Customer simulation system, method, device, storage medium and program product

Abstract

The embodiment of the invention provides a customer simulation system, method and device, a storage medium and a program product. The system comprises a customer agent and a customer service agent. Wherein the customer agent is used for generating a first cue word in combination with historical dialogue content generated in a multi-round dialogue process with the customer service agent, inputting the first cue word into a first large language model so as to enable the first large language model to output a reference question in a current round of dialogue process, and sending the reference question to the customer service agent; and the customer service agent is used for generating a second cue word in combination with the historical dialogue content, the reference question and the product description information of the target product, inputting the second cue word into a second large language model to enable the second large language model to output a reference answer corresponding to the reference question, and sending the reference answer to the customer agent. According to the scheme, real experienced customer service personnel do not need to participate, the labor cost is low, a verbal skill library does not need to be used in the whole working process, and the flexibility is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a customer simulation system, method, device, storage medium and program product. BACKGROUND

[0002] At present, a traditional customer service personnel needs to undergo a series of training before formally taking up a post, and can be a real customer service personnel only after passing a relevant certification examination, and then serves customers. In the process of detecting the ability of a new customer service personnel, the prior art generally pre-establishes a dialogue library, which stores a plurality of sets of dialogue information of customers and customer service personnel. In specific implementation, an experienced customer service personnel plays a customer (for ease of description, hereinafter referred to as a simulation customer), the simulation customer asks questions to a novice customer service personnel based on the dialogue library, and observes the performance of the novice customer service personnel, i.e., whether the reply of the novice customer service personnel accurately locates and solves the question raised by the simulation customer. According to the performance of the novice customer service personnel, it can be judged that the service level of the novice customer service personnel, and it can be determined whether the novice customer service personnel has the ability to truly serve customers.

[0003] The present application relates to the technical field of artificial intelligence, and in particular to a customer simulation system, method, device, storage medium and program product.

[0004] 1. An experienced customer service personnel is needed to play a role, and the labor cost is high.

[0005] 2. The dialogue library is used, and the flexibility is poor. Specifically, with the continuous development of science and technology, the industry information of various industries updates and changes at a high speed, and the dialogue library cannot adapt to the update speed of the industry information. SUMMARY

[0006] The customer simulation system, method, device, storage medium and program product provided by the embodiments of the present application do not need a real experienced customer service personnel to participate, the labor cost is low, and the dialogue library does not need to be used in the whole working process, and the flexibility is high.

[0007] In a first aspect, the embodiments of the present application provide a customer simulation system, which comprises a customer intelligent agent for simulating a target type of customer, and a customer service intelligent agent for simulating a customer service personnel.

[0008] The first large language model is included in the customer intelligent agent, and the second large language model is included in the customer service intelligent agent.

[0009] The customer agent is configured to generate a first prompt word in combination with historical dialogue content generated in a multi-round dialogue process with the customer service agent, input the first prompt word into the first large language model, so that the first large language model outputs a reference question in a current round of dialogue process, and send the reference question to the customer service agent, wherein the first prompt word includes customer feature information of the target type customer and first dialogue purpose information.

[0010] The customer service agent is configured to generate a second prompt word in combination with the historical dialogue content, the reference question and product description information of the target product, input the second prompt word into the second large language model, so that the second large language model outputs a reference answer corresponding to the reference question, and send the reference answer to the customer agent, wherein the second prompt word includes second dialogue purpose information, and the second dialogue purpose information and the first dialogue purpose information form an antagonistic relationship.

[0011] In a second aspect, an embodiment of the present application provides a customer simulation method, which is applied to a customer agent for simulating a target type customer generated based on customer feature information of the target type customer, and the customer agent includes a first large language model. The method comprises:

[0012] A first prompt word is generated in combination with historical dialogue content generated in a multi-round dialogue process with a customer service agent, wherein the first prompt word includes first dialogue purpose information corresponding to a target product and customer feature information of the target type customer, and the customer service agent includes a second large language model;

[0013] The first prompt word is input into the first large language model, so that the first large language model outputs a reference question in a current round of dialogue process;

[0014] The reference question is sent to the customer service agent, so that the customer service agent generates a second prompt word in combination with the historical dialogue content, the reference question and product description information of the target product, and inputs the second prompt word into the second large language model, so that the second large language model outputs a reference answer corresponding to the reference question, and the reference answer is sent to the customer agent, wherein the second prompt word includes second dialogue purpose information, and the second dialogue purpose information and the first dialogue purpose information form an antagonistic relationship.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, which comprises a memory, a processor, a communication interface, wherein the memory stores executable code, and when the executable code is executed by the processor, the processor executes the video transcoding method in the first aspect.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a communication interface; wherein the memory stores executable code, and when the executable code is executed by the processor, the processor performs the customer simulation method according to the second aspect.

[0017] In a fourth aspect, an embodiment of the present application provides a non-transitory machine readable storage medium, which stores executable code, and when the executable code is executed by a processor of an electronic device, the processor performs the customer simulation method according to the second aspect.

[0018] In a fifth aspect, an embodiment of the present application provides a computer program product, which comprises a computer program, and when the computer program is executed by a processor of an electronic device, the processor performs the customer simulation method according to the second aspect.

[0019] The customer simulation system provided by the embodiment of the present application combines the customer agent for simulating the target type customer with the customer service agent for simulating the customer service, performs multi-round dialogue between the customer agent and the customer service agent, and makes the two agents constantly confront each other in the dialogue process, so as to make the customer agent constantly ask more questions about the target product, thereby helping the customer service agent constantly improve the service ability.

[0020] In the confrontation between the customer agent and the customer service agent in each round of dialogue, the first dialogue purpose information in the first prompt word corresponding to the first large language model in the customer agent and the second dialogue purpose information in the second prompt word corresponding to the second large language model in the customer service agent are mainly used. It should be understood that, since the second dialogue purpose information and the first dialogue purpose information form an antagonistic relationship, that is, the purpose of the customer agent is to constantly ask questions and try not to easily agree to purchase the target product, and constantly challenge the answers of the customer service agent, while the purpose of the customer service agent is to constantly provide rich information of the target product and promote the purchase. Under the guidance of this antagonistic relationship, the customer agent can constantly ask more questions about the target product.

[0021] Based on the above customer simulation system, when it is necessary to detect the ability of a new customer service personnel for a target product, the customer simulation system can set different customer feature information according to the demand, simulate different types of customer agents, and through the automatic multi-round interaction with the customer service agent, the simulated customer agent has a realistic human simulation effect, so that in the subsequent ability detection of the new customer service personnel, the customer agent and the customer service personnel can automatically dialogue, and the answer of the customer service personnel can be observed, without the participation of a real experienced customer service personnel, only the trained customer agent is used, the labor cost is low, and the flexibility is high. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1 A structural schematic diagram of a customer simulation system provided by the embodiment of the present application is provided.

[0024] Figure 2 An application schematic diagram of a customer agent provided by the embodiment of the present application is provided.

[0025] Figure 3 An application schematic diagram of a customer service agent provided by the embodiment of the present application is provided.

[0026] Figure 4 Another structural schematic diagram of a customer simulation system provided by the embodiment of the present application is provided.

[0027] Figure 5 A flowchart of a customer simulation method in a customer simulation system provided by the embodiment of the present application is provided.

[0028] Figure 6 A structural schematic diagram of an electronic device provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0030] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in the embodiments of the present invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0031] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Where there is no conflict between the embodiments, the following embodiments and features can be combined with each other. Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0032] First, the terms or concepts involved in the embodiments of this invention will be explained:

[0033] AI Agent: A smart agent system with a large language model as its core controller. Essentially, it is an agent system that controls the large language model to solve user problems. It relies on the large language model as its core decision-making and processing unit and can use the logical reasoning ability of the large language model to decompose the solution steps of the user problem, call tools, and summarize the solution results of multiple steps.

[0034] Large language models refer to deep learning models trained on massive amounts of text data, possessing powerful expressive and generalization capabilities. They can not only generate natural language text but also deeply understand the meaning of text, handling various natural language tasks such as text summarization, question answering, and translation.

[0035] Currently, the process of assessing the capabilities of new customer service representatives involves experienced customer service personnel posing as customers (hereinafter referred to as simulated customers for ease of description). The simulated customers ask questions to the new customer service representatives based on a set of scripts, and the new representatives' performance is observed—specifically, whether their responses accurately pinpoint and resolve the simulated customer's issues. Based on this performance, the new customer service representative's service level is judged, determining whether they possess the ability to truly serve customers. However, this method has at least the following problems:

[0036] 1. It requires experienced customer service personnel to perform role-playing, resulting in high labor costs.

[0037] 2. Using a script database lacks flexibility. Specifically, with the continuous development of technology, industry information in various sectors is updated and changed rapidly, and using a script database cannot adapt to the pace of information updates.

[0038] In view of this, embodiments of the present invention provide a customer simulation system that solves the above problems through the following approach: A customer intelligence agent simulating a target type of customer is combined with a customer service intelligence agent simulating customer service. By engaging in multi-round dialogues and continuous interaction during these dialogues, the customer intelligence agent is encouraged to ask more questions about the target product, thereby helping the customer service intelligence agent continuously improve its service capabilities. Based on this, when the customer simulation system is needed to assess the capabilities of new customer service representatives for a target product, the system can be configured with different customer characteristic information to simulate different types of customer intelligence agents to assess their capabilities. Specifically, when assessing the capabilities of new customer service representatives, the customer intelligence agent can automatically converse with the customer service personnel, observing the personnel's responses. This eliminates the need for experienced customer service personnel; only a trained customer intelligence agent is required, resulting in lower labor costs. Furthermore, the system does not require a script library throughout the process, offering high flexibility.

[0039] Figure 1 A schematic diagram of the structure of a customer simulation system provided in an embodiment of the present invention is shown below. Figure 1 As shown, the system includes: a customer agent 10 for simulating a target type of customer, and a customer service agent 20 for simulating customer service. Customer agent 10 includes a first large language model, and customer service agent 20 includes a second large language model. Customer agent 10 generates a first prompt word by combining historical dialogue content generated during multiple rounds of dialogue with customer service agent 20. The first prompt word is input into the first large language model, causing the first large language model to output a reference question for the current round of dialogue, which is then sent to customer service agent 20. The first prompt word includes customer characteristic information of the target type of customer corresponding to the target product and information about the first dialogue purpose. Customer service agent 20 generates a second prompt word by combining historical dialogue content, the reference question, and product description information of the target product. The second prompt word is input into the second large language model, causing the second large language model to output a reference answer corresponding to the reference question, which is then sent to customer agent 10. The second prompt word includes information about the second dialogue purpose, which forms an adversarial relationship with the first dialogue purpose.

[0040] In practical applications, in order to make the simulated customer agent 10 more realistic and human-like, the customer agent 10 and the customer service agent 20 can automatically engage in multi-round dialogues. It should be noted that in the initial state, the historical dialogue content between the customer agent 10 and the customer service agent 20 is empty. Afterwards, as the number of dialogue rounds and the content of the dialogue between the two continue to increase, the historical dialogue content will be updated accordingly.

[0041] For ease of understanding, the following takes the target product as an insurance product as an example to illustrate the content of the dialogue between the customer agent 10 and the customer service agent 20. It is assumed that the customer agent 10 and the customer service agent 20 have carried out 5 rounds of dialogue since the initial state, and the specific dialogue content of the 5 rounds of dialogue is as follows:

[0042] Customer agent 10: Do you have any insurance product recommendations?

[0043] Customer service agent 20: This insurance helps you save money while being protected, such as illness, accidental injury, etc., and can also receive money in the future, and is worry-free for old age.

[0044] Customer agent 10: It sounds good. How much do you have to pay each year? Is there a minimum investment amount?

[0045] Customer service agent 20: It is very flexible, with a minimum investment of several thousand yuan per year, and you can plan according to your needs and invest a little every month.

[0046] Customer agent 10: If you need money in the middle, can you take it out, and will you be charged a lot of fees?

[0047] Customer service agent 20: You can take out part of it in advance, but there may be some fees or loss of part of the income, but it can still help you in an emergency.

[0048] Customer agent 10: I see. If something happens to me, how is the compensation amount calculated? Are there any restrictions?

[0049] Customer service agent 20: The compensation is based on the amount of insurance you buy, and the amount of compensation is paid according to the contract, and different types of accidents may have different restrictions and waiting periods, and we will explain in detail to ensure that you understand.

[0050] Customer agent 10: OK, what important terms are included in the contract, and what should I pay special attention to?

[0051] Customer service agent 20: The contract will specify the scope of protection, exclusion clauses, payment period, and claim settlement process, and pay special attention to the start time of the contract and the loss of withdrawal, and don't forget to fill in the health notice truthfully.

[0052] It should be noted that during the dialogue between the customer agent 10 and the customer service agent 20, either party can initiate the dialogue. In the above example, the customer agent 10 initiates the dialogue first, but this is not limited to this, and the customer service agent 20 can also initiate the dialogue first.

[0053] After the above-mentioned 5 rounds of conversation between the customer agent 10 and the customer service agent 20, the 6th round of conversation between the two can be regarded as the current conversation turn. During the conversation process of the current conversation turn, the first prompt word can be generated based on the customer characteristic information of the target type customer corresponding to the target product and the first conversation purpose information in combination with the content of the above-mentioned 5 rounds of conversation, and the first prompt word is input into the first large language model to make the first large language model output the reference question in the current round of conversation process. It should be understood that the reference question is mainly generated based on the first prompt word, so that the first large language model in the embodiment of the application not only considers the historical conversation content of the customer agent 10 and the customer service agent 20, but also comprehensively considers the customer characteristic information of the target type customer and the first conversation purpose information in the process of outputting the reference question based on the first prompt word, which can more realistically simulate a real customer.

[0054] The customer characteristic information of the target type customer can be: “works as an engineer, has a family, has a two-year-old baby, has limited income, and will have pressure to pay the premium”. The first conversation purpose information can be: “describe the confusion about the insurance product; ask more questions about the answers of the customer service agent 20”. Based on this, the first prompt word generated in combination with the customer characteristic information of the target type customer and the first conversation purpose information is input into the first large language model, and the reference question obtained can be “is there any additional service or welfare”.

[0055] After obtaining the reference question, the reference question can be sent to the customer service agent 20. Then, the customer service agent 20 generates the second prompt word in combination with the reference question and the above-mentioned historical conversation content, and the product description of the target product, and inputs the second prompt word into the second large language model to make the second large language model output the reference answer corresponding to the reference question. It should be understood that the reference answer is mainly generated based on the second prompt word, so that the customer service agent 20 in the embodiment of the application comprehensively considers the historical conversation content of the customer agent 10 and the customer service agent 20, and the product description of the target product in the process of outputting the reference question based on the second prompt word, which guarantees the comprehensiveness and accuracy of the reference answer and ensures that the customer service agent 20 has high service ability.

[0056] The second prompt word further includes the second conversation purpose information, and the second conversation purpose information forms an antagonistic relationship with the first conversation purpose information, that is, the purpose of the customer agent 10 is to constantly ask questions and try not to easily agree to purchase the target product, and constantly challenge the answers of the customer service agent, while the purpose of the customer service agent 20 is to constantly provide rich information of the target product to promote the purchase. Under the guidance of the antagonistic relationship, the customer agent 10 can constantly ask more diverse questions about the target product.

[0057] Further, the product description of the target product can be product information of an insurance product, and the product information records all contents related to the insurance product. The second dialogue purpose information can be "to make the customer understand and consider the recommended insurance product". Based on this, the second prompt word generated by combining the reference question, the historical dialogue content, the second dialogue purpose information, and the product description of the target product is input into the second large language model, and the reference answer corresponding to the reference question obtained can be "yes, the insurance product is attached with a regular cashback benefit".

[0058] After obtaining the reference answer, the reference answer is sent to the customer agent 10, that is, the current round of dialogue between the customer agent 10 and the customer service agent 20 is completed, and then the above operation can be repeated to continue the dialogue between the customer agent 10 and the customer service agent 20 until the dialogue ends. The trigger condition for ending the dialogue between the customer agent 10 and the customer service agent 20 can be that the dialogue round between the customer agent 10 and the customer service agent 20 reaches a set threshold (such as the dialogue round reaching 20 times), or one of the customer agent 10 and the customer service agent 20 appears an ending word such as "bye" and "goodbye", and the like, which are not listed here.

[0059] The customer simulation system provided by the embodiment of the application can continuously dialogue between the customer agent 10 and the customer service agent 20, and continuously dialogue between the customer agent 10 and the customer service agent 20 based on the first dialogue purpose information in the first prompt word and the second dialogue purpose information in the second prompt word in the process of multiple rounds of dialogue between the customer agent 10 and the customer service agent 20, so as to make the customer agent 10 continuously ask more diverse questions about the target product, thereby helping the customer service agent 20 continuously improve the service ability.

[0060] When using a customer simulation system to assess the capabilities of new customer service representatives for a target product, the system can be customized to simulate specific customer characteristics based on requirements. This allows for the creation of a tailored customer agent to simulate the corresponding customer type. During the subsequent capability assessment, the system automatically interacts with the customer service representative, observing their responses to automate the assessment. Each response from the new representative can be scored. If the response reaches a set score (e.g., 8 out of 10), the new representative is deemed qualified and can then serve real customers. Conversely, if the response does not reach the set score, the assessment is considered unqualified. Therefore, the customer simulation system provided in this embodiment can simulate real customers during the capability assessment of new customer service representatives, eliminating the need for experienced customer service personnel. It utilizes only a trained customer agent, resulting in lower labor costs and greater flexibility as it does not require a script library throughout the process.

[0061] The structure and application process of Customer Intelligence Agent 10 and Customer Service Intelligence Agent 20 are explained in detail below:

[0062] Figure 2 This is an application diagram of the client intelligent agent 10 provided in an embodiment of the present invention, such as... Figure 2 As shown, the customer intelligent agent 10 includes: a customer feature module 101, a reasoning and response module 102, a long-term memory module 103, and a short-term memory module 104.

[0063] The customer feature module 101 is used to obtain customer feature information and send it to the inference response module 102. Customer feature information may include, for example, the customer's name, gender, age, speaking style, occupation, family situation, hobbies, knowledge of the target product, social circle, etc., and will not be listed here.

[0064] The reasoning response module 102 is used to issue a reference question for the current dialogue round to the customer service agent 20 based on the customer characteristic information and the historical dialogue content between the customer agent 10 and the customer service agent 20. For example, in the previous dialogue round, the customer service agent 20 asked, "Hello, do you have any questions about the insurance I recommended to you before?" Then, the reasoning response module 102 could issue a reference question for the current dialogue round to the customer service agent 20, such as, "Oh, I'm not interested in the annuity insurance from last time. Do you have any recommendations for family-related insurance?" It should be noted that the historical dialogue content used by the reasoning response module 102 in this process is obtained from the long-term memory module 103 and the short-term memory module 104.

[0065] Specifically, the historical dialogue content includes the first historical dialogue content generated in a first time period and the second historical dialogue content generated in a second time period. The first time period is earlier than the second time period, and the data volume corresponding to the second historical dialogue content is a set value. The first historical dialogue content generated in the first time period is the content stored in the long-term memory module 103, while the second historical dialogue content generated in the second time period is the content stored in the short-term memory module 104. In practical applications, assuming that the data volume of the 20 rounds of dialogue content before the current dialogue round between customer agent 10 and customer service agent 20 reaches the set value, then the 20 rounds of dialogue content before customer agent 10 and customer service agent 20 can be considered as the second historical dialogue content, while the dialogue content before these 20 rounds can be considered as the first historical dialogue content (such as the dialogue content from round 20 to round 60 between customer agent 10 and customer service agent 20).

[0066] In practice, the long-term memory module 103 can be used to extract summary information from the first historical dialogue content to obtain the summary information corresponding to the first historical dialogue content. It should be understood that since the data volume of the first historical dialogue content is large, reading it in its entirety would consume a lot of time. Therefore, by extracting summary information from the first historical dialogue content, it is easier for the first language model to quickly analyze the first historical dialogue content when outputting the reference question, thereby improving work efficiency.

[0067] Furthermore, after obtaining the summary information, the first prompt word can be generated by combining the summary information with the second historical dialogue content. This ensures that the subsequent first language model can quickly and completely combine the historical dialogue information to output the reference question, thereby improving work efficiency while ensuring the authenticity of the output reference question.

[0068] To better understand the content of the first prompt word in the input language model, the following is a detailed example: "Let's play a game. You are a customer interested in financial management. You need to remember your role and personal experience. You need to use your skills to ask as many questions as possible about this insurance. You can only use simple, colloquial sentences to express your wishes. Now you can start communicating with the insurance sales agent. Please do not use formal language, use colloquialisms, or even verbal slang."

[0069] In this scenario, you will be an insurance purchaser. You will need to clearly and concisely express your problems and needs to customer service, and evaluate the solutions they provide. Your personal background includes: limited income, financial pressure to pay premiums, an engineer job, a family, and a two-year-old child. Your skills may include: 1. Clearly and concisely describing your problems and needs. 2. Asking follow-up questions when customer service's questions are unclear. 3. Asking counter-questions when customer service's questions are unclear.

[0070] It should be noted that in the above example, "role," "personal experience," and "skills" can all be replaced according to the actual situation to obtain the desired first prompt word.

[0071] Figure 3 This is an application diagram of the customer service intelligent agent 20 provided in an embodiment of the present invention, such as... Figure 3 As shown, the customer service agent 20 includes: a reasoning and response module 201, a long-term memory module 202, and a short-term memory module 203.

[0072] The reasoning response module 201 is used to send a reference answer for the current dialogue round to the customer agent 10 based on the reference question output by the customer agent 10, the product description information of the target product, and the historical dialogue content between the customer agent 10 and the customer service agent 20. For example, in the current dialogue round between the customer agent 10 and the customer service agent 20, if the reference question output by the customer agent 10 is "Oh, I wasn't interested in the annuity insurance last time. Do you have any family-related insurance recommendations?", then the reasoning response module 201 could send a reference answer for the current dialogue round to the customer agent 10 such as "Yes, we recently have an insurance product that is particularly suitable for children under 6 years old." It should be noted that the historical dialogue content used by the reasoning response module 201 in this process is obtained from the long-term memory module 202 and the short-term memory module 203.

[0073] Specifically, the historical dialogue content includes the first historical dialogue content generated in a first time period and the second historical dialogue content generated in a second time period. The first time period is earlier than the second time period, and the data volume corresponding to the second historical dialogue content is a set value. The first historical dialogue content generated in the first time period is the content stored in the long-term memory module 202, while the second historical dialogue content generated in the second time period is the content stored in the short-term memory module 203. In practical applications, assuming that the data volume of the 20 rounds of dialogue content before the current dialogue round between customer agent 10 and customer service agent 20 reaches the set value, then the previous 20 rounds of dialogue content between customer agent 10 and customer service agent 20 can be considered as the second historical dialogue content, while the dialogue content before these 20 rounds can be considered as the first historical dialogue content (such as the dialogue content from round 20 to round 60 between customer agent 10 and customer service agent 20).

[0074] In practice, the long-term memory module 202 can be used to extract summary information from the first historical dialogue content to obtain the summary information corresponding to the first historical dialogue content. It should be understood that since the data volume of the first historical dialogue content is large, reading it in its entirety would consume a lot of time. Therefore, by extracting summary information from the first historical dialogue content, it is easier for the first language model to analyze the first historical dialogue content during the output of the reference question, thereby improving work efficiency.

[0075] Furthermore, after obtaining the summary information, a second prompt word can be generated by combining the summary information, the second historical dialogue content, the reference question, and the product description information of the target product. This ensures that the subsequent second language model can quickly and completely combine the historical dialogue information to output the reference answer corresponding to the reference question during the output of the reference answer, thereby improving work efficiency while ensuring the authenticity of the output reference answer.

[0076] To better understand the specific content of the second prompt word in the second major language model, the following example illustrates the second prompt word: "Let's play a game. You are an insurance sales expert. You need to remember your role and personal experience. You must use your skills to get the customer to understand and consider purchasing the recommended financial insurance. You can only use one sentence to directly respond to the customer's mention of financial insurance. Now, start communicating with the customer's AI agent. Note that your workflow is: First, quickly understand the customer's current focus. Then, you must retrieve product information from the financial insurance product document library. Next, after understanding the product information, answer the customer's questions in conversational language, no more than one sentence. Finally, get the customer to understand and consider purchasing the recommended financial insurance."

[0077] In this role, you are an insurance sales specialist. You will need to answer customer questions based on a database of financial insurance product documentation (used to store information about financial insurance products). Your skills may include: 1. Answering questions based on the financial insurance product documentation database. 2. Responding to customer inquiries in conversational language. 3. Quickly identifying customer needs through brief conversations and customizing insurance recommendations.

[0078] It should be noted that in the above example, "role", "financial insurance product document library", and "skill" can all be replaced according to the actual situation to obtain the required second prompt word.

[0079] Figure 4 This is another structural schematic diagram of a customer simulation system provided in an embodiment of the present invention, such as... Figure 4 As shown, the system also includes a customer service response review agent 30 and a coach agent 40.

[0080] The customer service response review agent 30 includes a third language model. This agent generates third prompt words based on predefined rules and multiple reference answers, then inputs these prompt words into the third language model to determine whether the reference answers conform to the predefined rules. These reference answers are output by the customer service agent 20 during multi-turn dialogues with the customer agent 10.

[0081] The coach agent 40 includes a fourth language model. The coach agent 40 combines historical dialogue content and a list of question types to generate a fourth prompt word. This fourth prompt word is then input into the fourth language model, enabling it to select a target question type from the question type list. Based on the historical dialogue content and the target question type, the coach agent 40 generates a question suggestion, which is then sent to the client agent 10. The client agent 10 then outputs a reference question for the current round of dialogue. The question suggested in the question suggestion differs from the reference questions generated by the client agent 10 in the historical dialogue content.

[0082] In practical applications, it should be understood that the multiple reference answers output by customer service agent 20 during multi-round dialogues with customer agent 10 may not all conform to the set rules, and reference answers that do not conform to the set rules cannot be used. Therefore, this embodiment of the invention adds a customer service answer review agent 30 to the customer simulation system to determine whether the reference answers output by customer service agent 20 conform to the set rules. The result of this determination can be fed back to customer service agent 20 to optimize subsequent interaction outputs.

[0083] In practice, the customer service response review agent 30 generates a third prompt word based on set rules and multiple reference answers output by the customer service agent 20 during multi-round dialogues with the customer agent 10. This third prompt word is then input into a third language model, which determines whether the multiple reference answers conform to the set rules and outputs the response review result. For ease of understanding, let's assume the set rule is "cannot contain word A, cannot contain phrase B." As an optional approach, word A and phrase B can be sensitive terms within the target product industry, but are not limited to these. Then, after inputting the third prompt word based on this set rule into the third language model, the model can determine whether the multiple reference answers output by the customer service agent 20 during multi-round dialogues with the customer agent 10 contain word A and / or phrase B. If none of the multiple reference answers contain the specified word A and / or phrase B, then the third language model can output "The reference answer conforms to the rules". If the target reference answer contains word A and / or phrase B among the multiple reference answers, then the target reference answer does not conform to the specified rules. In this case, the third language model can output "The target reference answer does not conform to the rules".

[0084] Furthermore, the customer simulation system provided in this embodiment of the invention also includes a coach agent 40, which is essentially designed to adjust the difficulty of the first language model output problem.

[0085] Specifically, the coach agent 40 has a list of question types. The fourth language model contained in the coach agent 40 can select the target question type from the question type list based on the fourth prompt word, generate a question suggestion based on the historical dialogue content and the target question type, and send the question suggestion to the client agent 10. This allows the client agent 10 to refer to the historical dialogue content and the question suggestion when outputting the reference question in the current round of dialogue, thus ensuring the quality of the reference question.

[0086] To facilitate understanding, for example, the question type list could include: insurance product introduction, return budget, fee structure, coverage, financial flexibility, historical performance, market comparison, customer service, etc. The fourth prompt could be: "Please read the conversation between the buyer and seller and randomly select three question suggestions from the question type list to improve the insurance buyer's questioning skills. However, do not allow the insurance buyer to repeatedly ask the same type of question; only return three question suggestions and do not return other content."

[0087] In practice, assuming the fourth language model randomly generates three question suggestions related to "financial flexibility, historical performance, and customer service," after sending these three suggestions to the client agent 10, the client agent 10 will combine the historical dialogue content and the question suggestions to generate a first prompt word. Specifically, the client agent 10 will place the three question suggestions related to "financial flexibility, historical performance, and customer service" after the historical dialogue content to generate the first prompt word. In this way, the client agent 10 can refer to the historical dialogue content and question suggestions when outputting reference questions in the current round of dialogue, and can also avoid the question suggestion from repeating the reference questions in the historical dialogue content.

[0088] Below are examples illustrating three questions related to "financial flexibility, historical performance, and customer service":

[0089] 1. A suggestion for asking questions related to financial flexibility could be: "If I need a large sum of money urgently in the future, such as for my child's education or buying a house, can this insurance be used as collateral for a loan?"

[0090] 2. Suggested questions related to historical performance could be: "How has the insurance company performed in terms of investment? Is there any historical data you can share so I can compare and see the results of past investments?"

[0091] 3. Customer service-related questions and suggestions could include: "If I am not satisfied with the service or a better product becomes available, what losses or limitations will I face if I switch to other insurance policies?"

[0092] Based on the above, by using the coach agent 40, the primary language model in the customer agent 10 can be stimulated to output more challenging questions for the target product, thereby improving the questioning ability of the customer agent 10. This, in turn, helps the customer service agent 20 to continuously improve its service capabilities, ultimately resulting in a customer simulation system that can adapt to different question difficulties, has a more realistic and human-like effect, and a more powerful service capability.

[0093] Furthermore, in embodiments of the present invention, such as Figure 4 As shown, the system also includes a retrieval module 50. This retrieval module 50 maintains product description information for the target product. The customer service agent 20 sends a reference question to the retrieval module 50 and receives reference product description information corresponding to the reference question from the retrieval module 50. It then generates a second prompt by combining historical dialogue content, the reference question, and the reference product description information. The retrieval module 50 retrieves a portion of the product description information matching the reference question from the target product's product description information and generates reference product description information based on this portion.

[0094] In practice, after receiving the reference question for the current round from the customer agent 10, the customer service agent 20 can send the reference question to the retrieval module 50. The retrieval module 50 retrieves a portion of the product description information matching the reference question from the product description information of the target product, and generates reference product description information based on the sub-product description information. This reference product description information is used as a reference, and combined with historical dialogue content and the reference question to generate a second prompt word. Then, the second prompt word is input into the second language model to obtain a reference answer corresponding to the reference question. Specifically, the retrieval module 50 can be implemented as a Retrieval-augmented Generation (RAG) model.

[0095] By using the retrieval module 50, relevant product description information can be retrieved from the product description information of a target product with a large data volume. Reference product description information can then be generated based on this sub-product description information, improving work efficiency and the accuracy of the final reference answer. Furthermore, using the retrieval module 50 can enhance the customer service agent 20's understanding of the target product and strengthen its dialogue capabilities with the customer agent 10. For example, when a customer needs detailed cost information for a target product, the retrieval module 50 can search the product description information and calculate the corresponding cost based on the search results to meet the customer's needs.

[0096] The following will describe in detail a customer simulation method provided by an embodiment of the present invention. Figure 5 A flowchart of a customer simulation method in a customer simulation system provided by an embodiment of the present invention. This method is applied to a customer agent generated based on customer characteristic information of a target type customer for simulating that target type customer. The customer agent includes a first major language model, such as... Figure 5 As shown, the method includes the following steps:

[0097] 501. Based on the historical dialogue content generated during the multi-turn dialogue with the customer service agent, generate the first prompt word. The first prompt word includes the first dialogue purpose information corresponding to the target product and the customer characteristic information of the target type customer. The customer service agent includes the second language model.

[0098] 502. Input the first prompt word into the first language model so that the first language model can output the reference question in the current round of dialogue.

[0099] 503. Send the reference question to the customer service agent so that the customer service agent can generate a second prompt word by combining the historical dialogue content, the reference question and the product description information of the target product. Then, input the second prompt word into the second language model so that the second language model can output a reference answer corresponding to the reference question and send the reference answer to the customer service agent. The second prompt word includes the second dialogue purpose information, which forms an adversarial relationship with the first dialogue purpose information.

[0100] Figure 5 The method shown can perform the steps in the foregoing embodiments. For detailed execution process and technical effects, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0101] This invention provides an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 61, a memory 62, and a communication interface 63. The memory 62 stores executable code, which, when executed by the processor 61, enables the processor 61 to at least implement the client simulation method provided in the foregoing embodiments.

[0102] In addition, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the client simulation method provided in the foregoing embodiments.

[0103] This invention provides a computer program product comprising: a computer program that, when executed by a processor of an electronic device, enables the processor to at least implement the client simulation method provided in the foregoing embodiments.

[0104] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and application programs. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. The present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A client simulation system, characterized by, Comprise: A customer agent for simulating a target type of customer, and a customer service agent for simulating a customer service; The first large language model is included in the customer agent, and the second large language model is included in the customer service agent; The customer agent is used to generate a first prompt word in combination with historical dialogue content generated in a multi-round dialogue process with the customer service agent, input the first prompt word into the first large language model, so that the first large language model outputs a reference question in the current round of dialogue process, and sends the reference question to the customer service agent, wherein the first prompt word includes customer feature information and first dialogue purpose information of the target type of customer; The customer service agent is used to generate a second prompt word in combination with the historical dialogue content, the reference question and product description information of the target product, input the second prompt word into the second large language model, so that the second large language model outputs a reference answer corresponding to the reference question, and send the reference answer to the customer agent, wherein the second prompt word includes second dialogue purpose information, and the second dialogue purpose information and the first dialogue purpose information form an antagonistic relationship.

2. The system of claim 1, wherein, Further comprise: A customer service answer auditing agent, the third large language model is included in the customer service answer auditing agent; The customer service answer auditing agent is used to generate a third prompt word according to a set rule and a plurality of reference answers, input the third prompt word into the third large language model, so that the third large language model determines whether the plurality of reference answers conform to the set rule, wherein the plurality of reference answers are output by the customer service agent in a multi-round dialogue process with the customer agent.

3. The system of claim 1, wherein, Further comprise: A coach agent, the fourth large language model is included in the coach agent; The coach agent is used to generate a fourth prompt word in combination with the historical dialogue content and a question type list, input the fourth prompt word into the fourth large language model, so that the fourth large language model selects a target question type from the question type list, and generates a questioning suggestion according to the historical dialogue content and the target question type, and sends the questioning suggestion to the customer agent, so that the customer agent outputs a reference question in the current round of dialogue process, wherein the question indicated in the questioning suggestion is different from each reference question generated by the customer agent in the historical dialogue content.

4. The system of claim 3, wherein, The customer agent is used to generate the first prompt word in combination with the historical dialogue content and the questioning suggestion.

5. The system of any one of claims 1-4, wherein, Further comprise: A retrieval module, the product description information of the target product is maintained in the retrieval module; The customer service agent is used to send the reference question to the retrieval module, receive reference product description information corresponding to the reference question fed back by the retrieval module, and generate the second prompt word in combination with the historical dialogue content, the reference question and the reference product description information; The retrieval module is configured to search for part of product description information matching the reference question from product description information of the target product, and generate the reference product description information according to the part of product description information.

6. The system of any one of claims 1-4, wherein, The historical dialogue content includes first historical dialogue content generated in a first time period and second historical dialogue content generated in a second time period, the first time period is earlier than the second time period, and a data amount corresponding to the second historical dialogue content is a set value. The customer agent is further configured to extract summary information from the first historical dialogue content to obtain summary information corresponding to the first historical dialogue content, store the summary information and the second historical dialogue content, and generate the first prompt word in combination with the summary information and the second historical dialogue content.

7. The system of claim 6, wherein, The customer agent is further configured to extract summary information from the first historical dialogue content to obtain summary information corresponding to the first historical dialogue content, store the summary information and the second historical dialogue content, and generate the second prompt word in combination with the summary information, the second historical dialogue content, the reference question and product description information of the target product.

8. A client simulation method, characterized by, A customer agent for simulating a target type customer is applied to generation of customer feature information of the target type customer, the customer agent includes a first large language model, and the method includes: In combination with historical dialogue content that has been generated in a multi-round dialogue process of a customer agent, a first prompt word is generated, wherein the first prompt word includes first dialogue purpose information corresponding to a target product and customer feature information of the target type customer, the customer agent includes a second large language model; The first prompt word is input into the first large language model, so that the first large language model outputs a reference question in a current round of dialogue process; The reference question is sent to the customer agent, so that the customer agent generates a second prompt word in combination with the historical dialogue content, the reference question and product description information of the target product, and the second prompt word is input into the second large language model, so that the second large language model outputs a reference answer corresponding to the reference question, and the reference answer is sent to the customer agent, wherein the second prompt word includes second dialogue purpose information, and the second dialogue purpose information and the first dialogue purpose information form an antagonistic relationship.

9. An electronic device, comprising: Comprise: A memory, a processor, a communication interface; wherein the memory stores executable code, when the executable code is executed by the processor, the processor executes the customer simulation method of claim 8.

10. A non-transitory machine-readable storage medium, characterized in that, The non-transitory machine readable storage medium stores executable code, when the executable code is executed by the processor of the electronic device, the processor executes the customer simulation method of claim 8.

11. A computer program product, characterised in that, Comprise: A computer program which, when executed by a processor of an electronic device, causes the processor to perform the client simulation method of claim 8.

Images