Interaction method, device and equipment of intelligent customer service, readable storage medium and product

Abstract

The invention relates to an intelligent customer service interaction method, device and equipment, a readable storage medium and a product. The method comprises the steps of obtaining current question information input by a target user; based on a large language model, processing the current problem information to obtain user state information corresponding to the target user; based on a retrieval enhancement generation model, processing the user state information to obtain question answering information corresponding to the current question information; and outputting question answering information. By adopting the method, the intelligent customer reply accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of financial technology, and in particular to an interactive method, apparatus, device, readable storage medium, and product for intelligent customer service. Background Technology

[0002] Intelligent customer service has become a basic feature of the banking industry. Most major banks have equipped themselves with intelligent customer service systems to meet customers' daily inquiry needs and improve service efficiency and customer experience.

[0003] Different banks' intelligent customer service systems vary in the details and features of their functions. While their level of intelligence is constantly improving, there are still shortcomings. The main issue is that their language comprehension ability needs to be enhanced. When faced with some complex questions, intelligent customer service systems still provide inaccurate or irrelevant answers, which urgently need to be addressed. Summary of the Invention

[0004] Therefore, it is necessary to provide an interactive method, device, equipment, readable storage medium, and product for intelligent customer service that can improve the accuracy of intelligent customer responses, addressing the aforementioned technical problems.

[0005] Firstly, this application provides an interactive method for intelligent customer service, including:

[0006] Obtain the current question information input by the target user;

[0007] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0008] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0009] Output the solution information for the problem.

[0010] In one embodiment, the user state information includes user intent information, which is vectorized information; correspondingly, based on the retrieval enhancement generation model, the user state information is processed to obtain the question answer information corresponding to the current question information, including:

[0011] Determine the correlation between user intent information and different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model;

[0012] Based on the degree of correlation, the target solution vector is selected from each candidate solution vector;

[0013] Based on the target solution vector, generate the solution information corresponding to the current problem information.

[0014] In one embodiment, the target solution vector is selected from the candidate solution vectors based on each relevance degree, including:

[0015] Determine the target solution vector;

[0016] Among them, the target solution vector is the candidate solution vector with the highest correlation value.

[0017] In one embodiment, based on a large language model, the current question information is processed to obtain user status information corresponding to the target user, including:

[0018] Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information;

[0019] Based on the semantic understanding network in the large language model, semantic understanding of text information is performed to obtain the semantic information corresponding to the current question information;

[0020] The semantic information is vectorized to obtain the user intent information corresponding to the current question information.

[0021] In one embodiment, based on a large language model, the current question information is processed to obtain user status information corresponding to the target user, including:

[0022] Obtain historical problem information of the target user;

[0023] Given historical question information, the similarity between current question information and historical question information is determined based on a large language model.

[0024] If the similarity exceeds the preset similarity threshold, the historical question information and the current question information are combined as new current question information. Based on the large language model, the new current question information is processed to obtain the user status information corresponding to the target user.

[0025] In one embodiment, the user state information includes user intent information and user emotion information; correspondingly, the method further includes:

[0026] If the solution to the problem is not yet confirmed, the current problem information, user intent information, and user emotion information are output to the target customer service representative so that the target customer service representative can determine the corresponding solution information based on the current problem information, user intent information, and user emotion information.

[0027] Secondly, this application also provides an interactive device for intelligent customer service, comprising:

[0028] The problem acquisition module is used to acquire the current problem information input by the target user;

[0029] The first determination module is used to process the current problem information based on the large language model to obtain the user status information corresponding to the target user;

[0030] The second determining module is used to process user status information based on the retrieval enhancement generation model to obtain the answer information corresponding to the current question information;

[0031] The answer output module is used to output the answer information for the question.

[0032] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0033] Obtain the current question information input by the target user;

[0034] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0035] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0036] Output the solution information for the problem.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0038] Obtain the current question information input by the target user;

[0039] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0040] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0041] Output the solution information for the problem.

[0042] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0043] Obtain the current question information input by the target user;

[0044] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0045] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0046] Output the solution information for the problem.

[0047] The aforementioned intelligent customer service interaction method, device, equipment, readable storage medium, and product acquire the current question information input by the target user; process the current question information based on a large language model to obtain the user status information corresponding to the target user; and process the user status information based on a retrieval-enhanced generation model to obtain the question answer information corresponding to the current question information. In this process, the powerful semantic understanding and context awareness capabilities of the large language model are utilized to accurately parse the target user's question. Secondly, a RAG enhancement mechanism is introduced to retrieve the most relevant authoritative information fragments from a dynamically updated financial knowledge base in real time and inject them as context into the generation process of the large language model, making the question answer information generated by the large language model more accurate. In other words, the above process can improve the accuracy of intelligent customer service responses. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating the interaction method of an intelligent customer service system in one embodiment;

[0050] Figure 2 This is a flowchart illustrating the steps for determining the question-answering information in one embodiment;

[0051] Figure 3 This is a flowchart illustrating the steps for determining user status information in one embodiment;

[0052] Figure 4 This is a flowchart illustrating the user status information determination step in another embodiment;

[0053] Figure 5A This is an application environment diagram of an intelligent customer service interaction method in one embodiment;

[0054] Figure 5B This is a flowchart illustrating the interaction method of intelligent customer service in another embodiment;

[0055] Figure 6 This is a structural block diagram of an intelligent customer service interaction device in one embodiment;

[0056] Figure 7This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] Before introducing the embodiments of this application, it should be noted that intelligent customer service has become a basic function of the banking industry. Most major banks are equipped with intelligent customer service systems to meet customers' daily inquiry needs and improve service efficiency and customer experience. Whether it's a large state-owned bank or a joint-stock bank, their telephone banking systems and mobile banking apps all have intelligent customer service functions.

[0059] Different banks offer intelligent customer service systems that vary in the details and features of their functionalities. Some banks emphasize comprehensive functionality, providing not only basic text-based Q&A but also various interactive methods such as voice input and video customer service. Other banks focus on more distinctive aspects of their intelligent customer service interfaces and question guidance to better guide customers in using them. For example, Bank of Communications has a relatively complete basic setup, while also offering a smooth voice input experience and intelligent functions such as wealth management tools and a smart wealth management assistant. China Everbright Bank offers a variety of systems including intelligent customer service, digital human-themed branches, wealth management manager mini-stores, and video customer service.

[0060] Different banks' intelligent customer service systems vary in the details and features of their functional settings. While their level of intelligence is constantly improving, there are still shortcomings. These are mainly reflected in the need to enhance language comprehension capabilities and improve multi-turn dialogue capabilities. When faced with some complex questions, intelligent customer service systems still provide inaccurate or irrelevant answers. In particular, for highly specialized questions involving specific business processes, laws and regulations, the answers from intelligent customer service systems may not be precise or comprehensive enough.

[0061] The integration of intelligent customer service and human customer service is becoming increasingly close. When human customer service is busy, intelligent customer service can first screen and categorize customer questions, quickly resolving common and simple issues and reducing the workload of human customer service representatives. At the same time, intelligent customer service can also provide relevant information and suggestions to human customer service representatives, helping them to better answer customer questions. Most banks' intelligent customer service systems have achieved seamless transfer to human customer service. When a customer's question cannot be resolved through intelligent customer service, it can be easily transferred to human customer service. However, if the transfer process is not properly configured, the number of transfers to human customer service can increase significantly, leading to longer waiting times and reduced customer satisfaction when queues are required.

[0062] In light of this, this patent proposes an intelligent customer service management method and system that deeply integrates Large Language Model (LLM) and Retrieval-Augmented Generation (RAG) technologies, specifically designed for digital financial scenarios with high security and accuracy requirements. The core of the system lies in constructing a closed-loop intelligent service system encompassing "understanding-retrieval-generation-management." First, leveraging the powerful semantic understanding and context awareness capabilities of the LLM, it accurately analyzes complex, unstructured financial inquiries initiated by users through multiple channels (such as apps, online banking, and telephone), covering the entire business scope including account inquiries, product recommendations, transaction operations, and risk warnings. Second, it innovatively introduces a RAG enhancement mechanism. The system does not rely solely on the internal knowledge of the LLM but retrieves the most relevant authoritative information fragments from a dynamically updated financial knowledge base (including product manuals, regulatory policies, historical work orders, compliance regulations, and real-time market data) in real time and injects them as context into the generation process of the LLM. This effectively solves the "illusion" problem that may occur with LLM in specialized fields, significantly improving the accuracy, compliance, and timeliness of responses. Simultaneously, the system integrates intelligent routing and agent assistance modules. Based on dialogue content and sentiment analysis, it automatically judges the complexity of the problem and the customer's emotions, achieving seamless human-machine collaboration: simple problems are automatically handled by AI in a closed loop, while complex or sensitive problems are accurately transferred to human agents, and key information summaries and handling suggestions retrieved by RAG are pushed, greatly improving the efficiency of human service. In addition, the system has continuous learning and knowledge base self-evolution capabilities, automatically optimizing retrieval strategies and knowledge base content by analyzing user feedback and dialogue logs. This invention, through the collaboration of a large model and RAG, constructs a new paradigm of secure, accurate, explainable, and traceable digital financial intelligent customer service, achieving a unity of intelligent service, refined management, and controllable risk.

[0063] In one exemplary embodiment, such as Figure 1 As shown, an interactive method for intelligent customer service is provided. Taking the application of this method to intelligent customer service as an example, the following steps are included:

[0064] S110, Obtain the current problem information input by the target user.

[0065] The target users are those who have consultation needs. Understandably, given a large number of users, the intelligent customer service can serve a subset of them to respond to their inquiries. The target users are those currently being served by the intelligent customer service.

[0066] The current question information is used to characterize the target user's current inquiry question. Intelligent customer service can obtain this current question information in many ways, and this application does not impose any limitations on it.

[0067] For example, when the intelligent customer service detects a user within a preset distance, it identifies that user as the target user and outputs a question entry point, allowing the target user to enter the current question information based on the question entry point. For instance, it can output a question statement via voice, such as "How can I help you?"; or it can display the question statement and question entry point on the intelligent customer service screen, such as "Please enter your question below," etc.

[0068] Optionally, target users can also input the current problem information through multiple channels (such as applications, online banking, and telephone).

[0069] In one alternative implementation, an Application Programming Interface (API) Gateway can serve as a unified entry point for external requests and internal systems, responsible for request routing, protocol conversion, and authorization verification to ensure that external requests can enter the system securely and in accordance with regulations. Correspondingly, the data received by the API Gateway is acquired and used as the current question information input by the target user.

[0070] It should be noted that when the number of users waiting to be served exceeds the preset number, a large number of user requests can be evenly distributed to different processing nodes to avoid a single node crashing due to too many requests, thereby improving the overall throughput and availability of the system.

[0071] S120, based on a large language model, processes the current problem information to obtain the user status information corresponding to the target user.

[0072] User status information is used to characterize the current state of the target user, such as emotional state and inquiry state.

[0073] Among them, the large language model is the foundation of the "brain" of intelligent customer service. It is trained through large-scale unsupervised learning and has powerful language understanding and generation capabilities, providing underlying support for subsequent semantic understanding and answer generation.

[0074] Specifically, in this embodiment, the current problem information is input into the large language model to obtain the user status information corresponding to the target user.

[0075] S130: Based on the retrieval enhancement generation model, process the user status information to obtain the answer information corresponding to the current question information.

[0076] Specifically, in this embodiment, user status information is input into the retrieval enhancement generation model to obtain the answer information corresponding to the current question information.

[0077] S140, outputs the solution information for the problem.

[0078] For example, in this embodiment, the intelligent customer service can output question and answer information through text or voice, and this application does not limit it in any way.

[0079] In one alternative implementation, if the solution to the problem is not yet determined, the current problem information, user intent information, and user emotion information can be output to the target customer service representative, so that the target customer service representative can determine the solution to the current problem information based on the current problem information, user intent information, and user emotion information.

[0080] The target customer service representative can be an idle human customer service representative, or a human customer service representative who is about to provide service to the target user.

[0081] Among them, user intent information is used to characterize the inquiry intent expressed by the current question information; user emotion information is used to characterize the emotional state of the target user when outputting the current question information.

[0082] Specifically, when the intelligent customer service cannot answer the target user's current question information, the target user's current question information, user intent information, and user emotion information are output to the human customer service, providing relevant information and suggestions to help the human customer service better answer the customer's questions, and enabling a smooth transfer between intelligent customer service and human customer service.

[0083] In addition, in this embodiment, the large language model can also perform sentiment analysis on the dialogue content, automatically judge the complexity of the current problem information and the customer's emotions, and achieve seamless human-machine collaboration: simple problems are automatically handled by intelligent customer service in a closed loop; complex or sensitive problems are accurately transferred to human customer service and pushed key information summaries and handling suggestions retrieved by RAG, which greatly improves the efficiency of human service.

[0084] In the aforementioned intelligent customer service interaction method, the following steps are taken: First, the current question information input by the target user is obtained. Then, based on a large language model, this current question information is processed to obtain the user's corresponding user status information. Finally, based on a retrieval-enhanced generation model, the user status information is processed to obtain the corresponding question answer information. This process leverages the powerful semantic understanding and context awareness capabilities of the large language model to accurately parse the target user's question. Secondly, a RAG enhancement mechanism is introduced to retrieve the most relevant authoritative information fragments from a dynamically updated financial knowledge base in real time and inject them as context into the large model's generation process. This makes the question answer information generated by the large language model more accurate. In other words, the above process improves the accuracy of intelligent customer service responses.

[0085] Based on the technical solutions of the above embodiments, this application also provides an optional embodiment. In this optional embodiment, the user state information includes user intent information, which is vectorized information. In this case, the process of processing the user state information based on the retrieval enhancement generation model to obtain the question answer information corresponding to the current question information is refined.

[0086] See Figure 2 The steps for determining the answer information shown include:

[0087] S210, determine the correlation between user intent information and different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model.

[0088] In this embodiment, the vector database corresponding to the retrieval enhancement generation model is a real-time dynamically updated financial knowledge base, which includes authoritative information fragments of various data such as product manuals, regulatory policies, historical work orders, compliance regulations, and real-time market data.

[0089] Specifically, in this embodiment, the correlation between user intent information and different candidate solution vectors in the vector database is determined, so that the target solution vector can be selected from the candidate solution vectors based on the correlation.

[0090] S220, Select the target solution vector from each candidate solution vector based on the degree of correlation.

[0091] In one alternative implementation, a target solution vector can be determined; wherein the target solution vector is the candidate solution vector with the highest correlation value.

[0092] Specifically, the candidate solution vector corresponding to the highest correlation degree among all correlation degrees is taken as the target solution vector.

[0093] S230, Generate the answer information corresponding to the current question information based on the target answer vector.

[0094] For example, in this embodiment, the target solution vector can be input into a pre-trained information transformation model to obtain the solution information corresponding to the current problem information.

[0095] In the above embodiments, the most relevant authoritative information fragments are retrieved in real time from a dynamically updated financial knowledge base (including product manuals, regulatory policies, historical work orders, compliance regulations, and real-time market data), and these fragments are injected as context into the generation process of the large model. This effectively solves the "illusion" problem that may occur in professional fields for large language models, and significantly improves the accuracy, compliance, and timeliness of the answers.

[0096] Based on the technical solutions of the above embodiments, this application also provides an optional embodiment. In this optional embodiment, when the user state information includes user intent information, the process of processing the current question information based on a large language model to obtain the user state information corresponding to the target user is described in detail.

[0097] See Figure 3 The steps for determining user status information shown include:

[0098] S310, Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information.

[0099] Among them, the text conversion model is used to convert speech information into text information, realizing speech-to-text conversion.

[0100] S320, based on the semantic understanding network in the large language model, performs semantic understanding on text information to obtain the semantic information corresponding to the current question information.

[0101] S330, the semantic information is vectorized to obtain the user intent information corresponding to the current question information.

[0102] The semantic understanding network is used to store and retrieve vectorized representations of knowledge. When a user asks a question, the system vectorizes the question.

[0103] Specifically, in this embodiment, the text information corresponding to the current question information is input into the semantic understanding network. The semantic understanding network performs semantic understanding on the text information to obtain the semantic information corresponding to the current question information, and performs vectorization processing on the semantic information to obtain the user intent information corresponding to the current question information.

[0104] In the above embodiments, the current question information is converted into text information, enabling the large language model to process the current question information and obtain user intent information. Since the user intent information is vectorized, it lays the foundation for quickly retrieving the most relevant knowledge fragments from the vector database to generate question-answering information.

[0105] Based on the technical solutions of the above embodiments, this application also provides an optional embodiment. In this optional embodiment, the process of processing the current problem information based on a large language model to obtain the user status information corresponding to the target user is described.

[0106] See Figure 4 The steps for determining user status information shown include:

[0107] S410: Obtain historical problem information of the target user.

[0108] Among them, historical question information refers to the question information entered by the target user within a preset historical period.

[0109] The preset historical period can be determined based on human experience, through a large number of experiments, or by technicians according to their needs. For example, the preset historical period can be 1 day or 3 days, and this application does not impose any limitations on it.

[0110] S420: Given historical problem information, determine the similarity between current problem information and historical problem information based on a large language model.

[0111] For example, in this embodiment, the current problem information and historical problem information can be input into the similarity determination model to obtain the similarity between the current problem information and the historical problem information.

[0112] For example, in this embodiment, vocabulary extraction can also be performed on the current problem information and historical problem information, and the similarity between the current problem information and historical problem information can be determined based on the amount of vocabulary repetition between them. For example, reference similarities corresponding to different amounts of repetition can be predetermined, and the reference similarity corresponding to the amount of vocabulary repetition between the current problem information and historical problem information can be used as the similarity between the current problem information and historical problem information.

[0113] S430: If the similarity exceeds the preset similarity threshold, the historical question information and the current question information are combined as new current question information, and the new current question information is processed based on the large language model to obtain the user status information corresponding to the target user.

[0114] The preset similarity threshold can be determined based on human experience or through a large number of experiments; this application does not impose any restrictions on it.

[0115] In this embodiment, if the similarity exceeds a preset similarity threshold, the current question information and the historical question information are considered to be a multi-turn dialogue scenario for the same question. The dialogue history can be tracked, reasoning can be performed based on the context, and complex interactions such as user rhetorical questions and pronouns (such as "it" or "that") can be handled, making the dialogue more coherent and humanized, and improving the user experience.

[0116] In the above embodiments, when historical problem information exists, new current problem information is generated based on the historical problem information and the current problem information, and the new current problem information is processed. This can handle complex interactions such as user questions and references (such as "it" or "that"), enabling clarification when the user's expression is ambiguous, reducing ineffective communication, and improving problem-solving efficiency.

[0117] Based on the technical solutions of the above embodiments, this application also provides an optional embodiment. In this optional embodiment, the interaction method of intelligent customer service is applied to, for example... Figure 5A The overall architecture of the digital financial system shown in this application provides a detailed description of the intelligent customer service interaction method.

[0118] Figure 5A The system comprises an access layer, a large model core, and a RAG application layer. Among these:

[0119] The access layer includes:

[0120] API Gateway: As a unified entry point for external requests and internal systems, it is responsible for request routing, protocol conversion, and permission verification to ensure that external requests can enter the system securely and in accordance with regulations.

[0121] Load balancing: Distributes a large number of user requests evenly across different processing nodes, preventing a single node from crashing due to excessive requests and improving the overall throughput and availability of the system.

[0122] The core of the large model includes:

[0123] Pre-trained models are the foundation of the "brain" of intelligent customer service. They are trained through large-scale unsupervised learning and have powerful language understanding and generation capabilities, providing underlying support for subsequent semantic understanding and response generation.

[0124] Vector database: Used to store and retrieve vectorized representations of knowledge. When a user asks a question, the system vectorizes the question and then quickly retrieves the most relevant knowledge fragments from the vector database, providing accurate reference materials for the large model to generate answers. This is a key component of RAG (Retrieval Augmented Generation) technology.

[0125] Semantic understanding: responsible for parsing the natural language input by users, understanding their intentions, emotions and specific needs, and transforming vague and colloquial user expressions into structured information that the system can process, so that subsequent modules can perform accurate response processing.

[0126] The RAG application layer includes:

[0127] Customer service conversation management: Controls the entire customer service conversation process, including initiation, progression, and termination. It can recognize the context of the conversation, ensure its coherence, and guide users and clarify questions when appropriate, thereby improving the user's conversation experience.

[0128] Multi-turn interaction engine: Supports multi-turn interactive communication between intelligent customer service and users. It can remember historical dialogue information, reason based on context, and handle complex interactive scenarios such as follow-up questions and counter-questions from users, making the dialogue more logical and human-like.

[0129] Knowledge Update Module: Responsible for adding, modifying, and deleting knowledge. It promptly updates the system with new product information, business rules, frequently asked questions, and other knowledge, ensuring that the intelligent customer service's answers are always up-to-date and accurate.

[0130] The intelligent customer service interaction method provided in this embodiment is based on the collaboration of a large language model and RAG (Retrieval Enhanced Generative Model). Based on the target user's current question information, it generates more accurate question answer information. In the absence of generated question answer information, it outputs the target user's user intent information, user emotion information, and knowledge retrieved from RAG to the human customer service representative. This allows the human customer service representative to understand the target user's needs in advance, making the answer to the current question information more coherent and humanized, and improving the user experience.

[0131] See Figure 5B The interactive methods of the intelligent customer service shown include:

[0132] S501, Obtain the current problem information and historical problem information input by the target user;

[0133] S502, determine whether there is historical problem information for the target user. If yes, proceed to S503; otherwise, proceed to S505.

[0134] S503, based on a large language model, determines the similarity between current question information and historical question information;

[0135] S504, If the similarity exceeds the preset similarity threshold, the historical problem information and the current problem information are combined as the new current problem information; continue to execute S505;

[0136] S505, Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information;

[0137] S506, based on the semantic understanding network in the large language model, performs semantic understanding on text information to obtain semantic information and user sentiment information corresponding to the current question information;

[0138] S507, the semantic information is vectorized to obtain the user intent information corresponding to the current question information;

[0139] S508, based on the retrieval enhancement generation model, processes the user status information to obtain the answer information corresponding to the current question information;

[0140] S509, Determine whether the problem-solving information has been obtained. If yes, proceed to S510; otherwise, proceed to S511.

[0141] S510 outputs the solution information for the problem.

[0142] S511 outputs the current problem information, user intent information, and user emotion information to the target customer service representative, so that the target customer service representative can determine the corresponding problem answer information based on the current problem information, user intent information, and user emotion information.

[0143] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0144] Based on the same inventive concept, this application also provides an intelligent customer service interaction device for implementing the intelligent customer service interaction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more intelligent customer service interaction device embodiments provided below can be found in the limitations of the intelligent customer service interaction method described above, and will not be repeated here.

[0145] In one exemplary embodiment, such as Figure 6 As shown, an intelligent customer service interaction device is provided, including: a question acquisition module 610, a first determination module 620, a second determination module 630, and an answer output module 640, wherein:

[0146] The problem acquisition module 610 is used to acquire the current problem information input by the target user;

[0147] The first determining module 620 is used to process the current problem information based on the large language model to obtain the user status information corresponding to the target user;

[0148] The second determining module 630 is used to process the user status information based on the retrieval enhancement generation model to obtain the question answer information corresponding to the current question information;

[0149] The solution output module 640 is used to output the solution information for the question.

[0150] In one embodiment, the user state information includes user intent information, which is vectorized information; correspondingly, the second determining module 630 includes a first determining unit for determining the correlation between the user intent information and different candidate answer vectors in the vector database corresponding to the retrieval enhancement generation model; a second determining unit for selecting a target answer vector from each candidate answer vector based on each correlation; and a third determining unit for generating question answer information corresponding to the current question information based on the target answer vector.

[0151] In one embodiment, the second determining unit is specifically used to determine the target solution vector; wherein the target solution vector is the candidate solution vector with the highest correlation value.

[0152] In one embodiment, the first determining module 620 includes a fourth determining unit, used to input the current question information into a text conversion model to obtain the text information corresponding to the current question information; a fifth determining unit, used to perform semantic understanding on the text information based on the semantic understanding network in the large language model to obtain the semantic information corresponding to the current question information; and a sixth determining unit, used to vectorize the semantic information to obtain the user intent information corresponding to the current question information.

[0153] In one embodiment, the first determining module 620 includes an acquisition unit for acquiring historical question information of the target user; a seventh determining unit for determining the similarity between the current question information and the historical question information based on a large language model when the historical question information is acquired; and an eighth determining unit for taking the historical question information and the current question information together as new current question information when the similarity exceeds a preset similarity threshold, and processing the new current question information based on the large language model to obtain the user status information corresponding to the target user.

[0154] In one embodiment, user status information includes user intent information and user emotion information; correspondingly, the intelligent customer service interaction device also includes an information output module, which is used to output the current question information, user intent information and user emotion information to the target customer service representative when the question answer information has not been determined, so that the target customer service representative can determine the question answer information corresponding to the current question information based on the current question information, user intent information and user emotion information.

[0155] The modules in the aforementioned intelligent customer service interactive device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0156] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements an intelligent customer service interaction method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0157] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0158] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0159] Obtain the current question information input by the target user;

[0160] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0161] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0162] Output the solution information for the problem.

[0163] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0164] Determine the correlation between user intent information and different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model;

[0165] Based on the degree of correlation, the target solution vector is selected from each candidate solution vector;

[0166] Based on the target solution vector, generate the solution information corresponding to the current problem information.

[0167] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0168] Determine the target solution vector;

[0169] Among them, the target solution vector is the candidate solution vector with the highest correlation value.

[0170] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0171] Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information;

[0172] Based on the semantic understanding network in the large language model, semantic understanding of text information is performed to obtain the semantic information corresponding to the current question information;

[0173] The semantic information is vectorized to obtain the user intent information corresponding to the current question information.

[0174] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0175] Obtain historical problem information of the target user;

[0176] Given historical question information, the similarity between current question information and historical question information is determined based on a large language model.

[0177] If the similarity exceeds the preset similarity threshold, the historical question information and the current question information are combined as new current question information. Based on the large language model, the new current question information is processed to obtain the user status information corresponding to the target user.

[0178] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0179] If the solution to the problem is not yet confirmed, the current problem information, user intent information, and user emotion information are output to the target customer service representative so that the target customer service representative can determine the corresponding solution information based on the current problem information, user intent information, and user emotion information.

[0180] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0181] Obtain the current question information input by the target user;

[0182] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0183] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0184] Output the solution information for the problem.

[0185] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0186] Determine the correlation between user intent information and different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model;

[0187] Based on the degree of correlation, the target solution vector is selected from each candidate solution vector;

[0188] Based on the target solution vector, generate the solution information corresponding to the current problem information.

[0189] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0190] Determine the target solution vector;

[0191] Among them, the target solution vector is the candidate solution vector with the highest correlation value.

[0192] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0193] Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information;

[0194] Based on the semantic understanding network in the large language model, semantic understanding of text information is performed to obtain the semantic information corresponding to the current question information;

[0195] The semantic information is vectorized to obtain the user intent information corresponding to the current question information.

[0196] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0197] Obtain historical problem information of the target user;

[0198] Given historical question information, the similarity between current question information and historical question information is determined based on a large language model.

[0199] If the similarity exceeds the preset similarity threshold, the historical question information and the current question information are combined as new current question information. Based on the large language model, the new current question information is processed to obtain the user status information corresponding to the target user.

[0200] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0201] If the solution to the problem is not yet confirmed, the current problem information, user intent information, and user emotion information are output to the target customer service representative so that the target customer service representative can determine the corresponding solution information based on the current problem information, user intent information, and user emotion information.

[0202] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0203] Obtain the current question information input by the target user;

[0204] Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user;

[0205] Based on the retrieval enhancement generation model, user status information is processed to obtain the answer information corresponding to the current question information;

[0206] Output the solution information for the problem.

[0207] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0208] Determine the correlation between user intent information and different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model;

[0209] Based on the degree of correlation, the target solution vector is selected from each candidate solution vector;

[0210] Based on the target solution vector, generate the solution information corresponding to the current problem information.

[0211] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0212] Determine the target solution vector;

[0213] Among them, the target solution vector is the candidate solution vector with the highest correlation value.

[0214] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0215] Input the current problem information into the text conversion model to obtain the text information corresponding to the current problem information;

[0216] Based on the semantic understanding network in the large language model, semantic understanding of text information is performed to obtain the semantic information corresponding to the current question information;

[0217] The semantic information is vectorized to obtain the user intent information corresponding to the current question information.

[0218] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0219] Obtain historical problem information of the target user;

[0220] Given historical question information, the similarity between current question information and historical question information is determined based on a large language model.

[0221] If the similarity exceeds the preset similarity threshold, the historical question information and the current question information are combined as new current question information. Based on the large language model, the new current question information is processed to obtain the user status information corresponding to the target user.

[0222] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0223] If the solution to the problem is not yet confirmed, the current problem information, user intent information, and user emotion information are output to the target customer service representative so that the target customer service representative can determine the corresponding solution information based on the current problem information, user intent information, and user emotion information.

[0224] 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 this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0225] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory 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 can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0226] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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 application.

[0227] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An interactive method for intelligent customer service, characterized in that, The method includes: Obtain the current question information input by the target user; Based on the large language model, the current problem information is processed to obtain the user status information corresponding to the target user; Based on the retrieval enhancement generation model, the user status information is processed to obtain the question answer information corresponding to the current question information; Output the answer to the question.

2. The method according to claim 1, characterized in that, The user state information includes user intent information, which is vectorized information; correspondingly, the process of processing the user state information based on the retrieval enhancement generation model to obtain the question answer information corresponding to the current question information includes: Determine the correlation between the user intent information and the different candidate solution vectors in the vector database corresponding to the retrieval enhancement generation model; Based on the aforementioned correlation degree, a target solution vector is selected from the aforementioned candidate solution vectors; Based on the target solution vector, generate the solution information corresponding to the current problem information.

3. The method according to claim 2, characterized in that, The step of selecting the target solution vector from the candidate solution vectors based on the respective correlation degrees includes: Determine the target solution vector; The target solution vector is the candidate solution vector with the highest correlation value.

4. The method according to claim 2, characterized in that, The process of processing the current problem information based on the large language model to obtain the user status information corresponding to the target user includes: The current problem information is input into the text conversion model to obtain the text information corresponding to the current problem information; Based on the semantic understanding network in the large language model, the text information is semantically understood to obtain the semantic information corresponding to the current question information; The semantic information is vectorized to obtain the user intent information corresponding to the current question information.

5. The method according to any one of claims 1-4, characterized in that, The process of processing the current problem information based on the large language model to obtain the user status information corresponding to the target user includes: Obtain the target user's historical question information; Having obtained the historical question information, the similarity between the current question information and the historical question information is determined based on the large language model. If the similarity exceeds a preset similarity threshold, the historical question information and the current question information are combined as new current question information, and the new current question information is processed based on the large language model to obtain the user status information corresponding to the target user.

6. The method according to any one of claims 1-4, characterized in that, The user state information includes user intent information and user emotion information; correspondingly, the method further includes: If the solution to the problem is not yet determined, the current problem information, the user intent information, and the user emotion information are output to the target customer service representative, so that the target customer service representative can determine the solution to the current problem information based on the current problem information, the user intent information, and the user emotion information.

7. An interactive device for intelligent customer service, characterized in that, The device includes: The problem acquisition module is used to acquire the current problem information input by the target user; The first determining module is used to process the current problem information based on a large language model to obtain the user status information corresponding to the target user; The second determining module is used to process the user status information based on the retrieval enhancement generation model to obtain the question answer information corresponding to the current question information; The solution output module is used to output the solution information for the question.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images