machine learning training; and the extracted features are input into the GPT model for model training to construct the customer portrait generation agent.

[0014] In some embodiments of the first aspect of the present application, the inspection content of the customer portrait by the custom portrait inspection and polishing agent includes: completeness check, which is used to verify whether the customer portrait input by the user contains all necessary structured information, and generate a series of questions or prompts based on the missing information points to guide the user to supplement the missing information, and check the integrity of the customer portrait after the user supplements the information; personal expression check, which is used to check whether the customer portrait is expressed in the second person; key information point check, which is used to check whether the customer portrait contains at least the following key information points: personal identity, relationship with the agent, personal situation, attitude towards insurance, purchased insurance, etc.

[0015] In some embodiments of the first aspect of the present application, the method of constructing the situational introduction generation agent includes: using NLP technology to parse the customer portrait, identify and extract key information points of the virtual customer; using a machine learning algorithm to extract deep-level features containing implicit needs and personality traits of the virtual customer from the customer portrait; inputting the extracted key information points and deep-level features of the virtual customer into the GPT model for training so that it can understand and simulate the customer's behavior and language style; constructing one or more dialogue frameworks based on the characteristics of the customer portrait, and the dialogue frameworks cover the interaction paths of different business scenarios; in each dialogue framework, using the trained GPT model to generate dialogue content that conforms to the customer's characteristics; obtaining the user's evaluation and feedback information on the dialogue scene during the dialogue process, and optimizing the GPT model through multiple rounds of iterations.

[0016] In some embodiments of the first aspect of the present application, the method for constructing the training dialogue generation agent includes: receiving a customer portrait of a virtual customer from a customer portrait generation agent, receiving a polished user-defined customer portrait from a custom portrait checking and polishing agent, and receiving simulated dialogue scene content from a scenario introduction generation agent, and preprocessing and extracting features of the received data; the GPT model understands the customer's intentions and needs based on the input features and contextual information; uses the language patterns and structures of the insurance industry learned during its training to construct and output dialogue content that conforms to the role and scenario; wherein the language patterns and structures of the insurance industry include: the use of professional terms, inquiry and question-answering patterns, risk assessment and suggestions, claims process description, compliance and standardization, and customer service language.

[0017] To achieve the above-mentioned purpose and other related purposes, the second aspect of the present application provides an intelligent rehearsal method based on the collaborative work of multiple intelligent agents, and constructs multiple intelligent agents based on the GPT model to work together to perform the following: construct structured data instructions based on insurance industry customer data to guide the large model to generate customer portraits of virtual insurance customers with different personality characteristics; check the custom customer portraits generated in response to user operations, and polish the customer portraits that lack necessary structured information; generate dialogue scenarios for the insurance industry based on the customer portraits; output insurance business dialogue content based on the simulated customer portraits and dialogue scenarios as input; in the generated insurance industry dialogue scenarios, provide corresponding reply suggestions based on the questions raised by the virtual customers, combined with the customer portraits, scenario descriptions and the context of the dialogue; dynamically generate corresponding dialogue endings based on the context of the dialogue, the customer's emotional characteristics and demand characteristics, and the current interaction status.

[0018] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent rehearsal method based on the collaborative work of multiple intelligent agents.

[0019] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer implements the intelligent rehearsal method based on the collaborative work of multiple intelligent agents.

[0020] To achieve the above-mentioned purpose and other related purposes, the fifth aspect of the present application provides a computer device, including a memory, a processor and a computer program stored in the memory; the processor executes the computer program to implement the intelligent rehearsal method based on the collaborative work of multiple intelligent agents.

[0021] As described above, the intelligent rehearsal system, method, medium, device and program product based on the collaborative work of multiple intelligent agents of the present application have the following beneficial effects:

[0022] (1) Different types of virtual customers are created through large models, and multiple intelligent agents collaborate to form a complete training system. The training efficiency is also greatly improved. All data generated during the training process can be preserved intact and can be traced and analyzed.

[0023] (2) The large model can be used to create realistic and challenging conversation scenarios, and can also dynamically generate appropriate closing remarks based on the context of the conversation, the customer's emotions and needs, and the current state of interaction, thereby improving the natural fluency of the conversation and the service experience.

[0024] (3) The large model automatically generates virtual customers, and users can customize and polish customer portraits, providing trainees with a variety of interactive practice opportunities. At the same time, the conversation assistant provides real-time support and suggestions during the conversation, helping trainees better understand and respond to customer needs. It also conducts a comprehensive evaluation of the entire conversation record, provides feedback and improvement suggestions, and further improves the training effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Shown is a structural diagram of an intelligent rehearsal system based on the collaborative work of multiple intelligent agents in one embodiment of the present application.

[0026] Figure 2 Shown is a schematic diagram of a customer portrait of a virtual customer in one embodiment of the present application.

[0027] Figure 3Shown is a schematic diagram of the process of a customer portrait generation agent generating a customer portrait in one embodiment of the present application.

[0028] Figure 4 Shown is a schematic diagram of a GUI interface for checking items for a customer portrait in one embodiment of the present application.

[0029] Figure 5 Shown is a schematic diagram of the process of generating a simulated dialogue scene by a situational introduction generating agent in one embodiment of the present application.

[0030] Figure 6 Shown is a flowchart of an intelligent rehearsal method based on the collaborative work of multiple intelligent agents in one embodiment of the present application.

[0031] Figure 7 Shown is a schematic diagram of the structure of a computer device in one embodiment of the present application. DETAILED DESCRIPTION

[0032] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0033] Before further describing the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are applicable to the following interpretations:

[0034] <1> GPT (Generative Pre-trained Transformer): A natural language processing (NLP) model based on the Transformer architecture. The GPT model learns the patterns and structures of language by pre-training a large amount of text data, and can then be fine-tuned on a variety of language tasks, such as text generation, translation, question answering, etc. The core of the GPT model is the Transformer, which is an attention mechanism model that can process sequence data and capture long-distance dependencies.

[0035] <2> Agent: A concept in the field of computer science and artificial intelligence that refers to an entity that can perceive the state of the environment in a specific environment and make decisions to achieve specific goals. Agents can be in the form of software, such as chatbots and recommendation systems; they can also be in the form of hardware, such as self-driving cars and robots.

[0036] This application aims to simulate real customer conversation scenarios through the collaboration of multiple agents. These agents are carefully designed to perform specific tasks, such as understanding, responding, and sentiment analysis, and together form an efficient training environment. Through this multi-model collaborative approach, the system can provide a more natural, lively, and challenging communication experience, effectively improving users' communication skills and problem-solving abilities.

[0037] Figure 1 The structure diagram of an intelligent rehearsal system based on the collaborative work of multiple agents in an embodiment of the present invention is shown. The intelligent rehearsal system in this embodiment includes: a customer portrait generation agent 101, a custom portrait check and polishing agent 102, a situation introduction generation agent 103, a training dialogue generation agent 104, a speech assistant agent 105, an intelligent end judgment agent 106, a customer analysis idea agent 107 and a rehearsal evaluation agent 108.

[0038] The customer profile generation agent 101 is used to construct structured data instructions based on insurance industry customer data to guide the large model to generate customer profiles of virtual insurance customers with different personality traits. These customer profiles will simulate customers with different personalities and provide trainees with diverse interactive exercise opportunities. For example, the customer profile generation agent can generate a customer profile of a young single professional who may have high requirements for health insurance and accidental injury insurance; or generate a customer profile of a middle-aged housewife who may be more concerned about pension insurance and children's education insurance. In this way, trainees can interact with different types of customers in a simulated environment to improve their communication skills and sales capabilities.

[0039] The process of generating a customer profile of a virtual insurance customer by the customer profile generating agent 101 is as follows: Figure 3 As shown:

[0040] Step S3a: Collect customer data of the insurance industry and construct structured data instructions, and use the structured data instructions to preprocess and extract features of the collected customer data; the customer data is simulated customer data based on a real data structure and is desensitized.

[0041] Figure 2A schematic diagram of a customer portrait of a virtual customer in an embodiment of the present application is shown. Customer data in the insurance industry includes but is not limited to basic customer information, policy details, claim record information, etc. The basic customer information includes personal information (such as name, gender, age, occupation, educational background, etc.), contact information (such as telephone number, email address, etc.), family and financial status (such as family structure, income level, assets and liabilities, etc.), which are helpful in assessing the customer's ability to pay and insurance needs. Policy details include policy type (such as medical insurance, serious illness insurance, annuity insurance and other insurance types purchased by the customer), policy terms (such as insurance liability, exemption clauses, insurance amount, etc.), policy history (such as purchase time, policy term, renewal status and other purchase history information), etc. Claim record information includes the number of claims, claim amount, claim reason (such as disease, accident), etc. It is worth noting that the training data used in the embodiment of the present application is preferably desensitized data, which is simulated customer data. They are based on real data structures and are intended to protect the real information security of customers and ensure that such operations comply with relevant regulatory requirements. In this way, the present application can make full use of data to optimize models and systems without leaking customer personal information.

[0042] Structured data instructions refer to a series of rules and steps defined during the data preprocessing and feature engineering stages to convert raw data into a structured format suitable for machine learning model training. For example, structured data instructions may include: data cleaning instructions, feature selection instructions, feature conversion instructions, feature construction instructions, data encoding instructions, data segmentation instructions, data resampling instructions, time series feature extraction instructions, etc. Data cleaning instructions are used to delete or fill missing values, remove or correct outliers, standardize date formats, etc. Feature selection instructions are used to select features that are highly correlated with the target variable (such as insurance claim amount) and exclude features that are irrelevant or redundant to the target variable. Feature conversion instructions are used to uniquely encode categorical variables, such as converting from "male" and "female" to two binary features, and standardizing or normalizing continuous variables. Feature construction instructions are used to create new features, such as extracting years and months from dates, or deriving new features from original features, such as deriving categories such as "young men" and "middle-aged women" from age and gender features. Data encoding instructions are used to convert text data into numerical data. The data segmentation instruction is used to divide the data set into a training set and a test set, for example, according to the ratio of 70% training set and 30% test set. The data resampling instruction is used to balance the categories of the unbalanced data set using oversampling or undersampling techniques. The time series feature extraction instruction is used to extract trends, seasonality and other features from time series data.

[0043] Step S3b: Input the extracted features into the GPT model for model training to construct the customer portrait generation agent.

[0044] In some examples, the extracted features are combined to form a data set, and the data set is divided into a training set and a validation set in a certain ratio. The training set is input into the GPT model for training, thereby obtaining a customer portrait generation agent. In the process of training the GPT model using the training set, the model parameters are optimized by minimizing the loss function. The model learns how to predict the next most suitable token (the basic unit in the text, such as words, characters, strings, etc.) based on the given context, and the model weights are updated using the gradient descent algorithm to reduce the difference between the predicted value and the actual value. After the training is completed, the GPT model can generate a series of tokens based on the given context, and these tokens are combined to form a customer profile of the virtual insurance customer. Finally, the validation set is used to evaluate the performance of the model, check the accuracy and relevance of the generated customer portrait, and adjust the model parameters or training strategy based on the evaluation results to improve the generalization ability of the model.

[0045] It should be understood that the GPT model is a pre-trained language model based on the Transformer architecture, and its core is the self-attention mechanism, which enables the model to capture the dependencies between any two positions in the input sequence, no matter how far apart they are. During the training process, the model learns how to predict the next most likely token based on a given context, and this ability makes the GPT model very suitable for text generation tasks. When generating virtual customer portraits, the GPT model acts as a powerful text generator that can generate coherent, relevant, and personalized text descriptions based on the input context (which may be the customer's basic attributes or behavior patterns). This generation process is achieved by maximizing the probability of the next token, and the model continuously adjusts its predictions to match the patterns in the training data. In this way, the GPT model is able to learn how to generate detailed customer portraits based on customer data, including their possible preferences, behaviors, and needs.

[0046] The custom portrait checking and polishing agent 102 is used to check the custom customer portraits generated in response to user operations, and to polish the customer portraits that lack necessary structured information.

[0047] It should be noted here that the "user" referred to in the present invention refers to the agent who uses this intelligent training system for training. Therefore, the "agent" below actually has the same meaning as the "user". The "customer" referred to in the present invention refers to the virtual customer generated by the GPT model or customized by the user.

[0048] When the user chooses to input the customer portrait by himself, in order to ensure the completeness and richness of the portrait, the custom portrait checking and polishing agent 102 based on the GPT model parses the text input by the user, compares it with the predefined necessary information standards, and checks whether it contains all the necessary structured information or the information is not detailed enough.

[0049] In some examples, the GPT model-based custom portrait checking and polishing agent 102 checks the customer portrait: completeness check, personal expression check, key information point check, etc. The completeness check verifies whether the customer portrait input by the user contains all the necessary structured information, and generates a series of questions or prompts based on the missing information points to guide the user to supplement the missing information; the customer portrait after the user supplements the information is checked for completeness again, and natural language processing is performed to ensure the detail of the information and the accuracy of the expression. The personal expression check is to check whether the customer portrait is expressed in the second person. The key information point check is to check whether the customer portrait contains at least the following key information points: personal identity, relationship with the agent, personal situation, attitude towards insurance, etc. Among them, in the process of checking the key information points, the GPT model-based custom portrait checking and polishing agent 102 provides prompt information to the user for supplementation or automatically supplements according to the information provided by the user for the customer portrait that lacks key information points or does not meet the standard in terms of information detail. For example, if the user only enters the customer's name and age, the custom portrait checking and polishing agent 102 can automatically add some common information points, such as occupation, income level, family status, etc., to make the portrait closer to the real customer.

[0050] by Figure 4 Taking the displayed GUI interface as an example, the inspection items of customer portrait include:

[0051] 1. Whether it is expressed in the second person, such as "You are Ms. Wang".

[0052] 2. Whether it contains 3 or more of the following information points:

[0053] (1) Who are you? Or how do people usually call you? For example, “You are Ms. Wang.”

[0054] (2) How did you get to know the insurance agent? An old acquaintance? A complete stranger? Was he introduced by someone else?

[0055] (3) Your personal situation, such as whether you are married or have children, your income level, and your personality?

[0056] (4) What is your attitude towards insurance? Do you approve of insurance? Do you have insurance already?

[0057] …

[0058] It should be understood that Figure 4 The content shown is only an illustrative example to assist in understanding the relevant concepts. It is intended to provide an intuitive reference framework to help those skilled in the art better understand. However, in actual application, the specific situation may be different, so it should not be taken as Figure 4 Treated as fixed constraints.

[0059] The scenario introduction generation agent 103 is used to generate dialogue scenarios for the insurance industry based on customer portraits. The scenario introduction generation agent 103 uses advanced GPT large model technology to generate realistic simulated dialogue scenarios based on customer portraits. It not only captures the basic information of customers, but also deeply explores their implicit needs and personality characteristics through deep learning algorithms, creating a realistic and challenging exercise environment.

[0060] It is worth noting that, in the context of the insurance industry, the situational introduction generation agent 103 is able to identify and strategically hide certain real information about the customer, such as personality tendencies, living habits, etc., so as to design a more realistic and variable dialogue situation. For example, even if it is known that the customer is actually very introverted, the agent may set a scenario that requires extroverted interaction, forcing the trainee to respond with incomplete information, and exercising his or her adaptability and insight. In addition, the situational introduction generation agent 103 takes into account the diversity and complexity of insurance products, ensuring that each generated scenario can specifically train the trainee's sales skills and problem-solving abilities for specific products or services. In this way, trainees can face various possible challenges in a safe environment and be prepared for real situations encountered in actual work.

[0061] In the embodiment of the present application, the situation introduction generation agent 103 realizes the process of generating a realistic simulated dialogue scene according to the customer portrait based on the GPT large model. Figure 5 As shown:

[0062] Step S5a: Use NLP technology to analyze the customer portrait, identify and extract key information points of the virtual customer; use machine learning algorithms to extract deep features including implicit needs and personality traits of the virtual customer from the customer portrait; input the extracted key information points and deep features of the virtual customer into the GPT model for training, so that it can understand and simulate the customer's behavior and language style.

[0063] NLP technology refers to Natural Language Processing technology, a branch of artificial intelligence and linguistics that aims to enable computers to understand, interpret and generate content in human language. Key information points extracted through NLP technology include personal identity, relationship with the agent, personal situation, attitude towards insurance, etc.

[0064] Using machine learning algorithms to extract deep features containing the implicit needs and personality traits of virtual customers from customer portraits, the process includes: selecting features related to implicit needs and personality traits from the original data of customer portraits, such as purchase frequency, product preferences, feedback content, interaction frequency, etc. Using historical data to train machine learning models according to the selected features, such as decision trees, random forests, neural networks, etc., so that the deep learning model can identify implicit needs and personality traits from the features. It should be understood that implicit needs refer to needs that customers do not directly express or even realize themselves, and these needs are hidden in customers' behaviors, preferences, and feedback, and need to be discovered through analysis and mining. Generally speaking, implicit needs include preferences and interests, potential problem solutions, and future demand forecasts. Preferences and interests refer to preferences for certain products or services that are not clearly expressed, potential problem solutions refer to customers facing some problems but have not yet found solutions, and future demand forecasts refer to predicting possible future needs by analyzing customers' current behaviors.

[0065] Step S5b: Based on the characteristics of the customer portrait, construct one or more dialogue frameworks, which cover the interaction paths of different business scenarios; in each dialogue framework, use the trained GPT model to generate dialogue content that conforms to the customer characteristics.

[0066] To facilitate understanding, the following three dialogue frameworks and corresponding interaction paths are used as examples for explanation.

[0067] Dialogue framework 1: Customer consultation and service.

[0068] Business scenario: Customers have questions about insurance products and need to consult specific insurance terms, fees, claims procedures, etc.

[0069] Interaction Path:

[0070] 1. Customers ask specific questions.

[0071] 2. The virtual dialogue system provides accurate information and solutions based on customer problems.

[0072] 3. The customer provides feedback based on the information provided and may need further explanation or assistance.

[0073] 4. The virtual dialogue system provides further assistance or solutions based on customer feedback.

[0074] Conversation content generation method: Use the natural language processing capabilities of the GPT model to generate detailed answers to specific customer questions, such as insurance terms explanation, cost calculation, etc. By simulating real conversations, the model is trained to provide more humane and professional services, for example: "Hello, regarding the health insurance you mentioned, it covers hospitalization expenses, surgical expenses, etc. The specific terms are as follows..."

[0075] Dialogue Framework 2: Claims Processing

[0076] Business scenario: Customers need to submit claims applications or inquire about the claims process.

[0077] Interaction Path:

[0078] 1. The customer raises a claim demand or question.

[0079] 2. The virtual dialogue system guides customers through the claims process and provides necessary guidance and information.

[0080] 3.Customers provide necessary claims information and documents according to the instructions.

[0081] 4. After the virtual dialogue system confirms that the information is correct, submit the claim application.

[0082] Conversation content generation method: Use the GPT model to generate guidance dialogues for the claims process, for example: "In order to help you complete the claim quickly, please provide the following information: date of medical treatment, hospital, and the insurance policy number involved."

[0083] Dialogue Framework 3: Sales and Recruitment Training

[0084] Business scenario: Insurance agents need to improve their sales and recruitment capabilities and conduct simulation training through a virtual dialogue system.

[0085] Interaction Path:

[0086] 1. The agent enters the virtual dialogue system and selects a sales or recruitment scenario.

[0087] 2. The system generates simulated customer conversations based on the selected scenario.

[0088] 3. Agents interact with simulated customers and try to achieve sales or recruitment goals.

[0089] 4. The system provides feedback and improvement suggestions based on the agent’s performance.

[0090] Dialogue content generation: Use the GPT model to generate simulated customer dialogues, simulating real sales scenarios, for example: "Hello, I am considering buying a life insurance policy recently. Can you introduce it to me?" The system provides feedback based on the agent's answer to help them improve their communication skills and sales strategies.

[0091] Step S5c: Obtain the user's evaluation and feedback information of the dialogue scene during the dialogue process, and optimize the GPT model through multiple rounds of iterations based on this information.

[0092] According to the collected user feedback and the results of the automatic scoring model, the GPT model is optimized, including adjusting the parameters of the model, improving the model structure or adding more training data. The GPT model parameters that can be optimized include but are not limited to: learning rate, batch size, optimizer, regularization parameter, loss function weight, sequence length, context length, etc. The learning rate is a parameter that controls the parameter update step size of the model during training. The learning process can be accelerated or slowed down by adjusting the learning rate. The batch size determines the amount of data input into the model during each training. Adjusting the batch size can affect the convergence speed and stability of the model. Optimizers such as AdamW have different effects on the training efficiency and final performance of the model. Regularization parameters such as dropout rate are used to prevent the model from overfitting. These parameters can be adjusted to balance the generalization ability of the model. Loss function weight In multi-task learning, the loss function weights of different tasks can be adjusted to optimize the performance of the model on each task. For the GPT model, the sequence length determines the length of text that the model can handle. Adjusting the sequence length can affect the model's ability to handle long texts. In multi-round dialogues, the context length determines the amount of dialogue history information that the model can remember, which is crucial for generating coherent dialogues.

[0093] The training dialogue generation agent 104 is used to output the insurance business dialogue content based on the customer portrait simulated by the customer portrait generation agent 101, the customer portrait polished by the custom portrait checking and polishing agent 102, and the dialogue scene generated by the situation introduction generation agent 103 as input.

[0094] The training dialogue generation agent 104 uses advanced GPT large model technology to build the entire dialogue process and define the way of interaction. When playing a specific role, GPT not only clarifies its focus, but also carefully designs the communication method to promote more effective information exchange. In this way, we aim to provide a more personalized and efficient service experience. In the insurance industry, effective communication is the key to building customer trust and facilitating transactions. The training dialogue agent 104 simulates real sales conversations to help trainees learn how to guide conversations, how to deal with customer objections, and how to effectively sell products. In addition, the agent can also provide feedback and suggestions to help trainees continuously improve their communication skills.

[0095] In the embodiment of the present application, the implementation process of the training dialogue generation agent 104 is as follows:

[0096] First, the training dialogue generation agent 104 based on the GPT model preprocesses and extracts features from data such as customer portraits and dialogue scene content. It should be understood that these inputs can be multimodal data, such as text, voice or other forms of data. Preprocessing includes but is not limited to text cleaning, word segmentation, removal of stop words, etc., so that the model can better understand and process. Feature extraction refers to extracting key features from preprocessed data, such as keywords, phrases, semantic roles, etc., which are used for subsequent dialogue generation. These comprehensive inputs make the training dialogue closer to the actual situation, allowing trainees to practice and improve their sales skills in a risk-free environment. At the same time, the agent can also provide instant feedback and guidance based on the trainee's performance, so that they can grow into an excellent insurance agent faster.

[0097] Secondly, the GPT model understands the customer's intentions and needs based on the input features and contextual information; it uses the language patterns and structures of the insurance industry learned during its training to construct and output conversation content that is consistent with the role and scenario.

[0098] It should be noted that in the insurance industry, language patterns and structures usually refer to commonly used expressions, sentence patterns, professional terms, and dialogue processes in conversations. These patterns and structures can make conversations more fluent, professional, and in line with industry characteristics. For example: (1) Use of professional terms: The insurance industry has its own specific professional terms, such as "premium", "policy", "claim", "risk assessment", etc. The GPT model will learn the use of these professional terms during the training process so that it can use them accurately when generating conversations. (2) Question and answer pattern: In insurance consultation scenarios, there is usually a pattern of customer questions and insurance consultant answers. For example, a customer may ask: "What coverage does this insurance product include?" The insurance consultant needs to provide specific coverage content and terms. The GPT model will learn this question and answer pattern and imitate this structure when generating conversations. (3) Risk assessment and advice: When talking with customers, insurance consultants will conduct risk assessments based on the customer's situation and provide corresponding insurance advice. The GPT model will learn this evaluation and advice language pattern so that it can ask personalized insurance needs in the conversation. (4) Claims process description: During the claims process, insurance consultants need to explain the claims process and required materials to customers. The GPT model will learn the language pattern of this process description so that it can simulate the customer's real reaction in the conversation and inquire and question the claims steps. (5) Compliance and standardization: Conversations in the insurance industry need to follow strict compliance and standardization requirements. The GPT model will learn these requirements and ensure compliance with laws, regulations and company policies when generating relevant content. (6) Customer service language: The insurance industry emphasizes customer service, so the GPT model will simulate customers of various personalities to train trainees to learn how to communicate with customers using polite, empathetic and professional language to improve customer satisfaction. By learning these language patterns and structures, the GPT model can generate smooth, natural and professional conversation content in the insurance industry to meet customer needs and provide high-quality services.

[0099] The speech assistant agent 105 is used to provide corresponding answer suggestions in the generated insurance industry dialogue scenario based on the questions raised by the virtual customer, combined with the customer portrait, scenario description and the context of the dialogue.

[0100] Customer portraits include the customer's basic information, personality traits, consumption habits, etc. Scenario descriptions refer to the background information when the customer asks questions, such as pre-purchase consultation, after-sales service, or questions during product use. Contextual information refers to the customer's previous communication records, which is crucial to understanding the customer's current questions.

[0101] The speech assistant agent 105 uses the GPT model to provide real-time reply suggestions for students. As a pre-trained language model based on the Transformer architecture, the GPT model learns the common patterns and structures of the language by pre-training on large-scale text data. The GPT model is fine-tuned using customer portraits, scenario descriptions, and contextual data so that the GPT model learns how to predict the most appropriate reply ideas and suggestions based on the input customer data. For example, if a customer expresses doubts about an insurance product, the speech assistant agent will, after receiving the question, suggest to the student in real time to understand the customer's concerns first, and then provide relevant information and data to eliminate the customer's doubts. Or, if the customer is price-sensitive, the agent will suggest to the student to emphasize the value and long-term benefits of the product instead of just focusing on the price, etc. In addition, the speech assistant agent will also teach students some practical speech skills and strategies, such as how to use open-ended questions to guide conversations and how to use affirmative language to enhance persuasion. These skills and strategies can help students communicate better with customers and improve sales results.

[0102] After the GPT model is trained to become a speech assistant agent, it can generate ideas and suggestions for responses based on new customer questions and feedback, relying on the model's generation capabilities. That is, when a customer asks a new question or gives feedback, the speech assistant agent will analyze the input data and generate one or more response suggestions. These suggestions can be direct text replies or ideas and frameworks for responses for further refinement by human customer service. In addition, an evaluation mechanism can be set up to evaluate the effectiveness of responses through indicators such as user satisfaction surveys and response success rates, and the GPT model can be further optimized accordingly.

[0103] In some examples, the training process of the speech assistant agent based on the GPT model includes:

[0104] Step 1. Collect customer profiles, scenario descriptions, and context information, and preprocess these data as data sets for model training. Preprocessing includes data cleaning, formatting, and splitting the data set into training and test sets.

[0105] Step 2. Input the preprocessed training set into the GPT model for fine-tuning. During the fine-tuning process, a loss function (such as the cross entropy loss function) is used to calculate the difference between the model's predicted value and the actual value, and an optimization algorithm such as Adam or SGD is selected to update the model's weights to minimize the loss function.

[0106] Step 3. After the model training is completed, use the test set to test and evaluate the trained GPT model, and evaluate it through model indicators such as F1 score, precision or recall. Based on the evaluation results, decide whether the model's hyperparameters need to be further adjusted.

[0107] The intelligent ending judgment agent 106 is used to dynamically generate corresponding conversation ending words according to the context of the conversation, the emotional characteristics and demand characteristics of the customer, and the current interaction status.

[0108] In the embodiment of the present application, the training process of the intelligent termination judgment agent 106 based on the GPT model includes:

[0109] Step 1. Collect and preprocess data such as context, customer emotion and demand characteristics, and current interaction status. Preprocessing includes data cleaning, formatting, and splitting the data set into training and test sets.

[0110] Step 2. Use the preprocessed training set to input the GPT model for fine-tuning, so that it learns how to identify the natural end point of the conversation and provide an appropriate way to end the conversation based on the input conversation context, emotions, and needs. During the fine-tuning process, a loss function (such as the cross entropy loss function) is used to calculate the difference between the model's predicted value and the actual value, and an optimization algorithm such as Adam or SGD is selected to update the model's weights to minimize the loss function.

[0111] In some examples, the process by which the GPT model identifies natural ending points for a conversation includes:

[0112] First, the GPT model processes the input sequence, which is converted into a numerical form that the model can understand, that is, the text is converted into a sequence of index numbers through tokenization. If the text length exceeds the maximum sequence length set by the model (for example, 1024), only the first 1024 tokens are retained and the remaining tokens are deleted; if the length is insufficient, it is filled with specific tags (such as <|endoftext|>).

[0113] Secondly, the GPT model uses the Self-Attention Mechanism to process contextual information in the input sequence. This mechanism allows the model to generate the next word based on all the words in the input sequence, not just the previous word. The Self-Attention Mechanism calculates the weight score of each word in the input sequence to determine the degree of influence of each word on the generation of the next word.

[0114] Then, based on the current context, the GPT model generates a probability distribution for the next word. This distribution is based on the language patterns learned during model training, and the model predicts the probability of each possible next word.

[0115] Next, the GPT model selects the output token from the probability distribution. This selection can be done through different strategies, such as greedy search (always choose the word with the highest probability), beam search (keep multiple candidate word sequences and choose the sequence with the highest score), or sampling (randomly select the next word according to the probability distribution).

[0116] Next, the generated words are added to the input sequence, and the above steps are repeated until a complete sentence is generated or the set generation length is reached. This process will be repeated until the model recognizes the natural end point of the conversation, that is, the generated output token meets the stopping condition. The stopping condition for the GPT model to recognize the end of the conversation can be a preset generation length, a specific end token (such as a period, question mark, etc.), or the probability distribution of the model predicting the next word, and the probability of the end token is significantly higher than the probability of other words. In this way, the model can determine that the conversation has ended naturally.

[0117] Step 3. After the model training is completed, use the test set to test and evaluate the trained GPT model, and evaluate it through model indicators such as F1 score, precision or recall. Based on the evaluation results, decide whether the model's hyperparameters need to be further adjusted.

[0118] It is worth noting that the traditional procedural ending method is usually based on fixed logic and rules, such as the conversation reaching a certain number of rounds, the appearance of specific keywords, or a preset time limit. This ending method lacks an in-depth understanding of the conversation context and the perception of user emotions, and therefore may not be able to adapt to complex conversation scenarios and changes in user needs. For example, if a user asks a new question or expresses a new need in a conversation, the procedural ending may not capture this, causing the conversation to end when the user still needs help. The intelligent agent used in the present invention utilizes advanced GPT large model technology, which can dynamically generate appropriate endings based on the context of the conversation, the customer's emotions and needs, and the current state of interaction. This approach is not only more natural and smooth, but also better meets customer expectations and improves the overall service experience.

[0119] The customer analysis idea agent 107 is used to trace and analyze the customer portrait, scene description and conversation record after the conversation is over, and learn the problem characteristics of the virtual customer.

[0120] The customer analysis agent 107 based on the GPT model uses data such as customer portraits, scenario descriptions, and conversation records to fine-tune the model. It uses the trained GPT model to identify the customer's key information and potential needs in combination with customer portraits, scenario descriptions, and conversation records. Based on historical data and industry knowledge, the GPT model can predict the customer's possible questions and concerns to help agents prepare in advance. Finally, the analysis results are organized into a report, which includes the customer's key information, potential needs, and predicted questions. In addition, the GPT model also uses the agent's feedback and the customer's subsequent interaction data for continuous learning and optimization of the model to improve the accuracy of the prediction and the relevance of the recommendations.

[0121] The drill evaluation agent 108 is used to generate evaluation information for the training drill based on the dialogue record after the dialogue is completed.

[0122] Specifically, the GPT model is fine-tuned using annotated conversation record data, and the model can generate evaluation information based on the content of the conversation. Using the fine-tuned GPT model, the conversation record is input, and the model gradually generates evaluation information through autoregression, and controls the diversity and accuracy of the generated evaluation information through beam search and other technologies. The generated evaluation information is checked for grammar and logical coherence, and the generated evaluation information is compared with the actual evaluation, and feedback is collected for further optimization of the model. Finally, the generated evaluation information is organized into a report and provided to the agent or relevant personnel. The report can be presented in the form of charts, dashboards, etc., which is more intuitive.

[0123] It should be understood that regressive generation is a text generation method that relies on the model's ability to predict the next word based on the current sequence of words that have been generated. In the GPT model, this generation method is implemented through the following steps: the conversation record is fed into the model as an input sequence; the model uses a self-attention mechanism to understand the context and predict the next most likely word; the model generates text word by word, with each step relying on the previously generated words; this process is repeated until a complete evaluation information is generated or the preset word limit is reached. Beam search is a common technique used in text generation. Instead of selecting only the word with the highest probability at each step, it retains multiple candidate words (forming a "beam") and considers the combined probability of these candidate words. This method can improve the quality and relevance of the generated text because it considers multiple possible word sequences instead of relying solely on the single word with the highest probability. By adjusting the size of the beam, the degree of exploration during the generation process and the quality of the final result can be controlled. Therefore, by combining autoregressive generation and techniques for controlling the quality of generation, the GPT model is able to generate evaluation information that is both consistent with the content of the conversation and has a certain degree of accuracy and diversity. The application of these techniques enables the model to better adapt to different evaluation needs and scenarios.

[0124] In the above, each intelligent agent of the intelligent training system in the embodiment of the present invention is explained in detail. In the following, the intelligent training method, device, medium, etc. will be further described in conjunction with the accompanying drawings.

[0125] like Figure 6 As shown, a flowchart of an intelligent rehearsal method based on the collaborative work of multiple agents in an embodiment of the present invention is shown. The intelligent rehearsal method constructs multiple agents based on the GPT model, so that the collaborative work is performed as follows:

[0126] Step S61: Construct structured data instructions based on insurance industry customer data to guide the large model to generate customer portraits of virtual insurance customers with different personality characteristics.

[0127] Step S62: Check the customized customer portraits generated in response to user operations, and polish the customer portraits that lack necessary structured information.

[0128] Step S63: Generate a dialogue scenario for the insurance industry based on the customer portrait.

[0129] Step S64: Based on the simulated customer portrait and dialogue scenario as input, output the insurance business dialogue content.

[0130] Step S65: In the generated insurance industry dialogue scenario, corresponding answer suggestions are provided based on the questions raised by the virtual customer, combined with the customer portrait, scenario description and the context of the dialogue.

[0131] Step S66: Dynamically generate corresponding conversation ending remarks according to the conversation context, customer emotional characteristics and demand characteristics, and current interaction status.

[0132] It should be understood that the specific process of each module executing the above corresponding steps has been described in detail in the above system embodiment, and for the sake of brevity, it will not be repeated here.

[0133] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" represent examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0134] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can represent: a, b, c, ab, ac, bc or abc, where a, b, c can be single or multiple.

[0135] Figure 7 is a schematic block diagram of a computer device provided in an embodiment of the present application. Figure 7 As shown in FIG. 1 , the computer device includes: at least one processor 701, a memory 702, at least one network interface 703 and a user interface 705. The various components in the device are coupled together via a bus system 704. It can be understood that the bus system 704 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 704 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 704 is not described in detail. Figure 7 In the specification, various buses are labeled as bus systems.

[0136] The user interface 705 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0137] It is understood that the memory 702 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of exemplary but not limiting explanation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.

[0138] The memory 702 in the embodiment of the present invention is used to store various categories of data to support the operation of the electronic terminal 700. Examples of these data include: any executable program for operating on the electronic terminal 700, such as an operating system 7021 and an application 7022; the operating system 7021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 7022 can include various applications, such as a media player (Media Player), a browser (Browser), etc., for implementing various application services. The XX method provided by the embodiment of the present invention can be included in the application 7022.

[0139] The method disclosed in the above embodiment of the present invention can be applied to the processor 701, or implemented by the processor 701. The processor 701 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 701 or the instruction in the form of software. The above processor 701 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 701 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 701 may be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0140] In an exemplary embodiment, the electronic terminal 700 may be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD) to execute the aforementioned method.

[0141] According to the method provided in the embodiment of the present application, the present application also provides a computer program product, which includes: computer program code, when the computer program code is run on a computer, it enables the computer to execute an intelligent rehearsal method based on the collaborative work of multiple intelligent agents.

[0142] According to the method provided in the embodiment of the present application, the present application also provides a computer-readable storage medium, which stores a program code. When the program code runs on a computer, the computer executes the above method.

[0143] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program and / or a computer. By way of illustration, both applications running on a computing device and a computing device can be components. One or more components may reside in a process and / or an execution thread, and a component may be located on a computer and / or distributed between two or more computers. In addition, these components may be executed from various computer-readable media having various data structures stored thereon. Components may, for example, communicate through local and / or remote processes according to signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system and / or a network, such as the Internet interacting with other systems through signals).

[0144] Those of ordinary skill in the art will appreciate that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0145] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0146] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0147] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0149] In the above embodiments, the functions of each functional unit can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When loading and executing computer program instructions (programs) on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media integrations. Available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., high-density digital video discs (DVDs), or semiconductor media (e.g., solid state disks (SSDs), etc.).

[0150] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0151] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0152] In summary, the present application proposes an intelligent rehearsal system, method, medium, device and program product based on the collaborative work of multiple intelligent agents. Different types of virtual customers are created through large models. Multiple intelligent agents collaborate to form a complete training system to solve the problem of standardization of training effects caused by manual training. The training efficiency is also greatly improved. All data generated during the training process can be preserved intact for traceability and analysis. By creating realistic and challenging dialogue scenes through large models, appropriate closing remarks can be dynamically generated according to the context of the dialogue, the emotions and needs of the customer, and the current interactive state, which improves the natural fluency and service experience of the dialogue. Virtual customers are automatically generated through large models, and users can also customize customer portraits and polish them, providing a variety of interactive rehearsal opportunities for the college. At the same time, there are also speech assistants to provide real-time support and suggestions during the dialogue to help students better understand and respond to customer needs. The entire dialogue record will also be comprehensively evaluated, feedback and improvement suggestions will be provided, and the training effect will be further improved. Therefore, this application effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0153] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. An intelligent drill system based on the collaborative work of multiple intelligent agents, characterized in that: include: Customer profile generation agent, which is used to build structured data instructions based on insurance industry customer data to guide the big model to generate customer profiles of virtual insurance customers with different personality traits; A custom portrait checking and polishing agent is used to check the custom customer portraits generated in response to user operations and polish the customer portraits that lack necessary structured information; Scenario Introduction Generative Agent, which is used to generate conversation scenarios in the insurance industry based on customer profiles; Training dialogue generation agents to output insurance business dialogue content based on simulated customer profiles and dialogue scenarios as input; The speech assistant agent is used to provide corresponding answer suggestions in the generated insurance industry dialogue scenario based on the questions raised by the virtual customer, combined with the customer portrait, scenario description and the context of the dialogue; The intelligent ending judgment agent is used to dynamically generate the corresponding conversation ending words according to the context of the conversation, the customer's emotional characteristics and demand characteristics, and the current interaction status.

2. The intelligent drill system based on the collaborative work of multiple agents according to claim 1 is characterized in that: The intelligent drill system also includes any one or both of the following intelligent agents: The customer analysis thinking agent is used to trace and analyze the customer portrait, scenario description and conversation record after the conversation is over, and learn the problem characteristics of the virtual customer. The drill evaluation agent is used to generate evaluation information for the training drill based on the dialogue records after the dialogue is completed.

3. The intelligent drill system based on the collaborative work of multiple agents according to claim 1 is characterized in that: The method for constructing the customer portrait generation intelligence includes: Collect customer data of the insurance industry and construct structured data instructions, and use the structured data instructions to preprocess and extract features of the collected customer data; the structured data instructions are used to convert the raw data into a structured format suitable for machine learning training; The extracted features are input into the GPT model for model training to construct the customer portrait generation agent.

4. The intelligent drill system based on the collaborative work of multiple agents according to claim 1 is characterized in that: The inspection contents of the customer portrait by the custom portrait inspection and polishing agent include: Completeness check is used to verify whether the customer profile input by the user contains all the necessary structured information, and generate a series of questions or prompts based on the missing information points to guide the user to supplement the missing information, and recheck the completeness of the customer profile after the user supplements the information; Personal pronoun check, used to check whether the customer portrait is expressed in the second person; Key information point check is used to check whether the customer portrait contains at least the following key information points: personal identity, relationship with the agent, personal situation, and attitude towards insurance.

5. The intelligent drill system based on the collaborative work of multiple agents according to claim 1 is characterized in that: The scenario introduces the construction method of generating an intelligent agent including: Use NLP technology to analyze the customer portrait, identify and extract key information points of the virtual customer; use machine learning algorithms to extract deep features including implicit needs and personality traits of the virtual customer from the customer portrait; input the extracted key information points and deep features of the virtual customer into the GPT model for training, so that it can understand and simulate the customer's behavior and language style; According to the characteristics of the customer portrait, one or more dialogue frameworks are constructed, and the dialogue frameworks cover the interaction paths of different business scenarios; in each dialogue framework, the trained GPT model is used to generate dialogue content that meets the customer characteristics; The user's evaluation and feedback information of the dialogue scene during the conversation is obtained, and the GPT model is optimized through multiple rounds of iterations.

6. The intelligent drill system based on the collaborative work of multiple agents according to claim 1, characterized in that: The method for constructing the training dialogue generating intelligent agent includes: The customer portrait generation agent receives the customer portrait of the virtual customer, the customized customer portrait after polishing is received from the customized portrait checking and polishing agent, and the simulated conversation scenario content is received from the scenario introduction generation agent, and the received data is preprocessed and feature extracted; The GPT model understands the customer's intentions and needs based on the input features and contextual information. It uses the language patterns and structures of the insurance industry learned during training to construct and output dialogue content that is consistent with the role and scenario. The language patterns and structures of the insurance industry include: the use of professional terms, inquiry and question-and-answer patterns, risk assessment and recommendations, claims process instructions, compliance and standardization, and customer service language.

7. An intelligent drill method based on the collaborative work of multiple intelligent agents, characterized in that: Based on the GPT model, multiple agents are built to work together to perform the following: Construct structured data instructions based on insurance industry customer data to guide the big model to generate customer portraits of virtual insurance customers with different personality characteristics; the customer data is simulated customer data based on real data structure and is desensitized; Check customized customer profiles generated in response to user actions and polish those that are missing necessary structured information; Generate conversation scenarios for the insurance industry based on customer portraits; Based on the simulated customer portrait and conversation scenario as input, output the insurance business conversation content; In the generated insurance industry dialogue scenario, corresponding answer suggestions are provided in real time based on the questions raised by the virtual customer, combined with the customer portrait, scenario description and the context of the dialogue; The corresponding conversation ending remarks are dynamically generated based on the conversation context, customer emotional characteristics and demand characteristics, and the current interaction status.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intelligent rehearsal method based on the collaborative work of multiple intelligent agents as described in claim 7 is implemented.

9. A computer program product, characterized in that The computer program product includes computer program code, and when the computer program code is executed on a computer, the computer implements the intelligent rehearsal method based on the collaborative work of multiple intelligent agents as described in claim 7.

10. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the intelligent rehearsal method based on the collaborative work of multiple intelligent agents as described in claim 7.

Images