Customer service switching method, device, equipment and medium

Abstract

The embodiment of the invention discloses a customer service transfer method and device, equipment and a medium. The scheme comprises the following steps: acquiring session information of a user and a first customer service in a first skill group; performing information extraction on the session information to obtain target information; providing the target information to a skill decision model, and determining a second skill group capable of processing the problem of the user; the second skill group comprises at least one second customer service; the second skill group and the first skill group are used for processing different types of problems; the skill decision model is a model obtained by performing fine tuning training on a language model based on historical session information; and sending a service request for the user to a second customer service in the second skill group.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, equipment and medium for customer service transfer. Background Art

[0002] With the continuous development of online platforms, users can complete shopping, video watching, resource trading and other services through online platforms. In order to improve user experience, many platforms will provide customer service functions to solve problems encountered by users in the process of using the platform. In actual applications, usually after the user selects the customer service, the platform will randomly assign a customer service staff to the user. When the customer service staff cannot solve the user's problem, the customer service staff is required to transfer the user to another customer service based on his or her own experience. The entire transfer process is highly dependent on the customer service staff's own experience or knowledge, and has high requirements for the customer service staff who perform the transfer operation.

[0003] Therefore, how to provide a more convenient customer service transfer method is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The embodiments of this specification provide a method, device, equipment and medium for customer service transfer to improve the convenience of the customer service transfer method.

[0005] To solve the above technical problems, the embodiments of this specification are implemented as follows: A method for customer service transfer provided in an embodiment of this specification includes: Obtaining the conversation information between the user and the first customer service representative in the first skill group; Extracting the session information to obtain target information; The target information is provided to a skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service representative; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning a language model based on historical session information; Sending a service request for the user to a second customer service representative in the second skill group.

[0006] An embodiment of this specification provides a customer service transfer device, including: A session information acquisition module, used to acquire session information between a user and a first customer service representative in a first skill group; A target information extraction module, used to extract information from the session information to obtain target information; A transfer processing module is used to provide the target information to a skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning the language model based on historical session information; A request sending module is used to send the service request for the user to the second customer service in the second skill group.

[0007] A customer service transfer device provided in an embodiment of this specification includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtaining the conversation information between the user and the first customer service representative in the first skill group; Extracting the session information to obtain target information; The target information is provided to a skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service representative; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning a language model based on historical session information; Sending a service request for the user to a second customer service representative in the second skill group.

[0008] An embodiment of the present specification provides a computer-readable medium on which computer-readable instructions are stored. The computer-readable instructions can be executed by a processor to implement a customer service transfer method.

[0009] At least one embodiment in the present specification can achieve the following beneficial effects: by extracting the conversation information between the user and the first customer service in the first skill group, obtaining the target information and providing the target information to the skill decision model, using the skill decision model to determine the second skill group that can handle the user's problem, and sending the service request for the user to the second customer service in the second skill group, it is possible to avoid relying heavily on the personal experience or knowledge of the first customer service to determine the second skill group for transfer, thereby improving the convenience of customer service transfer.

[0010] On the other hand, since the skill decision model is trained through historical conversation information, the skill decision model can determine the second skill group more accurately and efficiently, and can also improve the accuracy of transferring customer service and improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings required for use in the embodiments or the prior art description are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.

[0012] Figure 1 It is a schematic diagram of the overall architecture of a customer service transfer method provided in an embodiment of this specification; Figure 2 It is a flowchart of a customer service transfer method provided in an embodiment of this specification; Figure 3 It is a swimlane diagram of a customer service transfer method provided in an embodiment of this specification; Figure 4 It is a structural schematic diagram of a customer service transfer device provided in an embodiment of this specification; Figure 5 It is a structural diagram of a customer service transfer device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solutions and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of one or more embodiments of this specification.

[0014] For ease of understanding, some terms in the embodiments of this specification are explained as follows: Call Center (CC): is a customer service platform built using telephone communication technology. Users can connect to the system by calling or contacting online to obtain automatic or manual customer service. Call center systems can be widely used in customer service, technical support, sales, marketing and other fields, and can support automatic call distribution, interactive voice response, queue management, multi-channel access and other functions.

[0015] Customer Service System: A technical platform for managing and optimizing customer service and support. It can be used to help companies communicate effectively with customers, solve customer problems, and improve customer satisfaction. It can also provide services such as multi-channel access, automated response, case management, and customer information management.

[0016] Retrieval-Augmented Generation (RAG): is a framework that combines information retrieval and generative models to improve the performance of natural language processing tasks. Specifically, it can be used to query a given input, find the most relevant text or text fragments to the input query, and generate more accurate and contextual text based on the most relevant text content found. It can enhance contextual understanding and provide more accurate and reliable answers.

[0017] Large model fine-tuning: Based on the existing language model, new or specific content is learned for it. The fine-tuned model can provide more accurate answers that meet the needs.

[0018] Low-Rank Adaptation (LoRA): Also known as low-rank adjustment, it is a lightweight technology that can efficiently fine-tune the model and is suitable for large-scale pre-trained models. By introducing low-rank matrices into the model to reduce the number of parameters that need to be updated during fine-tuning, it can reduce computing costs and memory consumption while maintaining model performance. It can also improve the performance of the model in specific scenarios by fine-tuning on specific tasks.

[0019] Interactive Voice Response (IVR): It guides customer calls through an automated system, allowing customers to select services by voice or keystroke, reducing waiting time.

[0020] Automatic Speech Recognition (ASR) is a technology that converts speech signals into text. It is widely used in many fields, such as voice assistants, customer service systems, meeting records, speech translation, etc.

[0021] In actual applications, during the communication process between customer service and users, if the customer service cannot solve the problem consulted by the user, the customer service will transfer the user's call to other business lines or skill groups, and the transferred customer service will handle it. Business lines will continue to increase with the development of business, so the number of skill groups will also increase. When the customer service initiates the transfer, it needs to know exactly which business each skill group can handle, so that the user can be transferred to the customer service who can solve the user's problem. This method mainly relies on the customer service staff's own experience and knowledge.

[0022] In some other technologies, the system can decide how to perform intelligent transfers based on a series of predefined rules. These rules can include judgment conditions based on factors such as user information, conversation content, historical records, and corresponding transfer targets. It is also possible to organize entities and their relationships in the field in the form of a graph based on the structured knowledge representation form of knowledge graphs to represent customer information, product information, and common problems. However, these two methods require business personnel to continuously optimize and adjust as the business changes and customer needs develop. It is difficult to fully automate and difficult to adapt to new business needs.

[0023] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0024] In order to solve the defects in the prior art, this solution provides the following embodiments: Figure 1 The figure is a schematic diagram of the overall architecture of a method for customer service transfer in an embodiment of this specification.

[0025] like Figure 1 As shown, the solution may include a first skill group 1, a user 2, a server 3, and a second customer service 4. The first skill group 1 may be the skill group that the server 3 assigns to the user 2 when receiving a service request sent by the user 2; the first customer service in the first skill group 1 may communicate with the user 2 through voice or text. The server 3 may obtain the session information between the first customer service and the user 2. If the first customer service cannot solve the problem of the user 2, the server 3 may determine the second skill group 4 that can solve the problem consulted by the user 2 based on the session information, and transfer the user 2 to the second customer service in the second skill group 4, so that the second customer service can serve the user 2 and solve the problem of the user 2.

[0026] Next, a customer service transfer method provided in an embodiment of the specification will be specifically described with reference to the accompanying drawings.

[0027] Figure 2A flowchart of a method for customer service transfer provided in an embodiment of this specification. From a program perspective, the execution subject of the process can be a program installed on a server, specifically a program installed on a server of a customer service platform, a customer service system, or a call center system. From a hardware perspective, the execution subject of the process can be a server, specifically a customer service platform, a customer service system, or a call center system.

[0028] like Figure 2 As shown, the process may include the following steps: Step 202: Obtain the session information between the user and the first customer service representative in the first skill group.

[0029] A skill group may represent one or more persons who have a series of specialized skills in a certain field or multiple fields. The skill group in the embodiments of this specification may represent a set of customer service personnel who can handle specific types of requests or problems in the field of customer service. The first skill group may include one or more first customer service personnel. If the first skill group includes multiple first customer service personnel, a first customer service person may be randomly selected to provide services to the user, or the first customer service person who provides services to the user may be determined based on preset conditions, such as the busyness being less than the preset busyness, or the first customer service person who provides services to the user may be determined in other ways. The first customer service person may be a manual customer service person or a machine customer service person. The session information may be the content of the communication between the user and the first customer service person. The session information may be voice information; it may also be text, pictures and other information.

[0030] In actual applications, users can access the customer service system by dialing a customer service phone number. The customer service system can prompt the user to select a corresponding service so as to determine the corresponding first skill group based on the service selected by the user. The customer service system can also assign the user's phone number to the first customer service in the first skill group so that after the first customer service answers the call, he can communicate with the user to handle the user's request.

[0031] In actual applications, users can also access the customer service system by clicking on the customer service control displayed on the user terminal interface. The customer service system can display a conversation interface on the user terminal based on the user's access request. The customer service system can directly assign the first customer service in the first skill group to the user without the user selecting an operation. Or after the user sends a request information in the conversation interface, the customer service system assigns the first customer service to the user based on the user's request information, so that the user can communicate with the first customer service in the conversation interface so that the first customer service can handle the user's request.

[0032] Step 204: extract the session information to obtain target information.

[0033] The target information may be information that can be used and processed by the skill decision model. If the conversation information is voice conversation information, the text conversation information obtained by converting the voice conversation information into text may be used as the target information; or the voice conversation information may be used directly as the target information without processing the voice conversation information. If the conversation information is text conversation information, the text conversation information may be used directly as the target information without processing the text conversation information.

[0034] In the embodiments of this specification, in addition to the above-mentioned session information, the target information may also include user intent, which may be the user's call intent identified by the server based on the session information between the first customer service and the user, and may specifically be the user's demands, the problem the user wants to solve, etc. The target information may also include a service summary, which may be the summary information obtained by summarizing the session information between the first customer service and the user. The target information may also include information such as user intent and service summary.

[0035] Step 206: Provide the target information to a skill decision model to determine a second skill set that can handle the user's problem.

[0036] Among them, the second skill group includes at least one second customer service; the second skill group and the first skill group are used to handle different categories of problems; the skill decision model is a model obtained by fine-tuning the language model based on historical session information.

[0037] In the embodiments of this specification, the skill groups can be classified based on business types, such as membership recharge business, payment business, financial management business, transfer business and other types of business; they can also be classified based on platform functions, such as communication function, information notification function, account management function and other functions; they can also be classified based on the user's shopping stage, such as pre-sales stage, sales stage and after-sales stage; the skill groups can also be divided in other ways, such as dividing skill groups based on the front end and the back end.

[0038] The categories of user questions handled by the first customer service in the first skill group and the second customer service in the second skill group may be different. For example, the first customer service in the first skill group can handle questions related to the membership recharge service consulted by users; the second customer service in the second skill group can handle questions related to the financial management service consulted by users.

[0039] In practical applications, a small amount of training data can be used to fine-tune the language model to obtain a skill decision model, so that the skill decision model can know the relevant skill information of each skill group, and then determine the second skill group that can solve the user's problem based on the conversation information between the first customer service and the user.

[0040] The training data can be obtained based on historical conversation information or based on expert experience. The specific fine-tuning training method can be a method of fine-tuning a pre-trained large language model, such as fine-tuning the entire model, fine-tuning a fixed part of the layers, LoRA fine-tuning technology, Adapter Tuning fine-tuning technology, and other fine-tuning methods. A pre-trained language model can refer to a language model that has been preliminarily trained based on a large amount of data, such as the BERT model, GPT-4 model, RoBERTa model, XLNet model, Electra model, and other models.

[0041] Step 208: Send the service request for the user to the second customer service in the second skill group.

[0042] In the embodiments of this specification, the service request may at least include user identification information, so that the second customer service can determine the user to be communicated based on the user identification information in the service request. The user identification information can be the user's phone number or the user's ID information. The second customer service can be any customer service personnel in the second skill group; or it can be a customer service personnel determined from the second skill group based on information such as busyness, idleness, and matching degree.

[0043] In actual applications, the customer service interface of the second customer service can display the user's incoming call information. The second customer service can answer and transfer the user's call in the customer service interface and communicate with the user to solve the problem consulted by the user.

[0044] In actual applications, in addition to providing customer service transfers between skill groups, the server can also provide customer service transfers within a skill group. Customer service transfers within a skill group can also be determined based on the busyness and idleness of the customer service of the skill group to other customer service staff who can provide services to the user. For example, if the first customer service of the first skill group is too busy to handle the user's request in time, the user can be transferred from the first customer service of the first skill group to other customer service staff in the first skill group, so that other relatively idle customer service staff in the first skill group can provide services to the user to solve the user's problem.

[0045] It should be understood that the order of some steps in the methods described in one or more embodiments of this specification can be interchanged according to actual needs, or some steps can be omitted or deleted.

[0046] Figure 2The method extracts the conversation information between the user and the first customer service in the first skill group to obtain the target information and provides the target information to the skill decision model, and uses the skill decision model to determine the second skill group that can handle the user's problem, and sends the service request for the user to the second customer service in the second skill group, thereby avoiding highly relying on the personal experience or knowledge of the first customer service to determine the second skill group for transfer, and improving the convenience of customer service transfer.

[0047] On the other hand, since the skill decision model is trained through historical conversation information, the skill decision model can determine the second skill group more accurately and efficiently, and can also improve the accuracy of transferring customer service and improve user experience.

[0048] On the other hand, even if the server updates the skill group, it can adapt to the updated skill group by optimizing the skill decision model. There is no need for excessive training and learning for each customer service staff, which can reduce the learning cost of customer service staff and help reduce the cost of customer service.

[0049] based on Figure 2 The method, the examples of this specification also provide some specific implementation plans of the method, which are described below.

[0050] As an implementation manner, optionally, before providing the target information to the skill decision model in the embodiments of this specification, the method may also include: obtaining a transfer request for the user submitted by the first customer service.

[0051] The transfer request may be generated based on the operation of the first customer service clicking a transfer button in the customer service interface; or may be generated based on other operations, such as based on a transfer voice indicating transfer by the first customer service.

[0052] As an implementation method, the transfer request may also be automatically generated by the server when the server determines through the conversation information between the first customer service and the user that the first customer service cannot solve the problem consulted by the user, without the need for the first customer service to perform a transfer operation. For example, the server identifies preset keywords from the conversation between the first customer service and the user, or identifies that the content of the first customer service's answer has a low match with the content consulted by the user, and may automatically initiate a transfer request. The preset keywords may be sentences in the speech content of the first customer service, such as: I don't quite understand what you mean; I'm sorry, I can't solve the problem you mentioned, etc., which indicate that the first customer service cannot solve the user's problem. The preset keywords may also be sentences in the customer's speech content, such as: The answer is wrong; The problem I want to solve is not this, etc., which indicate that the user feels that the first customer service cannot solve the problem.

[0053] As an implementation mode, the transfer request may also be automatically generated by the server when it determines, based on the current busyness of the first customer service, that the first customer service is relatively busy and cannot provide customer service to the user in time.

[0054] In actual applications, if the server automatically initiates a transfer request, the server can generate notification information and send the notification information to the customer service interface of the first customer service for display, so that the first customer service can know that the server is transferring the user to other customer services based on the notification information displayed on the customer service interface. Or before the server initiates the transfer request, the server can generate prompt information and send the prompt information to the customer service interface of the first customer service for display, so that the first customer service can choose to confirm the transfer of the user to other customer services based on the prompt information, or cancel the transfer of the user to other customer services. The server can generate a transfer request after determining that the first customer service has completed the transfer confirmation operation.

[0055] As an implementation method, the transfer request may also be generated based on the user's operation or instruction. For example, during a conversation between a user and a first customer service representative, the first customer service representative may provide the user with a variety of service information, and the user may determine the service to be used from the multiple service information by voice, text, click, etc. according to actual needs. After obtaining the service determined by the user to be used, the server may transfer the user to the skill group that provides the service.

[0056] In actual applications, after obtaining the transfer request, the server can obtain the session information from the database recording the session information between the user and the first customer service, so that the server can process the session information and transfer the user to other customer service.

[0057] In actual applications, the first customer service can also be a transfer customer service. For example, after the user dials the customer service number, the first customer service asks the user what problem he wants to solve, and after determining the problem the user wants to solve, transfers the user to other skill groups. For example, after the first customer service receives the call from the user, the first customer service directly provides the user with a variety of services such as query services, processing services, etc. After the user says that he wants to make an inquiry, the user can be transferred to the skill group corresponding to the query service. Or, after the first customer service receives the call from the user and hears the user's request, it can initiate a transfer request and transfer the user to a skill group that can solve the user's request. The above session information may include information such as the user's problem information and user request information.

[0058] In order to improve the accuracy of the transferred skill group, the embodiments of this specification may also use RAG technology to perform information retrieval, so as to use the information retrieved by RAG to improve the accuracy of the skill decision model. Optionally, the method described in the embodiments of this specification may also include: based on RAG technology, retrieving related information matching the session information from a knowledge base; the knowledge base contains at least one of the skill information and common problem information of each skill group; the skill information includes the skill group name, the skill group service content, the professional field of the skill group, and at least one of the historical transfer cases handled by the skill group; the problem information includes question inquiry information, the answer information corresponding to the question inquiry information, and the skill group information that handles the question inquiry information; Providing the target information to the skill decision model may specifically include: providing the target information and the associated information to the skill decision model.

[0059] In the embodiments of this specification, the skill decision model can be a language model under the RAG architecture. The knowledge base can be mainly used to provide reference data for the skill decision model to improve the decision-making ability of the skill decision model. The server can search for information from the knowledge base through the retrieval module in the RAG architecture, and integrate the found information based on the generation module to generate related information that can be input into the skill decision model.

[0060] Frequently asked questions can be obtained by sorting and summarizing historical user questions. The skill group name information can be the skill group number or the general name of the skill group, such as the name of the after-sales processing group. The skill group service content can be used to describe the services that the skill group can provide. The professional field of the skill group can indicate the business field to which the skill group belongs, such as the financial field, payment field, travel field, etc.

[0061] The historical transfer cases handled by the skill group can be successful cases handled by the skill group, or failed cases that the skill group did not successfully handle. The historical transfer cases can include case content, such as the customer service that handled the user's request, the conversation information between the customer service and the user, the customer service feedback information, etc., and can also include case number information. The skill group information that handles the question inquiry information can be identification information such as the skill group number and name; it can also be information such as the number and name of the customer service that handled the problem, and then the skill group can be determined based on the customer service information.

[0062] In addition to the above information, skill information may also include information such as the processing flow of the skill group. The processing flow of the skill group may be a process for instructing customer service to provide services to users, such as the first step is to listen to customer demands, the second step is to express understanding of customer demands, the third step is to provide solutions to users, and the fourth step is to collect user feedback and evaluation information.

[0063] As an implementation mode, in the embodiments of this specification, conversation keywords such as subject, object, verb, noun and other words in the conversation can be extracted from the conversation information, and the extracted conversation keywords are matched with the information in the knowledge base to obtain the associated information matching the conversation information. The associated information may at least include the skill information of the skill group matching the conversation information in the knowledge base, the common problem information, etc. The server may also use the generation module in the RAG framework to integrate the information matched from the knowledge base and provide it to the large model.

[0064] The server can also use semantic matching to search for related information from the knowledge base. For example, it can identify the context and sentiment in the conversation information through natural language processing technology, determine the actual needs of the user from the conversation information, and search for related information that meets the actual needs of the user from the knowledge base based on the actual needs of the user. The server can also search for related information from the knowledge base based on input queries, multimodal information retrieval, collaborative filtering, etc.

[0065] In the embodiments of this specification, the efficiency of obtaining and generating target information is improved by intelligently integrating information retrieval and generation mechanisms through RAG technology, so that the skill decision model can output accurate skill groups based on the latest data and successful cases, thereby improving the accuracy of decision-making.

[0066] In actual applications, if customer service personnel need to conduct business learning, the information in the knowledge base can also be provided to customer service personnel to improve their problem understanding and problem solving capabilities.

[0067] The knowledge base in the embodiments of this specification can be constructed based on the historical service information processed by the customer service platform. Optionally, before providing the target information to the skill decision model, the embodiments of this specification can also include: obtaining historical knowledge data; the historical knowledge data includes at least one of historical session information, skill information of each skill group, and user feedback and evaluation information; converting the historical knowledge data into a preset format to obtain knowledge data in a preset format; indexing and storing the knowledge data in the preset format according to the corresponding entries.

[0068] In the embodiments of this specification, historical conversation information may be information about ended conversations, and may include at least one of various information such as conversation records, conversation summaries, and customer demands in the conversations. As an implementation method, historical knowledge data may include voice conversation information or text conversation information that successfully solves user problems, and may also include voice conversation information or text conversation information that fails to successfully solve user problems.

[0069] The skill information may be the same as or similar to the information contained in the aforementioned skill information, and may specifically include information such as the skill name, skill group service content, and the professional field of the skill group.

[0070] User feedback and evaluation information may include evaluation information given by users to customer service staff providing services, evaluation information of users on the services handled, information on problem resolution provided by users, and user satisfaction with services.

[0071] After the server detects that the customer service and the user have ended the conversation, an evaluation window may be displayed in the user interface, and the user may fill in or select evaluation information. For example, the evaluation window may display evaluation information such as very satisfied, satisfied, average, dissatisfied, and very dissatisfied for the user to choose, or a blank box may be reserved in the evaluation window for the user to fill in the evaluation information. The server may also prompt the user to make an evaluation in other ways, such as sending an evaluation SMS to the user, displaying an evaluation link in the conversation interface between the user and the customer service, etc.

[0072] The preset format in the embodiments of this specification may be a format that facilitates data retrieval, for example, it may be a JSON format, or it may be an XML format, a CSV format, or the like.

[0073] The knowledge base may contain multiple pieces of knowledge data. One piece of knowledge data may represent relevant information of a conversation, such as the conversation content information of the conversation group, the skill group or customer service information participating in the conversation, the problem information involved in the conversation, the user's feedback and evaluation information on the conversation result, etc. The conversation content information may represent at least one of the voice information, text information, and picture information during the communication between the user and the customer service.

[0074] The knowledge base may contain fields representing various information such as skill information and common problem information. For example, skill information includes fields such as skill group name, skill group service content, skill group professional field, and historical transfer cases handled by the skill group. Common problems include fields such as question inquiry information, answers corresponding to question inquiry information, and skill groups that handle question inquiry information. Each entry may correspond to a field, and an index may be created based on each entry so that each entry can be queried based on the index to obtain the information contained in each entry. The index may be a type of index selected from B-tree index, hash index, and full-text index. In actual applications, Elasticsearch may be used to create an index and store knowledge data in the knowledge base, or other types of search and analysis engines may be used, such as Solr, OpenSearch, Meilisearch, and other engines.

[0075] In actual applications, after the conversation between the user and the customer service ends, the relevant information generated during the processing of the problem, such as the conversation information, skill group information, and the user's feedback and evaluation information, can be stored. The server can determine whether the user's problem is successfully solved through the conversation information or the user's feedback and evaluation information. If the user's problem is successfully solved, the above relevant information can be stored in the knowledge base. If the user's problem is not successfully solved, the above relevant information can be stored in other databases, which can also be included in the server. The server can also directly store the above relevant information in the knowledge base without judging whether the user's problem is successfully solved.

[0076] In actual applications, in order to facilitate calling, an API interface can also be designed for the knowledge base, so that the server can use the API interface to call the knowledge base and obtain related information matching the user session.

[0077] To facilitate the explanation of the API interface, this specification provides an example of a knowledge base query API interface, and the specific example is as follows: API Endpoint: POST / api / v1 / search Request example: { "query": "How to handle customer complaints", "skill_group": "Customer Service" } Return example: { "success": true, "results": [ { "id": "1", "question": "How to handle customer complaints?", "answer": "Follow the four-step approach: 1. Listen to customers; 2. Express understanding; 3. Provide solutions; 4. Follow up on feedback.", "related_skill_group": "Customer Service", "examples": [ "Quickly respond to customer issues", "Follow up after complaints to ensure customer satisfaction" ] }]} In the above example, query can represent the keyword or question that the user wants to query, such as "how to handle customer complaints". Skill_group can represent the skill group of interest specified by the user, which can improve the relevance of the retrieval, such as the skill group named "Customer Service". The user can also not specify the skill group. Success can indicate that the request was successful, specifically, the request was successfully retrieved and the system responded. Results can represent an array of matched knowledge base entries. The id contained in each entry can represent the unique identifier of the entry, question can represent the collected question, answer can represent the answer to the question, related_skill_group can represent the related skill group, examples can represent an example array, and the example array can be used to show the actual situation related to the problem.

[0078] As an implementation manner, optionally, the conversation information in the embodiments of this specification includes voice conversation information; and extracting information from the conversation information to obtain target information specifically includes: converting the voice conversation information into text conversation information.

[0079] In actual applications, if the acquired conversation information includes conversation information in voice form, the voice conversation information can be converted into text conversation information through ASR technology. After conversion into text information, the server can send the text information to the skill decision model, so that the skill decision model directly processes based on the text information to obtain a second skill group that can handle user problems.

[0080] ASR technology can be a technology that uses a Transformer model, a multimodal fusion model, etc. to convert voice conversation information into text conversation information.

[0081] As an implementation method, in order to improve the accuracy of the skill decision model, the main content of the conversation information can also be provided to the skill decision model. The above-mentioned information extraction of the conversation information to obtain the target information can specifically include: using the summary extraction model to extract the service summary information of the text conversation information; the service summary information includes at least one of customer demand information, problem classification information, content summary information, customer emotion classification information, and information on whether the customer problem is solved.

[0082] Providing the target information to the skill decision model may specifically include: providing the service summary information to the skill decision model.

[0083] In the embodiments of this specification, the summary extraction model can be obtained by training a language processing model, such as a BERT model, based on training data. Customer demand information can represent the specific needs expressed by the user. Problem classification information can represent the category to which the user's question belongs, such as technical support, bill inquiry, etc. Content summary information can represent the main content of the conversation information between the first customer service and the user, and can be obtained by briefly summarizing the conversation information through keywords such as words representing the subject, predicate, and object in the conversation. Customer emotion classification information can represent the user's satisfaction with the current call, or the user's emotional state in the conversation. Information on whether the customer's problem is solved can indicate whether the first customer service currently providing services to the user has solved the user's demands. Information on whether the customer's problem is solved can be specifically used to determine whether it is necessary to provide a transfer service for the user. If the first customer service has solved the user's demands, the conversation with the user can be ended without transfer; if the first customer service has not solved the user's demands, a transfer service can be provided to the user, and other customer services can be transferred to provide services to the user.

[0084] In order to more clearly illustrate the generation of a service summary, a service summary example is provided, and the specific example is as follows: { "service_summary": { "customer_request": "I would like to know my billing details.", "issue_category": "Bill Query", "content_summary": "The customer inquired about the specific details of the bill, including the composition of the charges and the payment deadline. The agent explained the charges on the bill in detail and confirmed the payment date." }} customer_request can indicate customer request information. issue_category can indicate issue classification information. content_summary can indicate content summary information.

[0085] If the conversation between the user and the customer service ends, the above service summary example may also include: "emotion_classification": "Satisfied", "resolved": true Among them, emotion_classification can represent the customer emotion classification information. Resolved can represent whether the customer's problem has been solved. True can represent successful resolution. If it has not been resolved, false can be used to represent unsuccessful resolution.

[0086] In actual applications, the server can send the service summary information to the skill decision model; it can also send the service summary information and text conversation information to the skill decision model. The server can also retrieve related information from the knowledge base based on the service summary information, and send at least one of the retrieved related information, service summary information and text conversation information to the skill decision model. The server can also share the service summary with the transferred customer service so that the transferred customer service can understand the user's demands and solve them in a timely manner.

[0087] As an implementation method, the embodiments of this specification also provide a training method for a summary extraction model. Optionally, the method described in the embodiments of this specification may also include: obtaining historical conversation texts; annotating the historical conversation texts to obtain training data containing annotation information; the annotation information includes at least one of customer demand information, problem classification information, content summary information, sentiment classification information, and information on whether the problem is solved corresponding to the historical conversation texts; and using the training data to train a pre-trained language model to obtain the summary extraction model.

[0088] In the embodiments of this specification, the historical conversation text can be obtained based on the historical text conversation information between the user and the customer service. The historical conversation text can also be obtained based on the historical voice conversation data between the user and the customer service. Specifically, it can be the historical voice conversation data between the user and the customer service, the noise in the historical voice conversation data is removed, and the historical voice conversation data after the noise is removed is converted into text conversation data. The noise can specifically be the pause part, the repeated part, etc. in the historical voice conversation data. The server can use the model to annotate the historical conversation text; it can also use a combination of manual and model methods to annotate the historical conversation text. The training data can also include historical conversation text. At least part of the annotated data can be used as a training set, the training set can be input into the pre-trained speech model, and the pre-trained speech model can be fine-tuned using the LoRA fine-tuning technology to obtain a summary extraction model.

[0089] In actual applications, the server can evaluate the obtained summary extraction model. If the evaluation result meets the requirements, the obtained summary extraction model can be put into use. If the evaluation result does not meet the requirements, the model can continue to be trained using the training set until the evaluation result meets the requirements. The evaluation result meeting the requirements can be that the loss value of the trained summary extraction model is less than or equal to the preset loss value; it can also be that the quality of the service summary generated by the trained summary extraction model can meet the preset standard, or it can be evaluated in other ways that can evaluate the model quality, which is not specifically limited here.

[0090] In practical applications, the quality and accuracy of the service summary generated by the service summary model can be evaluated regularly or irregularly. If the evaluation result fails to meet the preset requirements, the service summary model can be optimized so that the service summary model can output accurate service summary information based on the input conversation text information.

[0091] In the embodiments of this specification, the sentiment classification information may be obtained by identifying the conversation information based on an existing sentiment classification model; or may be obtained by identifying the conversation information through a trained sentiment recognition model.

[0092] The specific training method of the emotion recognition model can be as follows: Acquire text data containing emotional information; annotate the text data containing emotional information to obtain emotional feature data; the emotional feature data at least includes emotional intensity information and emotional type information; obtain a training data set based on the text data containing emotional information and the emotional feature data; train the model using the training data set to obtain a trained emotion recognition model.

[0093] The text data containing emotional information in the embodiments of this specification may be obtained from multiple data sources, such as social media, comments, forums, news articles, etc. Before annotating the text data containing emotional information, the text data may be preprocessed, specifically by removing noise data, string normalization, word segmentation, stem extraction, or word form restoration. The emotional feature data may be obtained by processing the text data based on an existing feature extraction model, or by manual annotation. The feature extraction model may specifically include models such as an emotional vocabulary model, a bag of words model, and a subject model. The server may use the training data set to train a deep learning model, such as a recurrent neural network RNN, a long short-term memory network LSTM, a convolutional neural network CNN, and other deep learning models; the training data set may also be used to train the pre-trained model by fine-tuning transfer learning to improve the accuracy of emotional classification. The emotion recognition model obtained after training can output the probability of each emotion category in the conversation information, and the emotion categories may include positive, neutral, and negative categories.

[0094] The server can also use evaluation indicators such as accuracy, recall rate, F1 value, etc. to evaluate the performance of the emotion recognition model, and can also verify the effectiveness of the model through methods such as cross-validation and confusion matrix, so as to determine whether to optimize the emotion recognition model based on performance evaluation and effectiveness.

[0095] In actual applications, the emotion recognition model can be used to identify the emotional changes of customers during the conversation in real time, so as to dynamically adjust the transfer strategy. For example, when a customer expresses dissatisfaction, the server can prioritize transferring the user to a customer service representative with a higher skill level to solve the problem more effectively and improve customer satisfaction. At the same time, it enhances the personalization and accuracy of the interaction with the customer, which helps to optimize the overall service experience.

[0096] As an implementation method, the above summary extraction model can also be called through an API interface. Optionally, the method described in the embodiment of this specification further includes: encapsulating the summary extraction model as an API interface; The extracting service summary information of the text conversation information by using a summary extraction model specifically includes: The summary extraction model is called through the API interface; and the service summary information of the text conversation information is extracted using the summary extraction model.

[0097] In the embodiments of this specification, if the evaluation of the summary extraction model meets the requirements, the summary extraction model can be encapsulated as an API interface. The server can use the API interface to call the summary extraction model, and input the session information that meets the summary extraction model input format into the summary extraction model, and can obtain the model output result information of the summary extraction model, and then can feed back the model output result information as service summary information to the customer service platform.

[0098] In the embodiments of this specification, the pre-trained language model can be fine-tuned using customer service field-specific data to improve the model's accuracy in identifying user needs and matching skill sets. As an implementation method, optionally, the method also includes: obtaining historical service record information; the historical service record information includes historical text conversation information converted from historical service voice conversation information; annotating the historical service record information to obtain training sample data containing annotated information; the annotated information includes customer demand information contained in the historical service record information and skill set information for processing the customer demand information; and using the training sample data to fine-tune the pre-trained large language model to obtain the skill decision model.

[0099] In the embodiments of this specification, if the conversation between the customer service and the user is a text conversation, the text conversation may not be converted, and the text conversation may be used as historical service record information. The historical service record information may be obtained from the terminal application used by the user, or may be data obtained from the application server.

[0100] The annotation information may be obtained by processing the historical service record information using a model, or may be obtained by manual annotation, or may be obtained by annotating the historical service record information by combining manual and model annotation. Customer demand information may indicate the problem information that the user consults with the customer service, or the problem information that the user wants to solve. The skill group information used to process the customer demand information may include information such as the professional field of the skill group, the name of the skill group, and the service content of the skill group.

[0101] The annotation information may also include service tag information, transfer note information, etc. The service tag information may be used to indicate the type of the user's question. The transfer note information may be the content information of the customer service note that provided services to the user before the transfer.

[0102] In practical applications, the pre-trained large language model can also be trained based on the information in the knowledge base to enhance the decision-making ability of the skill decision model and improve the adaptability of the skill decision model to changing business needs.

[0103] As an implementation method, the training sample data may also include service summary information corresponding to the historical text conversation information.

[0104] In practical applications, one or more text information such as text conversation information, transfer note information, business label information, service summary information, customer demand information, and skill group information for processing the customer demand information in the training sample data can be vectorized to obtain vectorized information. The pre-trained large language model is trained using the vectorized information. The server can also use the validation set to evaluate the skill decision model obtained through training, and determine the performance of the skill decision model based on the evaluation results.

[0105] In actual applications, the server can periodically train the skill decision model to optimize model parameters and improve the accuracy of the model transfer skill group. For example, optimization training can be performed once a week, once every 15 days, and so on.

[0106] The server can perform non-periodic training on the skill decision model. For example, when the success rate of the skill group transferred by the skill decision model in solving user problems is less than or equal to the preset probability, the skill decision model is optimized. For example, if the preset probability is 90%, the skill decision model has transferred 100 users in a row, and 85 users' demands can be solved by the customer service in the skill group transferred by the skill decision model, and the success rate is 85%, while the success rate of 85% is less than the preset probability of 90%. The server can use the same or different data as the above training sample data to train the skill decision model again to optimize the model and improve the decision-making ability of the model. The server can also optimize the skill decision model when the skill group transferred by the skill decision model fails to solve the user's demands for a consecutive number of times reaching a preset number. The server can also set other conditions for optimizing the skill decision model so that when it is detected that the skill decision model reaches the optimization conditions, the skill decision model is optimized to improve the decision-making performance of the skill decision model.

[0107] In order to improve user satisfaction, a second skill group that provides customer service can be selected for the user based on the user's emotions. As an implementation mode, optionally, the method described in the embodiment of this specification may also include: determining the user's emotional information based on the session information; providing the target information to the skill decision model to determine the second skill group that can handle the user's problem, which may specifically include: providing the target information to the skill decision model to obtain multiple second skill groups that can handle the user's problem; selecting a target second skill group from multiple second skill groups based on the degree of negative emotion represented by the emotional information; the probability that the target second skill group can handle the problem is positively correlated with the degree of negative emotion.

[0108] In the embodiments of this specification, prompt words can be provided for the skill decision model to prompt the skill decision model to output multiple skill groups. Emotional information can be determined by the emotion recognition model obtained through the above training to identify voice conversation information or text conversation information. Emotional information can include user emotion categories, such as positive, negative and neutral, etc.; it can also include user specific emotions, such as anger, irritation, sadness, calmness, happiness, excitement, etc.; it can also include the probability of each user's specific emotion, such as the probability of anger is 90%, the probability of sadness is 10%, etc.; it can also include the degree of negative emotion of the user, such as the degree of negative emotion of the user is 10%.

[0109] The skill decision model in the embodiments of this specification can also obtain the success probability of each skill group in successfully handling the user's problem, and determine one or more second skill groups from the skill groups with success probabilities greater than or equal to the preset success probabilities, and then select the target second skill group according to the user's negative emotion level. Specifically, the skill group with a higher success probability among multiple second skill groups can be used as the target second skill group to serve users with higher negative emotion levels.

[0110] In actual applications, the server may also not consider the user's negative emotion level and directly select any one skill group from multiple second skill groups as the target second skill group. The server may also select the target second skill group based on the workload level in the second skill group, such as using the skill group with the lowest workload level in the second skill group as the target second skill group. The server may also make a selection based on the user's negative emotion level and the workload level of the skill group. If there is only one second skill group, this second skill group may be used as the target second skill group.

[0111] In order to avoid the user waiting time being too long, a skill group can be provided to the user based on the load level of the skill group. As an implementation mode, optionally, the method described in the embodiment of this specification may also include: determining the load level of the second skill group; if the load level of the second skill group is greater than or equal to a preset threshold, sending the service request for the user to the third customer service in the third skill group; the probability that the third skill group can handle the problem is less than the probability that the second skill group can handle the problem; sending the service request for the user to the second customer service in the second skill group specifically includes: if the load level of the second skill group is less than the preset threshold, sending the service request for the user to the second customer service in the second skill group.

[0112] In the embodiments of this specification, the workload level may be determined based on the number of queued users waiting for service provided by the skill group, or may be determined based on the ratio of queued users to customer service staff in the skill group.

[0113] The skill decision model can determine multiple skill groups whose probability of successfully handling the user's problem is greater than or equal to a preset probability threshold. The server can sort the skill groups from high to low based on the probability of successfully handling the user's problem of the multiple skill groups, and use the skill group with the highest probability of successfully handling the user's problem as the second skill group. If the load level of the second skill group is greater than or equal to the preset threshold, the skill group ranked next to the second skill group can be used as the third skill group. If the load level of the third skill group is less than the preset threshold, the user's service request can be sent to the third customer service in the third skill group. If the load level of the third skill group is greater than or equal to the preset threshold, the second best skill group can continue to be selected, such as the fourth skill group whose load level is ranked next to the third skill group.

[0114] In actual applications, the server can also comprehensively determine the skill group that can provide services to the user based on the user's negative emotion level and the workload level of the skill group. Specifically, a first weight value is set for the workload level of the skill group, and a second weight value is set for the user's negative emotion level. Multiple skill groups are calculated based on the first weight value and the second weight value to obtain the comprehensive probability of each skill group, and the skill group with the highest comprehensive probability can be used as the second skill group.

[0115] In the embodiment of this specification, the first customer service can make notes on the conversation content with the user, so that the second skill group output by the skill decision model is more accurate. Optionally, the method also includes: obtaining the note information provided by the first customer service; the note information includes the user demand information of the user written by the first customer service; Providing the target information to the skill decision model may specifically include: providing the target information and the remark information to the skill decision model.

[0116] In the embodiment of this specification, the first customer service can edit information in the filling area of ​​the customer service interface, and provide the information edited by the first customer service as remark information to the skill decision model. The customer service interface can also include option information for the customer service to select the reason for the transfer, and the content corresponding to the option selected by the first customer service can be provided as remark information to the skill decision model.

[0117] In the embodiment of this specification, the remark information may include the summary information of the conversation content written by the first customer service, the question information consulted by the user, and other information; it may also include the transfer option information selected by the first customer service on the transfer platform.

[0118] In actual applications, the skill decision model under the RAG architecture can also obtain the related information in the above-mentioned knowledge base. As an implementation method, various information such as related information, target information and remark information can be provided to the skill decision model so that the skill decision model can accurately decide the second skill group.

[0119] In order to facilitate the customer service receiving the transfer to provide better services to the user, as an implementation mode, optionally, the method described in the embodiment of this specification may also include: providing the service summary information to the second customer service.

[0120] In actual applications, the answering prompt information is displayed in the customer service interface, so that the second customer service can confirm the answering based on the answering prompt information and provide customer service to the user. The server can provide the service summary information to the second customer service before or after the second customer service confirms the answering.

[0121] In the embodiments of this specification, the server may also provide the remark information of the first customer service to the second customer service. The server may include one or more of the service summary information, the remark information of the first customer service, and the like in the user's service request and send it to the second customer service. Of course, the service summary information and the remark information of the first customer service may be provided to the second customer service after the service request is sent to the second customer service; the service summary information and the remark information of the first customer service may also be provided to the second customer service before the service request is sent to the second customer service.

[0122] In order to more clearly illustrate a customer service transfer method provided in an embodiment of this specification, Figure 3 A swimlane diagram of a customer service transfer method provided in an embodiment of this specification. Figure 3 As shown, the solution may include a data processing stage and a session transfer stage, which may specifically include: Step 302: The user sends an incoming call request via the user terminal.

[0123] In actual applications, the user's incoming call request may be generated based on the user's operation of dialing a customer service center; or may be generated based on the user's operation of clicking a customer service conversation window.

[0124] The server may obtain the incoming call request and execute step 304: determining a first skill group based on the incoming call request of the user.

[0125] In the embodiment of this specification, the server can determine the first skill group from multiple skill groups. The server can randomly determine any skill group from multiple skill groups as the first skill group. The incoming call request can carry the platform identification information of the user login, and the server can also determine the skill group of the corresponding platform as the first skill group based on the platform identification information.

[0126] Step 306: The first customer service representative in the first skill group has a conversation with the user.

[0127] In actual applications, the first customer service can answer the user's call through the customer service workstation; can also receive the user's text message through the customer service workstation. The first customer service can transmit and receive the voice stream between the user and the first customer service in real time through the voice dialogue unit, so that both parties can hear the voice.

[0128] The server may execute step 308: obtaining session information between the user and the first customer service.

[0129] In actual applications, the conversation information may be text conversation information, voice conversation information, picture information, etc. If the conversation information is voice information, the ASR technology may be used to convert the voice conversation information between the first customer service and the user into text conversation information. The voice conversation information may be converted in real time or after the conversation between the user and the first customer service ends. The service system may also not convert the voice conversation information.

[0130] Step 310: Extract service summary information of the session information using a summary extraction model.

[0131] In actual applications, the service summary records the main content of the communication process between the first customer service and the user. It can be used as the content input for the skill group decision in the transfer process, or it can provide the next customer service person after the transfer with the service record that needs to be reviewed. The summary extraction model can be obtained by training the pre-trained language model using the historical conversation text and the corresponding annotation information. The summary extraction model can identify the appeal information that represents the customer's intention and the content summary information, so as to provide the necessary context information for subsequent skill decisions.

[0132] If the first customer service does not have the time or ability to solve the user's problem, the first skill group may execute step 312: perform a transfer operation.

[0133] In the embodiment of this specification, the transfer operation can be an operation in which the first customer service selects a transfer reason in the customer service interface and clicks to submit, such as "I have not been trained on the user's problem; the user has multiple problems, I have solved some of them, but I have not been trained on the remaining problems; I need other customer service to assist in handling them" and other reasons to select and confirm. The transfer operation can also be an operation of clicking a transfer control.

[0134] In actual applications, the server may also determine through the session information between the first customer service and the user that the first customer service cannot solve the user's problem and automatically transfer the call.

[0135] The client server may generate a transfer request based on the transfer operation of the first client service, and the server may obtain the transfer request and execute step 314: using the summary information and the session information, retrieve matching related information from the knowledge base.

[0136] In the embodiments of this specification, RAG technology can be used to retrieve related information matching the summary information and text conversation information from the knowledge base to enhance the accuracy of the transfer decision and output the skill group that can provide the service. It is also possible to obtain the remark information made by the first customer service for the conversation content with the user, and retrieve information from the knowledge base in combination with the remark information.

[0137] Step 316: Submit the service summary information, session information, and associated information to the skill decision model to determine the second skill group.

[0138] In practical applications, the remark information can also be sent to the skill decision model to improve the accuracy of the skill decision model.

[0139] The server may generate a service request based on the second skill group output by the decision model, and execute step 318: sending the user's service request to the second skill group.

[0140] In actual applications, service summary information, session information, related information, remarks information and other information can be included in the service request and sent to the second skill group, so that the second skill group can better solve the user's demands; service summary information, session information, related information and other information can also be sent to the second skill group before or after sending the service request.

[0141] The second skill group may obtain the user service request sent by the server, and the second skill group executes step 320: the second customer service in the second skill group conducts a conversation with the user.

[0142] In actual applications, if the first customer service can solve the user's problem, the relevant information of the conversation between the first customer service and the user can be recorded and stored, and the user will not be transferred to customer service. If the first customer service fails to solve the user's problem, but the second customer service successfully solves the user's problem, the relevant information of the conversation between the second customer service and the user and the relevant information of the conversation between the first customer service and the user can be recorded and stored. If the first customer service and the second customer service fail to solve the user's problem, they can continue to transfer the user and record the relevant information of the conversation between each customer service and the user for storage. If the user's problem is still not solved after the entire conversation ends, the relevant information of the conversation between each customer service and the user can also be recorded and stored. The server can record the solution of each customer service. For example, the first customer service completely solves; the first customer service solves one user problem: order refund problem, and the second customer service solves the remaining user problems: order query problem, account abnormality problem; the first customer service and the second customer service both solve the user's problem.

[0143] In actual applications, the text conversation information between the first customer service and the user and the corresponding service summary information can also be stored. If the user's problem is solved, the text conversation information and the corresponding service summary information can also be stored in other databases to facilitate subsequent data analysis, optimize the model, and improve the transfer accuracy and success rate of the model.

[0144] The order of some steps in the method described in one or more embodiments of this specification can be interchanged according to actual needs, or some steps can be omitted or deleted. For example, the first customer service can perform a transfer operation before extracting the service summary; omit or delete step 314, etc. For steps that are similar or identical to the above content, refer to the above content and will not be repeated here.

[0145] Through the above method, on the one hand, the summary extraction model can be used to conduct in-depth analysis of the customer's conversation content, extract key information, identify customer demands, and train through LoRA technology to improve the accuracy of the customer's real needs.

[0146] On the one hand, the skill decision model can integrate multiple input data, such as text conversation information, the first customer service's remarks, service summary information, and related information. By conducting a comprehensive analysis of multiple input data, it ensures that multiple aspects of information are considered during the transfer process, thereby reducing the probability of misidentification caused by a single information source.

[0147] On the one hand, the combination of LoRA technology and RAG technology can enable the server to continuously learn and update in the background, so that when the business line is adjusted, the design skill group is added or deleted, the customer service does not need to perceive these changes, reducing the cost of information transmission and understanding, and no longer requiring frequent manual training. At the same time, customer service can access relevant knowledge at any time in actual work, obtain instant feedback, improve service efficiency, and further reduce learning and training costs.

[0148] On the one hand, the skill decision model automatically identifies the appropriate skill group, reducing the time customer service spends on identifying and selecting skill groups. Customer service does not need to intervene manually, and the server automatically initiates the transfer, improving overall efficiency.

[0149] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method. Figure 4 This is a schematic diagram of the structure of a customer service transfer device provided in an embodiment of this specification. Figure 4 As shown, the device may include: The session information acquisition module 402 is used to acquire the session information between the user and the first customer service in the first skill group; A target information extraction module 404 is used to extract the session information to obtain target information; The transfer processing module 406 is used to provide the target information to the skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning the language model based on historical session information; The request sending module 408 is used to send the service request for the user to the second customer service in the second skill group.

[0150] based on Figure 4 The present specification also provides some specific implementation schemes of the method, which are described below.

[0151] Optionally, the device further includes a call center module, which can be specifically used for: Obtain a transfer request for the user submitted by the first customer service.

[0152] Optionally, the device further includes a knowledge base module, which can be specifically used for: Based on the RAG technology, the associated information matching the session information is retrieved from the knowledge base; the knowledge base contains at least one of the skill information of each skill group and the common problem information; the skill information includes at least one of the skill group name, the skill group service content, the skill group professional field, and the historical transfer cases handled by the skill group; the problem information includes the question inquiry information, the answer information corresponding to the question inquiry information, and the skill group information for processing the question inquiry information; Providing the target information to the skill decision model specifically includes: The target information and the associated information are provided to a skill decision model.

[0153] Optionally, the session information includes voice session information; and the device further includes a voice module, which can be specifically used to: The voice conversation information is converted into text conversation information.

[0154] Optionally, the device further includes a text processing module, which can be specifically used for: The service summary information of the text conversation information is extracted using a summary extraction model; the service summary information includes at least one of customer demand information, problem classification information, content summary information, customer emotion classification information, and information on whether the customer problem is solved.

[0155] Optionally, the device further includes a summary service extraction model training module, which can be specifically used for: Acquire historical conversation texts; annotate the historical conversation texts to obtain training data containing annotation information; the annotation information includes at least one of customer demand information, problem classification information, content summary information, sentiment classification information, and information on whether the problem is solved corresponding to the historical conversation texts; and train a pre-trained language model using the training data to obtain the summary extraction model.

[0156] Optionally, the device further includes a model packaging module, which can be specifically used for: Encapsulating the summary extraction model as an API interface; The extracting the service summary information of the text conversation information by using the summary extraction model specifically includes: calling the summary extraction model through the API interface; and extracting the service summary information of the text conversation information by using the summary extraction model.

[0157] Optionally, the device may further include a skill decision model fine-tuning module, which may be specifically used for: Acquire historical service record information; the historical service record information includes historical text conversation information converted from historical service voice conversation information; annotate the historical service record information to obtain training sample data containing annotated information; the annotated information includes customer demand information contained in the historical service record information and skill group information for processing the customer demand information; use the training sample data to fine-tune the pre-trained large language model to obtain the skill decision model.

[0158] Optionally, the training sample data also includes service summary information corresponding to the historical text conversation information.

[0159] Optionally, the device further includes an emotion information recognition module, which can be specifically used for: Determining emotional information of the user according to the session information; The providing the target information to the skill decision model to determine the second skill group that can handle the user's problem specifically includes: providing the target information to the skill decision model to obtain multiple second skill groups that can handle the user's problem; selecting a target second skill group from multiple second skill groups according to the degree of negative emotion represented by the emotional information; and the probability that the target second skill group can handle the problem is positively correlated with the degree of negative emotion.

[0160] Optionally, the device further includes a skill group determination module, which can be specifically used to: Determine the load level of the second skill group; if the load level of the second skill group is greater than or equal to a preset threshold, send the service request for the user to a third customer service representative in a third skill group; the probability that the third skill group can handle the problem is less than the probability that the second skill group can handle the problem; The sending the service request for the user to the second customer service in the second skill group specifically includes: if the load level of the second skill group is less than the preset threshold, sending the service request for the user to the second customer service in the second skill group.

[0161] Optionally, the device can also be used for: Acquire the remark information provided by the first customer service; the remark information includes the user demand information of the user written by the first customer service; Providing the target information to the skill decision model specifically includes: providing the target information and the remark information to the skill decision model.

[0162] Optionally, the device may also be used to: provide the service summary information to the second customer service.

[0163] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0164] Figure 5 This is a schematic diagram of the structure of a customer service transfer device provided in an embodiment of this specification. Figure 5 As shown, the device 500 may include: at least one processor 510; and, A memory 530 in communication with the at least one processor; wherein, The memory 530 stores instructions 520 executable by the at least one processor 510. The instructions are executed by the at least one processor 510 to enable the at least one processor 510 to: Obtaining the conversation information between the user and the first customer service representative in the first skill group; Extracting the session information to obtain target information; The target information is provided to a skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service representative; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning a language model based on historical session information; Sending a service request for the user to a second customer service representative in the second skill group.

[0165] Based on the same idea, the embodiment of this specification also provides a computer-readable medium corresponding to the above method. The computer-readable medium stores computer-readable instructions, and the computer-readable instructions can be executed by a processor to implement the above customer service transfer method.

[0166] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. Figure 5 As for the device shown, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0167] 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, stored data, displayed data, 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 relevant data must comply with relevant laws, regulations and standards in the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0168] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with hardware entity modules. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0169] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0170] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0171] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0172] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0176] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0177] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0178] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0179] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0180] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0181] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0182] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for customer service transfer, comprising: Obtaining the conversation information between the user and the first customer service representative in the first skill group; Extracting the session information to obtain target information; Providing the target information to a skill decision model to determine a second skill group capable of handling the user's problem; The second skill group includes at least one second customer service; The second skill group and the first skill group are used to handle different types of problems; The skill decision model is a model obtained by fine-tuning the language model based on historical conversation information; Sending a service request for the user to a second customer service representative in the second skill group.

2. The method according to claim 1, before providing the target information to the skill decision model, further comprising: Obtain a transfer request for the user submitted by the first customer service.

3. The method according to claim 1, further comprising: Retrieving association information matching the session information from a knowledge base based on the RAG technology; The knowledge base contains at least one of the skill information and common problem information of each skill group; the skill information includes at least one of the skill group name, skill group service content, skill group professional field, and historical transfer cases handled by the skill group; the problem information includes question inquiry information, answer information corresponding to the question inquiry information, and skill group information for processing the question inquiry information; Providing the target information to the skill decision model specifically includes: The target information and the associated information are provided to a skill decision model.

4. The method according to claim 1, wherein the session information includes voice session information; and extracting information from the session information to obtain target information specifically comprises: The voice conversation information is converted into text conversation information.

5. According to the method of claim 4, extracting the session information to obtain target information specifically comprises: Extracting service summary information of the text conversation information using a summary extraction model; The service summary information includes at least one of customer demand information, problem classification information, content summary information, customer emotion classification information, and information on whether the customer problem is solved.

6. The method according to claim 5, further comprising: Get the historical conversation text; Annotating the historical conversation text to obtain training data containing annotation information; The annotation information includes at least one of customer demand information, problem classification information, content summary information, sentiment classification information, and information on whether the problem is solved corresponding to the historical conversation text; The pre-trained language model is trained using the training data to obtain the summary extraction model.

7. The method according to claim 5, further comprising: Encapsulating the summary extraction model as an API interface; The extracting service summary information of the text conversation information by using a summary extraction model specifically includes: Calling the summary extraction model through the API interface; The service summary information of the text conversation information is extracted using a summary extraction model.

8. The method according to claim 1, further comprising: Get historical service record information; The historical service record information includes historical text conversation information converted from historical service voice conversation information; The historical service record information is annotated to obtain training sample data containing annotated information; the annotated information includes customer demand information contained in the historical service record information and skill group information for processing the customer demand information; The training sample data is used to perform fine-tuning training on the pre-trained large language model to obtain the skill decision model.

9. According to the method of claim 8, the training sample data also includes service summary information corresponding to the historical text conversation information.

10. The method according to claim 1, further comprising: Determining the emotion information of the user according to the session information; Providing the target information to a skill decision model to determine a second skill group capable of handling the user's problem specifically includes: Providing the target information to a skill decision model to obtain a plurality of second skill groups capable of handling the user's problem; According to the degree of negative emotion represented by the emotional information, a target second skill group is selected from multiple second skill groups; the probability that the target second skill group can handle the problem is positively correlated with the degree of negative emotion.

11. The method according to claim 1, further comprising: determining a load level of the second skill group; If the load level of the second skill group is greater than or equal to a preset threshold, sending the service request for the user to a third customer service in a third skill group; The probability that the third skill group can handle the problem is less than the probability that the second skill group can handle the problem; The sending the service request for the user to the second customer service in the second skill group specifically includes: If the load level of the second skill group is less than the preset threshold, the service request for the user is sent to the second customer service in the second skill group.

12. The method according to claim 1, further comprising: Obtaining the remark information provided by the first customer service; The remark information includes the user demand information of the user written by the first customer service; Providing the target information to the skill decision model specifically includes: The target information and the remark information are provided to a skill decision model.

13. The method according to claim 5, further comprising: The service summary information is provided to the second customer service.

14. A customer service transfer device, comprising: A session information acquisition module, used to acquire session information between a user and a first customer service representative in a first skill group; A target information extraction module, used to extract information from the session information to obtain target information; A transfer processing module, used for providing the target information to a skill decision model to determine a second skill group capable of handling the user's problem; The second skill group includes at least one second customer service; The second skill group and the first skill group are used to handle different types of problems; The skill decision model is a model obtained by fine-tuning the language model based on historical conversation information; A request sending module is used to send the service request for the user to the second customer service in the second skill group.

15. A customer service transfer device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtaining the conversation information between the user and the first customer service representative in the first skill group; Extracting the session information to obtain target information; The target information is provided to a skill decision model to determine a second skill group that can handle the user's problem; the second skill group includes at least one second customer service representative; the second skill group and the first skill group are used to handle different types of problems; the skill decision model is a model obtained by fine-tuning a language model based on historical session information; Sending a service request for the user to a second customer service representative in the second skill group.

16. A computer-readable medium having computer-readable instructions stored thereon, wherein the computer-readable instructions can be executed by a processor to implement the customer service transfer method according to any one of claims 1 to 13.

Images