Intelligent sales method and device based on large model Agent, equipment and medium

Abstract

The invention discloses an intelligent sales method, device and equipment based on a large model Agent and a medium, and relates to the field of intelligent sales, the method comprises the following steps: using a large language model to carry out intention analysis on a user problem, determining whether a tool matched with the user problem exists, and obtaining a judgment result; if the judgment result is no, generating an answer to the user question by using a large language model and a system instruction, and feeding back the answer to the user; and if the judgment result is yes, determining a tool matched with the user question as a target tool, calling an external resource corresponding to the target tool by utilizing the target tool, generating an answer to the user question, and feeding back the answer to the user. According to the method, the practicability, the efficiency and the real-time performance of the large language model in a sales environment are improved, and the difficulty of intention recognition is reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent sales, and in particular to an intelligent sales method, device, equipment and medium based on a large model Agent. Background Art

[0002] With the rapid development of artificial intelligence (AI), especially the large language model (LLM) in the field of natural language processing (NLP), its powerful text understanding and generation capabilities have been witnessed in many fields. These models can quickly and accurately parse and respond to human language, greatly improving the efficiency and naturalness of human-computer interaction. Of course, these capabilities can also be used well in the fields of e-commerce and sales.

[0003] In today's society, e-commerce has become an important profitable industry with increasingly fierce competition. In order to improve purchase conversion rate and customer satisfaction, e-commerce platforms need to continue to innovate and provide more personalized and efficient services. However, traditional manual customer service has limitations such as work efficiency and working hours, making it difficult to meet the growing customer needs.

[0004] The introduction of AI technology has brought new possibilities to the e-commerce industry. Unlike humans, AI systems do not need to rest and can provide services around the clock, greatly improving service efficiency and purchase conversion rates.

[0005] In terms of data processing, AI is much more efficient than humans, making it faster and more efficient when adapting to sales work. At present, some studies have used fine-tuning large models to achieve relevant responses of the models at different sales stages, further improving sales results. However, the process of fine-tuning the model is very time-consuming and requires a large amount of data.

[0006] In terms of sales, although large models have shown great potential, they also have some technical shortcomings. The following are some of the current technical shortcomings of large models in sales:

[0007] 1. Inconvenient interaction with peripheral systems: In a highly digital sales environment, seamless integration between various systems is crucial. However, the big model currently lacks the ability to interact with payment systems or other key business systems. This means that although the big model can provide valuable sales insights or suggestions, it cannot automatically convert this information into actual sales actions, such as recommending products after querying the product list and updating the order status, adjusting inventory, or triggering subsequent service processes after placing an order. This limitation greatly reduces the practicality and efficiency of the big model in a complex sales environment.

[0008] 2. Intent recognition is difficult and inefficient: Due to the complexity and design limitations of large language models, they are not ideal for identifying user intent in a conversational environment. The diversity and complexity of user intent make accurately identifying user intent a very challenging task. Mainstream methods are relatively time-consuming and require multiple and detailed instruction executions through large models to barely complete the understanding of user intent. In actual application scenarios, it is not only necessary to identify user intent in general scenarios, but also to identify user intent in scenarios related to the current business, which will make it more difficult to identify user intent.

[0009] 3. Low knowledge update frequency and poor real-time performance: Mainstream methods require a long time to train large models, which may result in the model being unable to adapt to new market trends or changes in user behavior in a rapidly changing sales environment. In addition, when new products or new user needs emerge, large models may need to be retrained to adapt to these changes, which will affect their real-time performance.

[0010] In summary, although large models have many advantages in sales, they also have many technical disadvantages and challenges. Summary of the invention

[0011] The purpose of this application is to provide an intelligent sales method, device, equipment and medium based on a large model agent, which can improve the practicality, efficiency and real-time performance of the large language model in the sales environment and reduce the difficulty of intent recognition.

[0012] To achieve the above objectives, this application provides the following solutions:

[0013] In a first aspect, the present application provides an intelligent sales method based on a large model Agent, and the intelligent sales method based on a large model Agent includes:

[0014] Use a large language model to analyze the user's question intent, determine whether there is a tool that matches the user's question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool;

[0015] If the judgment result is no, the large language model and system instructions are used to generate an answer to the user's question and feed it back to the user;

[0016] If the judgment result is yes, the tool matching the user question is determined to be the target tool, and the target tool is used to call the external resources corresponding to the target tool to generate the answer to the user question and feed it back to the user.

[0017] Optionally, the system instruction is used to set the role, skills and precautions of the Agent that applies the large language model, so that the generated answers to the user questions are answers in a sales scenario.

[0018] Optionally, the target tool is used to call an external resource corresponding to the target tool to generate an answer to the user's question, specifically including:

[0019] Based on user questions, use the large language model to complete the business fields in the target tool;

[0020] Use the completed business fields in the target tool as input parameters, call external resources, and generate answers to user questions.

[0021] Optionally, based on the user's question, the large language model is used to complete the business fields in the target tool, including:

[0022] When the target tool is a sales tool, a large language model is used to extract keywords representing product types and / or product functions in the user's question, and the keywords are used to complete the business fields in the sales tool; wherein the business fields of the sales tool include product types and product functions;

[0023] When the target tool is an order placing tool, the product database is searched to determine the product number of the product selected by the user to be ordered, the mobile phone number is collected from the user in a counter-questioning manner, and the business field in the order placing tool is completed using the product number and the mobile phone number; wherein the business field of the order placing tool includes: mobile phone number and product number;

[0024] When the target tool is an after-sales tool, the mobile phone number is collected from the user in a counter-questioning manner or obtained from the conversation history, and the user's order information list is queried based on the mobile phone number for the user to select, and the order number of the order to be returned is obtained based on the user's selection result, and the business field of the after-sales tool is completed using the mobile phone number and the order number; wherein the business field of the after-sales tool includes the mobile phone number and the order number.

[0025] Optionally, the external resource corresponding to the sales tool is a product database. When the target tool is a sales tool, the completed business fields in the target tool are used as input parameters to call the external resource to generate an answer to the user's question, specifically including:

[0026] The business fields in the completed sales tool are used as conditions to query the commodity database to obtain commodity information as an answer to the user's question; the business fields in the sales tool include commodity types.

[0027] Optionally, the external resources corresponding to the order placement tool include peripheral systems, order databases, and payment systems; when the target tool is an order placement tool, the completed business fields in the target tool are used as input parameters to call external resources to generate answers to user questions, specifically including:

[0028] Based on the completed business fields in the order placement tool, call the peripheral system to create an order; the business fields in the order placement tool include mobile phone number and product number;

[0029] The order is added to the order information table, and the payment link corresponding to the order is output as the answer to the user's question; the payment link is used to call the payment system.

[0030] Optionally, the external resource corresponding to the after-sales tool is a return and exchange business system; when the target tool is an after-sales tool, the business fields in the completed target tool are used as input parameters, and the external resource is called to generate an answer to the user's question, specifically including:

[0031] Determine the order corresponding to the business field in the completed after-sales tool, and output the return and exchange link corresponding to the order as the answer to the user's question; the return and exchange link is used to call the return and exchange business system; the business field of the after-sales tool includes a mobile phone number and an order number.

[0032] In a second aspect, the present application provides an intelligent sales device based on a large model Agent, wherein the intelligent sales device based on a large model Agent applies the above-mentioned intelligent sales method based on a large model Agent, and the intelligent sales device based on a large model Agent includes:

[0033] An intent analysis module is used to use a large language model to perform intent analysis on user questions, determine whether there is a tool that matches the user question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool;

[0034] A non-business question processing module, which is used to generate an answer to the user's question using a large language model and system instructions if the judgment result is no, and feed it back to the user;

[0035] The business problem processing module is used to determine the tool matching the user's question as the target tool if the judgment result is yes, use the target tool to call the external resources corresponding to the target tool, generate the answer to the user's question, and feed it back to the user.

[0036] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned large model Agent-based intelligent sales methods.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned large model Agent-based intelligent sales methods.

[0038] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0039] The present application provides an intelligent sales method, device, equipment and medium based on a large model agent. The method applies a large prediction model, system instructions and tools, uses the knowledge and processing power of a large language model to realize the analysis of user intentions, uses system instructions to clarify the role of the intelligent agent, so that the large language model can better answer the user's questions in the sales field without using tools, and uses tools to call external APIs, access proprietary information knowledge bases and execute codes. The use of these tools expands the execution capabilities of the intelligent agent, enabling it to obtain real-time information or perform complex computing tasks by calling external resources when the internal knowledge of the model is insufficient, so that the general large language model can still accurately answer questions without the need for a lot of training without the need for relevant knowledge. The practicality, efficiency and real-time performance of the large language model in the sales environment are improved, and the difficulty of intent recognition is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A flowchart of an intelligent sales method based on a large model Agent provided in one embodiment of the present application;

[0042] Figure 2 A schematic diagram of an intelligent sales method based on a large model Agent provided in one embodiment of the present application;

[0043] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0046] In an exemplary embodiment, an intelligent sales method based on a large model agent is provided, such as Figure 1 As shown, the steps include steps 101 to 103. Among them:

[0047] Step 101, use a large language model to perform intent analysis on the user question, determine whether there is a tool that matches the user question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool.

[0048] Step 102: If the judgment result is no, the large language model and system instructions are used to generate an answer to the user's question and feed it back to the user.

[0049] Step 103, if the judgment result is yes, determine that the tool matching the user question is the target tool, use the target tool to call the external resources corresponding to the target tool, generate the answer to the user question, and feed it back to the user.

[0050] Implementing the above steps 101 to 103 improves the practicality, efficiency and real-time performance of the large language model in the sales environment and reduces the difficulty of intent recognition.

[0051] like Figure 2 As shown, in step 101, the user inputs the commodity to be purchased or the problem encountered in the input box to generate a user question, and the question format is a text format.

[0052] For example, "I want to buy basketball shoes", "I want to place an order", "I want to return goods".

[0053] In step 101, the Agent performs intent analysis on the user's question.

[0054] The Agent intelligent body consists of three parts: large model, system instructions, and tools.

[0055] Agent can be understood as a system that can make decisions independently and use a variety of tools to complete complex tasks. In this system, the universal large language model plays the role of "core scheduler". This scheduler is responsible for interpreting the user's natural language input, planning a series of executable actions, and then relying on tools and other components to gradually complete these actions.

[0056] The universal large language model provides a wide range of knowledge and processing capabilities, which is the key to the intelligent agent's ability to understand and respond to various questions. In this method, the universal large model can be selected from existing open source large models, such as the Wenxin Yiyan universal large model.

[0057] The system instruction is the prompt project in the fine-tuning of the large model. It is a descriptive text. The so-called prompt project is to let the large model know clearly and accurately what you want to do, and guide the model to fully explore its ability limit. The instruction can be the role you want the intelligent agent to play, what kind of results you want it to generate based on the user's input information, and detailed requirements for the content it generates. Its main function is to enable the general large language model to better answer users' questions in the sales field without using tools.

[0058] The use of tools in the Agent mainly involves the ability to call external APIs, access proprietary information knowledge bases, and execute code. The use of these tools expands the execution capabilities of the agent, enabling it to obtain real-time information or perform complex computing tasks by calling external resources when the internal knowledge of the model is insufficient. For example, when dealing with questions like "Who is Yao Ming's wife? How old is she now? What is the 0.76th power of her age?", the agent can use tools such as the general large language model interface, Baidu search interface, and mathematical calculation interface to achieve a general large language model that can accurately answer questions without relevant knowledge.

[0059] In the embodiment of the present application, the universal big model can select the currently existing open source big model, such as the Wenxin Yiyan universal big model, or other big models.

[0060] After the user submits a question, it is necessary to perform intent analysis on the user's question. The intent analysis method mainly uses the general large language model in the Agent to determine whether the user's question is related or matched with the tool description. If it matches, the corresponding tool will be called and the process within the tool will be executed. Compared with ordinary intent analysis, it makes it easier to determine whether the user's question is related or matched with the tool description to implement intent analysis, and the accuracy is also improved.

[0061] The tool description is a paragraph of text that describes the function or role of the tool. The tool description determines whether the Agent will call a tool and what tool to call after the user asks a question, so as to execute the corresponding business process. Therefore, the tool description information set should be accurate and clear in business description, which will make it easier to match the user's business question with the business tool description, thereby improving the efficiency and accuracy of intent recognition.

[0062] For example, the description of a sales tool could be: "Sales tool: when a user wants to buy a product or mentions a certain product, a list of related products will be recommended to the user"; the description of an ordering tool could be: "Ordering tool: when a user decides to place an order, an ordering link to the surrounding system will be provided to the user"; the description of an after-sales tool could be: "After-sales tool: when a user wants to return or exchange a product, a return or exchange application link to the surrounding system will be provided".

[0063] like Figure 2 As shown, in step 102, when there is no tool matching the user's question, the Agent's general model combines system instructions to answer the question.

[0064] When the user's question is small talk or unrelated to the business, after the Agent's intent analysis, the tool will not be triggered. The Agent will use the internal general big model to answer the user's question. The general big model will refer to the content in the system command to answer the user's question, so the content of the system command determines the scenario that the Agent can handle. In the sales scenario, the system command needs to be set to the knowledge in the sales field to guide it on how to answer the user's questions according to the user's needs. For example: in the sales scenario, set the system with requirements such as roles, skills, and precautions to ensure that the answers generated by the big model are consistent with the sales field scenario.

[0065] Example:

[0066] Role: You are an experienced and skilled salesperson.

[0067] Skill 1: Understand customer needs

[0068] 11. When customers visit, proactively and enthusiastically ask about their needs and preferences.

[0069] 12. Listen carefully to the customer's description and accurately grasp the customer's core needs through effective communication.

[0070] Skill 2: Product Recommendation

[0071] 21.Accurately recommend suitable products based on customer needs.

[0072] 22. Introduce the product’s features, advantages and usage in detail.

[0073] Skill 3: Resolving Customer Concerns

[0074] 31. Provide professional and clear answers to customers’ questions about products.

[0075] 32. If you encounter any uncertain questions, consult relevant information or colleagues in a timely manner to ensure that your response is accurate.

[0076] limit:

[0077] Only communicate with customers around sales-related topics and refuse to discuss irrelevant matters.

[0078] The output content must be organized according to the given format and cannot deviate from the framework requirements.

[0079] The text introducing the product should be concise and clear, highlighting the key points.

[0080] like Figure 2 As shown, in step 103, when there is a tool matching the user's question, the following steps 201-202 are included.

[0081] Step 201: Agent calls a tool.

[0082] Tools are an indispensable part of Agent intelligent bodies. Their structure is mainly composed of three core components: tool description, tool business field setting and tool logic processing layer. Tools are set up in advance for the use of agents. In this method, three tools are mainly set up, including sales tools, order tools and after-sales tools.

[0083] The tool description has been mentioned in step 102. The business fields of the tool include field types and field descriptions. The field types are several types commonly used in development, such as strings, numbers, etc. The description of the field needs to be combined with the specific business, and different businesses require different business field descriptions. For example, in a sales tool, two business fields, product category and product function, may be included. Among them, the field type of the product category is a string, and its description can be "shoe category, such as: sports shoes, basketball shoes, casual shoes, etc."; the field type of the product function is also a string, and its description can be "shoe function, such as: running, playing basketball, leisure, etc.".

[0084] The tool logic processing layer mainly relies on the judgment and use of the tool business fields to perform different logical operations. For example, when the information provided by the user is incomplete, the user can be asked to obtain more information; when the information provided is incorrect, the user can be guided to correct the error; when the information provided meets a certain business requirement, the corresponding third-party API can be called to implement a specific business; and interaction with peripheral systems can be achieved by returning jump links, etc.

[0085] When the Agent decides to call a tool after analyzing the user's question intent, it enters the internal execution process of the tool:

[0086] (1) Business field completion: First, business field completion is required. This is mainly done by using the general model in the Agent to extract appropriate text from the user's question based on the description of the business field to complete the content of the business field. For example, in the vending tool, when the user asks "I want to buy basketball shoes", the large model in the Agent will complete the content of the product type business field of the vending tool to "basketball shoes" based on the user's description.

[0087] (2) Business field judgment: After completion, it is necessary to judge whether certain necessary or non-empty business fields have been completed and whether the business fields have been completed correctly. If they are not completed or are incorrect, the corresponding prompt information will be returned. For example, when a user wants to place an order, the order tool will be called. If the mobile phone number business field in the order tool is not completed or is incorrect, the corresponding error prompt information will be returned, such as: "The mobile phone number you provided is incorrect" or "The mobile phone number you provided is incomplete" and other information.

[0088] (3) Logical processing: After the business fields are collected, the logical processing layer is entered. The logical processing layer mainly relies on the judgment and use of the tool business fields to perform different logical operations. For example, in the sales tool, the product list query interface can be called based on the product type as the search condition and the product list information is returned to the agent; in the ordering tool, the user's mobile phone number and product number can be used as input parameters to call the ordering interface and the return information of the ordering interface is returned to the agent; in the after-sales tool, the mobile phone number can be used as an input parameter to query the order details interface and the order details are returned to the agent.

[0089] The following introduces the creation process of the three tools: sales tool, order tool, and after-sales tool.

[0090] The sales tool needs to set two business fields: product category and product function. Both fields can be None. If neither is collected, the default product will be recommended. If collected, the collected business fields will be used as conditions to query the product database or call the business system to query the product interface to obtain product information, making the recommended products more accurate.

[0091] The ordering tool requires setting two business fields: mobile phone number and product number.

[0092] The mobile phone number must be provided by the user. If the user does not provide it, a prompt such as incomplete information provided by the user can be returned to the intelligent agent. The intelligent agent will then collect the missing necessary business field information from the customer in a counter-questioning manner based on the returned information.

[0093] The product number can be obtained from the conversation history, because when the user selects a product, the recommended product list queried through the database each time contains the product number information, and the product number can be obtained when the user decides to purchase the product.

[0094] When the mobile phone number and product number are collected, the peripheral system can be called to create an order for the user, and the mobile phone number and product number can be inserted into the order information table of the database to prepare for after-sales and other services.

[0095] After successful insertion, the payment link is returned to the user, and the user can jump to the payment system through the link.

[0096] The after-sales tool requires setting up two input business fields: mobile phone number and order number.

[0097] The mobile phone number must be provided by the customer or obtained from the conversation history. If it is not provided or not obtained from the conversation history, the processing method is the same as the above order placement tool, and the missing necessary business field information is collected from the customer in a counter-questioning manner.

[0098] After collecting the customer's mobile phone number, the customer's order information will be queried through the mobile phone number, and the order information will be listed, allowing the customer to choose the order to be returned.

[0099] After the mobile phone number and order number are collected, a return and exchange link will be provided to the customer, who will be redirected to the return and exchange business system through the link.

[0100] Step 202: The Agent returns information to the user in conjunction with the tool.

[0101] When the Agent calls the tool, the data results or information obtained by calling the tool will be returned to the Agent, such as: product list, product order details, incomplete information prompts, etc. The general large model in the Agent will integrate these results and information to respond to user questions.

[0102] Here are a few examples of scenarios for illustration.

[0103] 1. Chatting or scenarios unrelated to business, that is, when there is no tool matching the user's problem:

[0104] When the user's question cannot be matched with the tool description and the tool business field description, such as when the user chats with the model or asks about topics unrelated to the business, it is considered as a chat or unrelated to the business scenario. The user asked "Today's weather is really nice", and the user's question does not have a matching tool, so the Agent will not call the tool to answer the question. In this scenario, the Agent will directly use the general large language model combined with the instructions set in the Agent to answer the user's question. The instructions currently used are the sample instructions in step 3.

[0105] 2. Sales scenario, that is, when the target tool is a sales tool:

[0106] When the user question matches the sales tool description or the tool business field description, for example, when the user asks about the function of a product or the type of a product, it is considered to have triggered the sales scenario. The Agent will call the sales tool and execute the business logic in the sales tool.

[0107] For example, if a user asks "Do you have basketball shoes?", the Agent will use intent analysis to choose to call the sales tool to solve the user's problem. The tool will complete the product category as "basketball shoes" and use the "basketball shoes" category as the input parameter for calling the product information query interface. Finally, the queried product information is returned to the Agent. The general large model in the Agent combines the returned information and recommends the queried product information to the user.

[0108] 3. Payment scenario, that is, when the target tool is an order placing tool:

[0109] When the user question matches the order tool description or the tool business field description, for example, when the user asks to make a payment or confirms to purchase a certain item, it is considered to have triggered the payment scenario. The Agent will call the order tool and execute the business logic in the order tool.

[0110] For example, if a user asks "Let's order this pair of Li-Ning", the Agent will use intent analysis to choose to call the ordering tool to solve the user's problem. The tool needs to complete the product number and mobile phone number. Because the product number will be returned when the previous round of conversation is a recommended product list, when the user chooses to order a product in the recommended list, the product number will be automatically completed from the conversation history. However, the user did not provide the mobile phone number, so the tool will return a message similar to "incomplete mobile phone number information" by judging the mobile phone number business field. The general large model in the Agent combines the information returned by the tool to collect the mobile phone number from the user in the form of asking the user back. After collecting these two fields, the Agent can create an order for the user by calling the peripheral system, and insert the mobile phone number and product number into the order information table to prepare for after-sales and other services. After the insertion is successful, the general large language model responds to the user in combination with the payment link returned after calling the peripheral system in the tool. The user can jump to the payment system through the link. After the payment is completed, the system automatically changes the order status.

[0111] 4. After-sales scenario, that is, when the target tool is an after-sales tool:

[0112] When the user question matches the after-sales tool description or tool business field description, for example, when the user expresses the intention to return or exchange a product, it is considered to have triggered the after-sales scenario. The Agent will call the after-sales tool and execute the business logic in the after-sales tool.

[0113] For example, if a user asks "Your shoes broke after just two days of wearing, I want to return them", then the Agent will choose to call the after-sales tool to solve the user's problem through intent analysis. The tool needs to complete the order number and mobile phone number. The mobile phone number needs to be provided by the user. If the user does not provide it, it can be obtained by asking the user in the same way as the order placement tool. The order number is obtained by calling the query order interface of the peripheral system in real time by passing the mobile phone number to obtain the order information, and the order list is returned to the user for confirmation. After all business fields are collected, the return and exchange business can be triggered. At this time, the general large language model responds to the user in combination with the return and exchange application link returned after calling the peripheral system in the tool, so that the user can jump to the peripheral system to interact with it.

[0114] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0115] 1. Enhanced ability to ask questions:

[0116] Design a reverse question mechanism in the Agent instruction set and tool set to improve the accuracy and efficiency of the AI ​​sales system. By setting business fields in the Agent tool and binding optimized descriptions to each field, it can extract the corresponding values ​​from the conversation history, and then pre-set business fields in combination with user questions. Finally, it is determined whether to ask a reverse question based on whether the business field meets the pre-set conditions.

[0117] 2. Peripheral system interaction is more convenient and flexible:

[0118] Through the logical processing of the Agent tool, the business fields collected by the Agent can be used as input parameters to call the business system API in real time to obtain data or complete business operations, and realize interaction with peripheral systems or databases. Different business interactions can also be realized by judging the completed business fields of the Agent tool according to business needs. In addition, by returning the business system link to the user, a seamless jump between the dialogue and the business system can be realized, making the interaction between the model and the business system more convenient and flexible.

[0119] 3. Improved intent recognition efficiency:

[0120] By setting the tool description and business field description, the intent range of the current business scenario is clarified. The model can analyze and determine the user intent by combining the user question and tool description, and make business scenario intent judgment and business field intent completion.

[0121] 4. Real-time improvement:

[0122] After combining with Agent, the large model is no longer bound by training data. It can call the corresponding business system in real time to obtain business data according to different business needs through the Agent tool logic processing layer, so as to achieve real-time updating of the knowledge base.

[0123] In summary, the combination of large models and agents not only enables the model to conduct business-specific inquiries, but also makes the interaction with surrounding systems more convenient and flexible. At the same time, the efficiency of intent recognition has been significantly improved, and it can more accurately understand and respond to user needs. In addition, the improvement in real-time performance enables the model to update the knowledge base in real time to adapt to the ever-changing business environment. These improvements will help improve work efficiency in the sales field, increase customer satisfaction, and create more value for the company.

[0124] Based on the same inventive concept, the embodiment of the present application also provides a smart sales device based on a large model agent for implementing the above-mentioned. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations of one or more smart sales device embodiments based on a large model agent provided below can refer to the limitations of the smart sales method based on a large model agent above, and will not be repeated here.

[0125] In an exemplary embodiment, a smart sales device based on a large model Agent is provided, comprising:

[0126] The intent analysis module is used to use a large language model to perform intent analysis on user questions, determine whether there is a tool that matches the user question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool.

[0127] The non-business question processing module is used to use the large language model and system instructions to generate answers to user questions and feed them back to the user if the judgment result is no.

[0128] The business problem processing module is used to determine the tool matching the user's question as the target tool if the judgment result is yes, use the target tool to call the external resources corresponding to the target tool, generate the answer to the user's question, and feed it back to the user.

[0129] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an intelligent sales method based on a large model Agent is implemented.

[0130] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0131] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0132] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0133] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0134] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0135] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An intelligent sales method based on a large model agent, characterized in that: The intelligent sales method based on the large model agent includes: Use a large language model to analyze the user's question intent, determine whether there is a tool that matches the user's question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool; If the judgment result is no, the large language model and system instructions are used to generate an answer to the user's question and feed it back to the user; If the judgment result is yes, the tool matching the user question is determined to be the target tool, and the target tool is used to call the external resources corresponding to the target tool to generate the answer to the user question and feed it back to the user.

2. The intelligent sales method based on large model Agent according to claim 1 is characterized in that: The system instruction is used to set the role, skills and precautions of the Agent that applies the large language model, so that the answers to the generated user questions are answers in the sales scenario.

3. The intelligent sales method based on large model Agent according to claim 1 is characterized in that: Use the target tool to call the corresponding external resources of the target tool to generate answers to user questions, including: Based on user questions, use the large language model to complete the business fields in the target tool; Use the completed business fields in the target tool as input parameters, call external resources, and generate answers to user questions.

4. The intelligent sales method based on large model Agent according to claim 3 is characterized in that: Based on user questions, the large language model is used to complete the business fields in the target tool, including: When the target tool is a sales tool, a large language model is used to extract keywords representing product types and / or product functions in the user's question, and the keywords are used to complete the business fields in the sales tool; wherein the business fields of the sales tool include product types and product functions; When the target tool is an order placing tool, the product database is searched to determine the product number of the product selected by the user to be ordered, the mobile phone number is collected from the user in a counter-questioning manner, and the business field in the order placing tool is completed using the product number and the mobile phone number; wherein the business field of the order placing tool includes: mobile phone number and product number; When the target tool is an after-sales tool, the mobile phone number is collected from the user in a counter-questioning manner or obtained from the conversation history, and the user's order information list is queried based on the mobile phone number for the user to select, and the order number of the order to be returned is obtained based on the user's selection result, and the business field of the after-sales tool is completed using the mobile phone number and the order number; wherein the business field of the after-sales tool includes the mobile phone number and the order number.

5. The intelligent sales method based on large model Agent according to claim 3 is characterized in that: The external resource corresponding to the sales tool is a product database. When the target tool is a sales tool, the completed business fields in the target tool are used as input parameters to call the external resource to generate answers to user questions, including: The business fields in the completed sales tool are used as conditions to query the commodity database to obtain commodity information as an answer to the user's question; the business fields in the sales tool include commodity types.

6. The intelligent sales method based on large model Agent according to claim 3 is characterized in that: The external resources corresponding to the order placement tool include peripheral systems, order databases, and payment systems. When the target tool is an order placement tool, the completed business fields in the target tool are used as input parameters to call external resources and generate answers to user questions, including: Based on the completed business fields in the order placement tool, call the peripheral system to create an order; the business fields in the order placement tool include mobile phone number and product number; The order is added to the order information table, and the payment link corresponding to the order is output as the answer to the user's question; the payment link is used to call the payment system.

7. The intelligent sales method based on large model Agent according to claim 3 is characterized in that: The external resource corresponding to the after-sales tool is the return and exchange business system; when the target tool is the after-sales tool, the business fields in the completed target tool are used as input parameters to call the external resource to generate answers to user questions, including: Determine the order corresponding to the business field in the completed after-sales tool, and output the return and exchange link corresponding to the order as the answer to the user's question; the return and exchange link is used to call the return and exchange business system; the business field of the after-sales tool includes a mobile phone number and an order number.

8. An intelligent sales device based on a large model Agent, characterized in that: The intelligent sales device based on the large model Agent applies the intelligent sales method based on the large model Agent according to any one of claims 1 to 7, and the intelligent sales device based on the large model Agent includes: An intent analysis module is used to use a large language model to perform intent analysis on user questions, determine whether there is a tool that matches the user question, and obtain a judgment result; the tool is a sales tool, an order placement tool, or an after-sales tool; A non-business question processing module, for generating an answer to the user's question using a large language model and system instructions, and feeding it back to the user if the judgment result is no; The business problem processing module is used to determine the tool matching the user's question as the target tool if the judgment result is yes, use the target tool to call the external resources corresponding to the target tool, generate the answer to the user's question, and feed it back to the user.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent sales method based on a large model Agent as described in any one of claims 1-7.

10. 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 sales method based on a large model agent described in any one of claims 1 to 7 is implemented.

Images