Customer service allocation method and device, storage medium and program product

Abstract

The embodiment of the invention provides a customer service allocation method and device, a storage medium and a program product. According to the embodiment of the invention, the method comprises the steps: locking a first target service field from a plurality of service fields according to the multi-dimensional demand information of a target user, and obtaining the first target service field matched with first problem information; furthermore, in a first target service field, service score evaluation is performed on each AI customer service in the field based on information such as knowledge matching degree and service history, and a first target AI customer service is selected from a plurality of AI customer services, so that the first target AI customer service provides customer service for a target user. A two-stage strategy of first customer service shunting and then target customer service selection is adopted, and a candidate scale is reduced from a global state to a local state, so that the calculation amount and the response delay are remarkably reduced; and the target AI customer service is selected from the target service field adapted to the target user demand, so that the service quality and the service efficiency of the customer service are improved, and the use experience of the user is improved.

Description

Technical Field

[0001] This application relates to the field of computer processing technology, and in particular to a customer service assignment method, device, storage medium, and program product. Background Technology

[0002] With the widespread adoption of internet services, instant messaging applications such as chat tools (e.g., WeChat Work) are widely used as enterprise-level communication and customer service platforms. Traditional chat tools typically use simple polling or random assignment methods to assign customer service representatives to users. However, the services provided by customer service representatives assigned in this way may not match the actual needs of users and may be difficult to adapt to complex and ever-changing business requirements, resulting in poor customer service quality, low service efficiency, and ultimately impacting the user experience. Summary of the Invention

[0003] This application provides a customer service allocation method, device, storage medium, and program product to improve the adaptability of the services provided by the allocated customer service representatives to the actual needs of users, enhance the adaptability of customer service to business requirements, thereby improving customer service quality and efficiency, and enhancing the user experience.

[0004] This application provides a customer service allocation method applied to the server of a target application. The server maintains multiple service domains. The method includes: receiving first question information submitted by a target user in a current session, and obtaining business scenario information of the current session and user information of the target user; generating first multi-dimensional demand information corresponding to the target user based on the user information, the business scenario information, and the first question information; determining a first target service domain adapted to the first question information from the multiple service domains based on the first multi-dimensional demand information, the first target service domain including multiple AI customer service representatives; calculating a service score corresponding to each AI customer service representative based on the first multi-dimensional demand information, the knowledge information of the multiple AI customer service representatives, and attribute information on whether they have provided customer service to the target user, the service score reflecting the suitability of each AI customer service representative to provide customer service to the target user; selecting a first target AI customer service representative from the multiple AI customer service representatives based on the service scores of each AI customer service representative; and inputting the first question information and subsequent question information into the first target AI customer service representative so that the first target AI customer service representative can provide customer service to the target user.

[0005] This application also provides an electronic device, including: a memory and a processor; the memory for storing a computer program; and the processor, coupled to the memory, for executing the computer program to implement the steps in the methods described above.

[0006] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps in the methods described above.

[0007] This application also provides a computer program product, which includes a computer program / instructions that, when executed by a processor, enable the processor to implement the steps in the methods described above.

[0008] In this embodiment, a first target service domain is identified from multiple service domains based on the multi-dimensional needs information of the target user, thus completing intelligent customer service domain routing and obtaining a first target service domain that matches the first question information. Further, within the first target service domain, the services corresponding to each AI customer service representative in that domain are quantitatively scored based on information such as knowledge matching degree and service history. Based on the service score obtained from the quantitative scoring, a first target AI customer service representative is selected from multiple AI customer service representatives to provide customer service to the target user. Therefore, this two-level strategy of first routing customer service and then selecting a target customer service representative reduces the candidate scale from global to local, significantly reducing computational load and response latency. Furthermore, selecting a target customer service representative from the service domain that matches the multi-dimensional needs information of the target user after routing improves the adaptability of the services provided by the assigned customer service representatives to the user's actual needs, enhances the adaptability of customer service to business requirements, thereby improving customer service quality and efficiency, and enhancing the user experience. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0010] Figure 1 A flowchart illustrating a customer service allocation method provided for an exemplary embodiment of this application;

[0011] Figure 2a A schematic diagram of customer service traffic triage provided for an exemplary embodiment of this application;

[0012] Figure 2b A schematic diagram of the workflow of a multi-dimensional requirement model provided for an exemplary embodiment of this application;

[0013] Figure 2c A schematic diagram illustrating the working principle of an intent recognition model provided in an exemplary embodiment of this application;

[0014] Figure 2d A schematic diagram illustrating the working principle of the sentiment analysis model provided in an exemplary embodiment of this application;

[0015] Figure 2e A schematic diagram illustrating the working principle of the entity extraction model provided in an exemplary embodiment of this application;

[0016] Figure 2f A schematic diagram illustrating the working principle of a user profile analysis model provided in an exemplary embodiment of this application;

[0017] Figure 3 A schematic diagram of the structure of an electronic device provided for an exemplary embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0020] Building upon the background technology, this application addresses the technical problem that the services provided by customer service representatives assigned in the manner described in the background technology may not match the actual needs of users and may be difficult to adapt to complex and ever-changing business requirements, resulting in poor customer service quality and low service efficiency, thus affecting the user experience. In this embodiment, a first target service domain is identified from multiple service domains based on the multi-dimensional needs information of the target user, and intelligent customer service domain routing is completed to obtain a first target service domain that matches the first problem information. Furthermore, within the first target service domain, the services corresponding to each AI customer service representative in that domain are quantitatively scored based on information such as knowledge matching degree and service history. Based on the service score obtained from the quantitative scoring, a first target AI customer service representative is selected from multiple AI customer service representatives to provide customer service to the target user. Therefore, the two-tiered strategy of first triaging customer service and then selecting target customer service representatives reduces the candidate pool from global to local, significantly reducing computational load and response latency. Furthermore, by selecting target customer service representatives from service areas that match the multi-dimensional needs of the target users after triage, the strategy improves the matching degree between the services provided by the assigned customer service representatives and the actual needs of users, enhances the adaptability of customer service to business requirements, thereby improving the quality and efficiency of customer service and enhancing the user experience.

[0021] The following describes in detail a solution provided by an embodiment of this application, with reference to the accompanying drawings.

[0022] An exemplary embodiment of this application provides a customer service assignment system, which includes: a user terminal, an instant messaging application, a server for the instant messaging application, and a server for the target application.

[0023] In this context, the user terminal is the terminal used by the target user, and an instant messaging application refers to an application that allows two or more people to exchange messages in real time over a network. These applications support users sending text messages, pictures, videos, voice messages, and file transfers, and typically also provide voice and video call functionality. Examples of instant messaging applications include chat or conversation tools such as WeChat and Weibo.

[0024] This embodiment does not limit the implementation form of the user terminal. For example, the user terminal can be a smartphone, tablet computer, laptop computer, or desktop computer, etc.; another example is that the user terminal can also be a smart wearable device, such as a smartwatch, smart bracelet, etc.; yet another example is that the user terminal can also be various smart home appliances with display screens, such as smart TVs, smart large screens, or smart robots, etc. In addition, various applications can be installed on the user terminal. Applications can be standalone apps, mini-programs that depend on standalone apps, or web pages; this embodiment does not limit this. Among them, applications include, but are not limited to: instant messaging applications and target applications. Instant messaging applications can serve as chat or conversation tools for target applications to conduct conversations with target users based on instant messaging applications. Among them, target applications can be service applications, rental applications, shopping applications, etc.; this embodiment does not limit this.

[0025] This embodiment does not limit the implementation form of the instant messaging application's server and the target application's server. The server can be a single physical server, a cloud server, or a server array. The instant messaging application's server and the target application's server can have the same implementation form or different implementation forms.

[0026] In this embodiment, a communication connection can be established between the user terminal, the server of the instant messaging application, and the server of the target application. The server of the instant messaging application can act as a communication medium between the user terminal and the server of the target application. Specifically, the user terminal can establish a communication connection with the server of the instant messaging application by logging into the local instant messaging application to send information to the server of the instant messaging application. The server of the instant messaging application can then send the information received from the user terminal to the server of the target application through the communication connection. This allows the server of the target application to execute a customer service allocation method based on the information sent by the instant messaging application. The customer service allocation method is described in [reference needed]. Figure 1 The relevant descriptions of the embodiments will not be repeated here.

[0027] Figure 1 A customer service allocation method is provided as an exemplary embodiment of this application. This customer service allocation method is applied to the server of a target application, which maintains multiple service domains, such as… Figure 1 As shown, the customer service assignment method includes:

[0028] 101. Receive the first question information submitted by the target user in this session, and obtain the business scenario information to which this session belongs and the user information of the target user;

[0029] 102. Based on the target user's user information, business scenario information, and primary problem information, generate the primary multi-dimensional requirement information corresponding to the target user;

[0030] 103. Based on the first multi-dimensional demand information, determine the first target service area that matches the first problem information from multiple service areas. The first target service area includes multiple AI customer service.

[0031] 104. Based on the first multi-dimensional demand information, the knowledge information of multiple AI customer service representatives, and the attribute information of whether they have provided customer service to the target user, calculate the service score corresponding to each AI customer service representative. The service score reflects the suitability of each AI customer service representative to provide customer service to the target user.

[0032] 105. Based on the service scores of each AI customer service representative, select the first target AI customer service representative from among multiple AI customer service representatives;

[0033] 106. Input the first problem information and subsequent problem information into the first target AI customer service so that the first target AI customer service can provide customer service to the target user.

[0034] In this embodiment, each service area includes multiple AI customer service representatives. AI customer service refers to customer service representatives who provide services to target users based on artificial intelligence models. These multiple AI customer service representatives encompass different customer service types, such as... Figure 2a As shown, AI customer service representatives from various fields are stored in an AI customer service resource pool. AI customer service types include, but are not limited to: pre-sales AI customer service, after-sales AI customer service, complaint AI customer service, and other AI customer service. The specific services provided by different types of AI customer service representatives are determined by the knowledge base of each AI customer service representative, which guides the specific services provided. AI customer service representatives of the same type may provide the same or different services, for example, with different service focuses, also determined by the knowledge base of each AI customer service representative. For example, some pre-sales customer service representatives may focus on answering users' difficult questions, while others may focus on recommending products to users. Furthermore, the artificial intelligence model can be obtained by training an initial artificial intelligence model based on training samples from the customer service service field, or by fine-tuning training samples from the customer service service field. Other models involved in the following embodiments can also be obtained by training an initial model or fine-tuning a mature model based on training samples from the corresponding field. Each model can be a large model with parameters meeting the set parameter quantity requirements. The parameter quantity requirements are not specifically limited and can have different definitions in different scenarios or fields. For example, in some scenarios or domains, a large model refers to a model with a parameter scale of tens of billions, hundreds of billions, or even trillions. However, the definition of the number of parameters in this model is merely an example and does not constitute a limitation on the technical solution of this application. In addition, a large model can be an AI model, such as an AI large language model with powerful language understanding and generation capabilities.

[0035] In this embodiment, a session request sent by the target application can be received. The session request includes the user information of the target application and the business scenario information to which the current session belongs. The target application establishes a session interface with the target user for the current session, and associates the session interface with the business scenario information and the user information of the target user, laying the foundation for obtaining the business scenario information to which the current session belongs and the user information of the target user in the future.

[0036] In this embodiment of the application, the target user can submit problem information during the current session, which can be referred to as the first problem information for ease of distinction and description. Furthermore, the business scenario information to which the current session belongs and the target user's user information can also be obtained.

[0037] This embodiment does not limit the implementation method of the target user submitting problem information and obtaining the business scenario information and user information of the target user in this session. For example, in response to the target user's trigger operation, a session interface with the target user is displayed. The trigger operation can be triggered by the target user or automatically set by the system. After the session interface associated with the target user is determined, the system will automatically initiate the trigger operation, but it is not limited to this. In addition, before responding to the trigger operation, the target user can log in to the instant messaging application based on their account information. Through the application page of the instant messaging application, advertising information can be displayed to the target user. This advertising information can be the products or services provided by the target application. The advertising information belongs to business scenario information, including but not limited to: business identifier / business scenario identifier, business details information, and the enterprise account information of the instant messaging application associated with the target application to which the advertising information belongs. The business details information can be the introduction information of the products or services provided. The enterprise account information of the instant messaging application can be, for example, the enterprise WeChat account information in WeChat. Furthermore, the advertising information can be identified through the instant messaging application to establish a session connection between the target user's terminal and the server of the target application, and in the process of establishing the session connection, the user information and business scenario information of the target user can be obtained. The target user's information includes, but is not limited to: the target user's instant messaging application user account information, basic personal information, geolocation information, and device information. User account information includes, but is not limited to: user account, historical behavior data within the target application, coupons and balance within the target application. Historical behavior data includes, but is not limited to: historical order data, historical browsing data, and historical consultation data. Basic personal information includes, but is not limited to: age, gender, occupation, educational background, marital status, income level, and appearance. Geolocation information includes, but is not limited to: the target user's city, region, and permanent address. Device information includes, but is not limited to: the user's device type, device model, and operating system version.

[0038] In this embodiment, during the process of identifying advertising information, a consultation request is sent to the server of the target application. The consultation request includes user information and business scenario information. When the server of the target application receives the consultation request initiated by the target user through the identification operation of advertising information, it can determine whether the target user's terminal and the server of the target application have established a session connection based on the consultation request. That is, it can determine whether the enterprise account has been added based on the business account information in the business scenario information contained in the consultation request. If it has been added, it is determined that a session connection has been established; if it has not been added, it is determined that a session connection has not been established.

[0039] In an optional embodiment, if it is determined that the session connection has not been established, in response to the consultation request, a session connection between the target user's terminal device and the server is established through the target application based on user information and business scenario information, and a session interface with the target user is created; the user information, business scenario information and session interface are associated.

[0040] In practical applications, advertising information can be implemented as an advertisement page for a product or service. This page contains business details and an identification code (such as a QR code). The identification code recognition function of the instant messaging application can identify this code to obtain the business scenario information and user information related to the corresponding product or service. Taking a moving scenario as an example, the product or service information in the business details could be relevant information from a promotional page for a moving product or service. After obtaining the business scenario information and user information, if no session connection has been established between the target user's terminal and the target application's server, a session connection is established between the target user's instant messaging application account and the enterprise account based on the business account information in the business scenario information. This establishes a session connection between the target user's terminal and the target application's server, enabling the enterprise account to provide services to the target user.

[0041] In another alternative embodiment, if it is determined that a session connection has been established, the session interface associated with the target user can be determined in response to the consultation request based on user information and business scenario information. For example, a session interface identifier associated with the target user can be determined based on user information and business scenario information, and the corresponding session interface can be determined based on the session interface identifier.

[0042] In practical applications, taking WeChat as an example of an instant messaging application, if a session connection has been pre-established, the target application's WeChat account page (such as the homepage) can display an online customer service control. First, in response to an inquiry request, the session interface associated with the target user is determined based on user information and business scenario information. Then, in response to the target user's triggering action on the online customer service control, that session interface is displayed. If a session connection has not been pre-established, it can be established using the method described in the above embodiments. Afterward, the online customer service control can be displayed on the target application's WeChat account page (such as the homepage), and in response to triggering actions on that control, a session interface with the target user is displayed.

[0043] In this embodiment, after obtaining the business scenario information of the current session, the user information of the target user, and the first question information, first multi-dimensional demand information corresponding to the target user can be generated based on the user information, business scenario information, and first question information. The first multi-dimensional demand information represents user needs from multiple dimensions, including but not limited to: target user intent, emotional level, entity information, and profile information. Specifically, the target user intent can represent the service domain or session type of the user session, and target user intents correspond to different categories (intent classifications), including but not limited to consultation (such as product information, price, service, etc.), complaint (such as complaints about quality, service, after-sales, etc.), suggestion, and other categories. Figure 2d As shown, emotional levels correspond to emotional tags, which include emotional type tags and urgency level tags. Emotional type tags include, but are not limited to, positive, negative, and neutral. Urgency level tags include, but are not limited to, low, medium, and high. Entity information (also known as key entities) refers to key information in the dialogue. Entity information can be, for example, a product name, service type, or specific question. Profile information refers to a dynamic data set constructed by collecting, integrating, and analyzing users' historical behavioral data, interests, preferences, and value contributions, etc., to systematically describe user characteristics, predict needs, and support decision-making. Profile information includes, but is not limited to, historical behavior analysis information, interest point identification information, and user value assessment information. User value assessment information refers to the economic value level (such as high-net-worth / potential users) and non-economic value contribution (such as activity level and reputation) quantified by comprehensively considering indicators such as user lifetime value, willingness to pay, and social influence. By coupling user information, real-time business scenarios, and problem information, a first multi-dimensional demand information covering intent, emotion, entity, and long-term preferences is extracted. This multi-dimensional demand information provides accurate reference information for the subsequent two-level triage of domain-first and customer service-second, reducing the candidate range from global to local and reducing invalid matching calculations. On the other hand, it makes implicit demands explicit, which facilitates the selection of target AI customer service in the future.

[0044] This application does not limit the implementation method of generating multi-dimensional demand information. For example, multi-dimensional demand analysis can be performed based on semantic information of the target user's user information, business scenario information, and first question information. Another example is that user profile analysis can be performed based on the target user's user information, business scenario information, and first question information, and the first multi-dimensional demand information can be determined based on the profile information obtained from the analysis. Yet another example is that semantic information can be input into a multi-dimensional demand analysis model, which can adopt a dual-branch parallel structure. This dual-branch parallel structure can include a conversation branch and a profile branch. The conversation branch consists of a bidirectional memory network (Bi-LSTM) and a self-attention mechanism to process the current dialogue context, while the profile branch consists of a Transformer encoder to process the target user's long-term behavioral sequence. The conversation branch outputs the current conversation demand triples from the current conversation dimension, and the triples are concatenated to form the first current response demand information vector. The triples include intent category (e.g., inquiry / complaint / purchase) and sentiment polarity. (Positive / Neutral / Negative), Key Entity List (product, fault, time, etc.); The profiling branch analyzes from the perspective of normalized needs, taking the user's historical interaction sequence as input (read in real time via user ID), and captures long-term preferences through self-attention, thereby outputting the demand intensity distribution and value stratification labels. The demand intensity distribution and value stratification labels together constitute the first profile information vector, where the demand intensity distribution represents the tendency weight of pre-sales, after-sales, value-added, and care needs, and the value stratification labels include, but are not limited to, high / medium / low value labels; The conversation demand vector and the profile information vector are fused and concatenated to obtain the first multi-dimensional demand information based on the target user's user information, business scenario information, and first question information. A more preferred implementation method for generating the first multi-dimensional demand information corresponding to the target user can be found in the relevant description of the following embodiments, which will not be repeated here.

[0045] In this embodiment, after obtaining the first multi-dimensional demand information, a first target service domain matching the first problem information can be determined from multiple service domains based on the first multi-dimensional demand information. The first target service domain includes multiple AI customer service representatives. Therefore, by accurately identifying the matching service domain through multi-dimensional demand information, the candidate range of AI customer service representatives is narrowed, which helps reduce matching time and error rate, thereby improving matching efficiency and accuracy. Figure 2a As shown, for the specific implementation of determining the first target service domain that matches the first problem information from multiple service domains based on the first multi-dimensional demand information, please refer to the relevant description of the following embodiments, which will not be repeated here.

[0046] Furthermore, after determining the primary target service area, target AI customer service representatives can be selected from this area to provide services to the target user. Since factors such as the first multi-dimensional demand information, the knowledge information of multiple AI customer service representatives, and their prior experience providing customer service to the target user influence the service quality of the AI ​​customer service representatives, a service score can be calculated for each AI customer service representative based on these factors. This allows for the selection of the target customer service representative based on their service scores within the primary target service area. The service score reflects the suitability of each AI customer service representative for providing customer service to the target user. The knowledge information of the multiple AI customer service representatives can be provided by their respective knowledge bases. Therefore, by comprehensively considering multi-dimensional demand information, knowledge matching, and quantitative scoring of service history, the suitability of each AI customer service representative becomes readily apparent, accurately identifying the most suitable representative for the target user, reducing trial-and-error transfers, and improving service continuity and user satisfaction. For the first multi-dimensional demand information, the knowledge information of multiple AI customer service representatives, and the attribute information of whether customer service has been provided to the target user, the specific implementation method for calculating the service score corresponding to each AI customer service representative can be found in the relevant description of the following embodiments, and will not be repeated here.

[0047] Furthermore, after obtaining the service scores for each AI customer service representative, a primary target AI customer service representative can be selected from multiple AI customer service representatives based on their service scores. Thus, by selecting the primary target AI customer service representative based on service scores, the most suitable resource is directly locked in, avoiding mismatches caused by manual screening and random assignment, shortening the response chain, and improving selection efficiency and service experience. For the specific implementation method of selecting the primary target AI customer service representative from multiple AI customer service representatives based on their service scores, please refer to the relevant description in the following embodiments, which will not be repeated here.

[0048] Furthermore, after obtaining the first target AI customer service representative, the first question information and subsequent question information can be input into the first target AI customer service representative so that the first target AI customer service representative can provide customer service to the target user.

[0049] The technical solutions provided in the above embodiments of this application lock a first target service domain from multiple service domains based on the multi-dimensional needs information of the target user, and complete the intelligent diversion of customer service domains to obtain a first target service domain that matches the first problem information. Further, within the first target service domain, the services corresponding to each AI customer service representative in that domain are quantitatively scored based on information such as knowledge matching degree and service history. Based on the service score obtained from the quantitative scoring, a first target AI customer service representative is selected from multiple AI customer service representatives to provide customer service to the target user. Thus, this two-level strategy of first diverting customer service and then selecting a target customer service representative reduces the candidate scale from global to local, significantly reducing computational load and response latency. Furthermore, selecting a target customer service representative from the diverted service domain that matches the multi-dimensional needs information of the target user improves the adaptability of the services provided by the assigned customer service representatives to the user's actual needs, enhances the adaptability of customer service to business requirements, thereby improving customer service quality and efficiency, and enhancing the user experience.

[0050] In this embodiment, the multiple AI customer service representatives in the first target service domain may include historical AI customer service representatives who have provided services to the target user. If such historical AI customer service representatives belong to the first target service domain, their knowledge base will be richer, enabling them to provide more comprehensive services to the target user and thus improving the user's service experience. Based on this, the selection of the first target AI customer service representative from among the multiple AI customer service representatives, according to the service scores corresponding to each AI customer service representative, includes: identifying whether the multiple AI customer service representatives included in the first target AI service domain include at least one historical AI customer service representative who has provided services to the target user; if so, attempting to select an AI customer service representative whose service score meets a first preset condition from the at least one historical AI customer service representative as the first target AI customer service representative; and if no historical AI customer service representative whose service score meets the first preset condition is selected from the at least one historical AI customer service representative, selecting a non-historical AI customer service representative whose service score meets a second preset condition from among the non-historical AI customer service representatives included in the multiple AI customer service representatives as the first target AI customer service representative; if not, selecting an AI customer service representative whose service score meets a third preset condition from among the multiple AI customer service representatives as the first target AI customer service representative. The first, second, and third preset conditions may each include a service score, allowing for the selection of AI customer service representatives based on service scores. Furthermore, since historical AI customer service representatives have provided services to the target user, when determining whether a historical AI customer service representative can serve as the target customer service representative, the service score included in the first preset condition can be greater than or equal to the service scores included in the second and third preset conditions, thereby enhancing the reliability of historical AI customer service representatives. The service scores included in the second preset condition and the third preset condition can be the same or different. This embodiment first identifies historical AI customer service representatives, prioritizing the reuse of those who have served the target user, leveraging their existing understanding of user needs and a more complete domain knowledge base to achieve service continuity and depth enhancement; if a historical AI customer service representative does not meet the conditions, a high-scoring non-historical customer service representative is selected as a second choice, ensuring that suitable resources are always available. The entire process is further refined after domain screening, taking into account familiarity and capability matching, reducing the cost of repeated communication, and improving user satisfaction.

[0051] In this embodiment, when it is identified that among the multiple AI customer service representatives included in the first target AI service domain, there is no at least one historical AI customer service representative who has provided services to the target user, or the at least one historical AI customer service representative is unable to provide accurate customer service to the target user, the specific explanation of selecting the target AI customer service representative from among the multiple AI customer service representatives included in the first target service domain is as follows.

[0052] In this embodiment, based on the target user's user information, business scenario information, and first question information, the first multi-dimensional demand information corresponding to the target user is generated. This includes: performing semantic understanding on the target user's user information, business scenario information, and first question information to obtain semantic information; inputting the semantic information into a multi-dimensional demand analysis model to perform demand analysis on the target user from the dimensions of the current session and normalized demand to obtain the first current session demand information and the first profile information corresponding to the target user, which serve as the first multi-dimensional demand information. The first profile information reflects the target user's normalized demand information. Normalized demand information refers to the relatively stable and predictable service demand pattern repeatedly exhibited by the target user within a continuous observation period (such as the past 30 days or longer), including the intensity of inclination towards specific business areas (pre-sales consultation, after-sales support, value-added services, etc.), preference for customer service interaction methods, and long-term focus on product / service functions. This normalized demand information is extracted from historical interaction data by the user profile analysis model and serves as a long-term demand baseline independent of a single session, used to combine with the current session demand to form a more complete multi-dimensional demand information. It should be noted that the first portrait information can be regarded as the portrait information in the above embodiments.

[0053] In some optional embodiments, semantic understanding is performed on the target user's user information, business scenario information, and first question information to obtain semantic information. This includes: preprocessing the three types of original text—user information, business scenario information, and first question information—to form a unified lexical space, such as word segmentation, stop word removal, part-of-speech tagging, and entity recognition (or stemming); concatenating the processed text into a context sequence, and outputting a multi-dimensional context embedding vector through a Bidirectional Encoder Representations from Transformers (BERT) or RoBERTa pre-trained model; weighting and concatenating the context vector with the user profile embedding and business scenario label vector at the feature layer to obtain a comprehensive semantic vector; and reducing the dimensionality of the comprehensive semantic vector through a lightweight fully connected network and mapping it to a fixed-dimensional semantic information representation for direct use by downstream multi-dimensional requirement analysis models. In this embodiment, the three types of original text are uniformly mapped to high-quality, low-dimensional semantic information. On the one hand, the pre-trained model can fully capture deep semantics and eliminate ambiguity; on the other hand, the weighted concatenation ensures the dynamic integration of user profiles and scenario information, guaranteeing the completeness and personalization of requirements. In addition, after dimensionality reduction by the lightweight network, the downstream model can be directly reused, significantly reducing the amount of computation and improving the accuracy of semantic understanding and the efficiency of demand analysis. This helps to improve the accuracy of identifying target AI customer service, and improve the service quality and user experience of target AI customer service.

[0054] In some alternative embodiments, such as Figure 2b As shown, the multi-dimensional demand analysis model includes: an intent recognition model, a sentiment analysis model, an entity extraction model, and a user profile analysis model. Each model within the multi-dimensional demand analysis model can also be considered a network layer of the multi-dimensional demand analysis model. Semantic information is input into the multi-dimensional demand analysis model to analyze the target user's demands from both the current session dimension and the normalized demand dimension, obtaining the target user's first current session demand information and first profile information. This includes: inputting semantic information into the intent recognition model to identify the intent category of the current session based on the semantic information, obtaining target intent information as a first current session demand information; inputting semantic information into the sentiment analysis model to analyze the target user's sentiment category in the current session based on the semantic information, obtaining target sentiment information as a first current session demand information; inputting semantic information into the entity extraction model to extract target entity information related to the current session from the question information based on the semantic information and entity information templates, as a first current session demand information; and inputting semantic information into the user profile analysis model to analyze the target user's user profile based on the semantic information, obtaining the target user's first profile information. By leveraging multiple models within a multi-dimensional demand analysis model, user needs can be analyzed in parallel from different dimensions. This allows for the simultaneous output of session-level and routine demands. The intent recognition model identifies business intent, the sentiment analysis model quantifies emotional fluctuations, the entity extraction model accurately extracts key elements, and the user profiling analysis model analyzes long-term preferences. These four complementary approaches form a high-resolution, two-dimensional demand profile, which can be directly reused in subsequent traffic routing processes. This eliminates the delays and errors associated with traditional manual annotation or rule configuration, enabling accurate understanding of needs, domain targeting, and customer service matching. Consequently, it significantly reduces transfer rates and improves the accuracy and efficiency of customer service.

[0055] Optionally, such as Figure 2cAs shown, semantic information is input into the intent recognition model. Based on the semantic information, the intent category of the current conversation is identified to obtain target intent information. This includes: inputting semantic information into the intent recognition model; performing key information extraction based on the semantic information to obtain key semantic information, which is the key semantic information corresponding to the keywords in the initial dialogue content surrounding the first question; performing intent classification based on the key semantic information to determine the target intent type (or intent classification) suitable for the current round of dialogue from multiple known intent types; and performing intent recognition under the target intent type to identify the target intent information suitable for the current round of dialogue. For example, intent types include, but are not limited to, consultation (such as product information, price, service, etc.), complaint (such as complaints about quality, service, after-sales, etc.), suggestion, and other types. In other words, known intent types include, but are not limited to: consultation intent, purchase intent, complaint intent, feedback intent, and other intents. Among them, consultation intent refers to the target user's desire to obtain information or answer questions, such as "What is the price of the product?" or "How do I apply for the service?"; purchase intent refers to the target user's clear purchase need, such as "I want to buy this product"; complaint intent refers to the target user's dissatisfaction with the product or service and their desire to express dissatisfaction or request a solution, such as "I am not satisfied with this service"; feedback intent refers to the target user's suggestion or feedback on the product or service, such as "I think this feature can be improved"; other intents refer to intents that do not fall into the above explicit categories, such as casual conversation or unexpressed intent.

[0056] Specifically, based on key semantic information, an intent classification operation is performed to determine the target intent type that is suitable for the current round of dialogue from multiple known intent types. Optional implementation methods include: extracting semantic information of each known intent type; calculating a first similarity between each key semantic information and the semantic information of each target intent type; if there is a first target similarity greater than the first similarity threshold, at least one intent type corresponding to the first target similarity is taken as the target intent type.

[0057] Specifically, under the target intent type, an intent recognition operation is performed to identify the target intent that is compatible with the current round of dialogue under the target intent type. Optional implementation methods include: extracting the semantic information of each known intent under the target intent type to obtain the semantic information of each intent under each intent type; calculating the second similarity between each key semantic information and the semantic information of each intent under each intent type; if there is a target similarity greater than the second similarity threshold, at least one intent corresponding to the target similarity is taken as the target intent information.

[0058] Optionally, such as Figure 2dAs shown, semantic information is input into the sentiment analysis model, and the target user's sentiment category in this conversation is analyzed based on the semantic information to obtain target sentiment information. This includes: inputting semantic information into the sentiment analysis model, the model extracting features and classifying the input data through its internal neural network structure to obtain a sentiment level probability distribution, which represents the probability that the dialogue content input by the target user belongs to different sentiment level categories; and selecting the sentiment level category whose probability value meets the sentiment level threshold as the target user's target sentiment information according to the preset sentiment level category and probability threshold. Emotional intensity categories include, but are not limited to, the following: very positive, positive, neutral, negative, very negative, and the level of need for the product or service (high, medium, low). "Very positive" refers to users expressing extreme satisfaction, excitement, or high appreciation, such as "Fantastic!" or "Very satisfied!"; "Positive" refers to users expressing satisfaction, happiness, or positive emotions, such as "Not bad!" or "Pretty good!"; "Neutral" refers to users expressing neutral emotions without a clear emotional bias, such as "It's okay." or "It's alright."; "Negative" refers to users expressing dissatisfaction, disappointment, or negative emotions, such as "Not so good." or "Not very satisfied."; "Very negative" refers to users expressing extreme dissatisfaction, anger, or extremely negative emotions, such as "Terrible!" or "Very angry!".

[0059] Optionally, such as Figure 2eAs shown, semantic information is input into the entity extraction model. Based on the semantic information and entity information template, the model extracts target entity information related to the current conversation from the question information. This includes: inputting semantic information, user profile information, and business scenario information into the key information extraction model; integrating the semantic information, user profile information, and business scenario information to form a unified input data structure; performing word segmentation on the integrated input data, splitting the text data into word or phrase sequences; performing part-of-speech tagging and dependency parsing on the segmented data to extract the syntactic structure information of the text; performing vectorization on the processed data, converting the text data into a numerical vector form that the model can process, specifically including word embedding, Bidirectional Encoder Representations from Transformers (BERT) or other natural language processing techniques; performing semantic extraction and keyword recognition on the numerical vector data to obtain a set of keywords or key phrases representing important information in the user input text; and filtering the keywords or key phrases according to the preset key information template to obtain target keywords, which are then used as target entity information. The key information template is a predefined list of keywords or rules used to guide the model in extracting keywords related to the business scenario. For example, for different business scenario templates, in a pre-sales consultation scenario, the key information template might include keywords such as "product model," "price," and "function"; in an after-sales service scenario, the key information template might include keywords such as "fault description," "warranty period," and "repair method." User profile templates: Based on interest tags and behavioral preference tags in the user profile information, the key information template is dynamically adjusted. For example, for a "technology enthusiast" user, the key information template might include keywords such as "technical parameters" and "latest features." The primary target key information includes, but is not limited to, the following types: business-related keywords, user need keywords, emotional keywords, and behavioral keywords. Among them, business-related keywords refer to keywords directly related to the user's current business scenario, such as "product model," "price," and "fault description"; user need keywords refer to keywords reflecting the user's specific needs, such as "purchase," "consultation," and "complaint"; emotional keywords refer to keywords reflecting the user's emotional state, such as "satisfied," "dissatisfied," and "angry"; and behavioral keywords refer to keywords reflecting the user's behavioral patterns, such as "login," "purchase," and "browse."

[0060] Optionally, such as Figure 2fAs shown, semantic information is input into the user profile analysis model, and the user profile of the target user is analyzed based on the semantic information to obtain the first profile information of the target user. This includes: inputting semantic information into the profile analysis model, using the user's unique identifier as an index, retrieving the target user's interaction records (including dialogue text, click behavior, purchase records, and complaint information) from the distributed cache Redis within a specified time period to form a time series; the user profile analysis model consists of two layers of Transformer Encoders (such as a hidden dimension of 256) to process the above time series, capture long-term behavioral data, and introduce a multi-head attention mechanism to weight and fuse the current semantic vector with historical behavioral features; furthermore, historical behavior analysis, interest point identification, and demand priority evaluation are performed. Specifically, historical behavior analysis is performed using Transformer, and the output global representation is used to map the behavior sequence to a predefined label set through a clustering algorithm (K-means) to generate behavior labels, such as high-frequency after-sales repair requests and low-frequency price inquiries. Based on attention weights, the most relevant historical behavior segments to the current semantics are extracted for interest point identification, resulting in interest labels, such as smart home and extended warranty services. The RFM model is used to calculate user value stratification for value assessment, resulting in value assessment results, including but not limited to high value, medium value, and low value. The semantic vector and historical profile are combined to assess demand priority, resulting in a multi-dimensional demand weight vector (such as pre-sales, after-sales, value-added services, and care), which is then normalized, such as to the [0,1] interval.

[0061] In this embodiment, based on the first multi-dimensional demand information, the knowledge information of multiple AI customer service representatives, and the attribute information of whether they have provided customer service to the target user, the service score corresponding to each AI customer service representative is calculated. This includes: calculating the confidence score of each AI customer service representative based on the first multi-dimensional demand information and the knowledge information of each AI customer service representative, where the confidence score reflects whether the AI ​​customer service representative is suitable to provide customer service to the target user; calculating the suitability score of each AI customer service representative based on the first multi-dimensional demand information and the attribute information of whether each AI customer service representative has provided customer service to the target user, where the suitability score reflects the degree to which the AI ​​customer service representative is suitable to provide customer service to the target user; and calculating the service score of each AI customer service representative based on their confidence score and suitability score. By comprehensively evaluating confidence and suitability to calculate the service score, this method achieves multiple technical effects, including accurate matching, personalized service, efficiency improvement, user experience optimization, and cost control, significantly improving the overall performance and service quality of the customer service system.

[0062] In one optional embodiment, the first multi-dimensional demand information includes the first current session demand information and the first profile information. Based on the first multi-dimensional demand information and the knowledge information of each AI customer service representative, the confidence score of each AI customer service representative is calculated, including: determining the first score corresponding to each AI customer service representative based on whether the knowledge information of each AI customer service representative contains the first question information or information related to the first question information; the first score corresponding to the first type of AI customer service representative is greater than the first score corresponding to the second type of AI customer service representative, where the first type of AI customer service representative refers to AI customer service representatives whose knowledge information contains the first question information or information related to the first question information, and the second type of AI customer service representatives whose knowledge information does not contain the first question information or information related to the first question information; Based on the analysis of the initial session's demand information, it determines whether the target user has expressed any intention to refuse to use or express dissatisfaction with the provided customer service from any specific AI customer service representative. A second score is generated for each AI customer service representative based on this expression. The second score for a third type of AI customer service representative is higher than that for a fourth type. The third type refers to AI customer service representatives who have not been refused use or expressed dissatisfaction by the target user, while the fourth type refers to AI customer service representatives who have been refused use or expressed dissatisfaction. A specific AI customer service representative can be some or all of multiple AI customer service representatives. A confidence score is generated for each AI customer service representative based on their first and second scores. The first and second scores are two key indicators used to evaluate whether each AI customer service representative is suitable to provide customer service to the target user. Using these two scores together allows for a more comprehensive evaluation of each AI customer service representative's suitability, optimizes resource allocation, and improves the efficiency and quality of AI customer service.

[0063] Optionally, based on whether the knowledge information of each AI customer service representative contains the first question information or information related to the first question information, the first score corresponding to each AI customer service representative is determined, including: preprocessing the knowledge base of each AI customer service representative, constructing an inverted index, where the key of the inverted index is a keyword or entity in the knowledge base, and the value is a set of AI customer service representative identifiers containing these keywords or entities. For example, if the knowledge base of AI customer service representative A contains keywords such as "product model X100" and "price query", then in the inverted index, the values ​​corresponding to "product model X100" and "price query" contain the identifier of AI customer service representative A; preprocessing the first question information, including natural language processing steps such as word segmentation, stop word removal, part-of-speech tagging, and entity recognition, to extract... The keywords and entities in the question, for example, if a user's question is "I want to inquire about the price of product model X100", the preprocessed keywords and entities extracted are "product model X100" and "price". These extracted keywords and entities are then matched against the inverted index to check if each AI customer service representative's knowledge base contains these keywords or entities. For each AI customer service representative, the number of keywords or entities contained in their knowledge base is counted. For example, if AI customer service representative A's knowledge base contains "product model X100" and "price inquiry", while AI customer service representative B's knowledge base only contains "product model X100", then AI customer service representative A has 2 matches, and AI customer service representative B has 1 match. Based on the statistical results, a first score is assigned to each AI customer service representative. The specific rules are as follows: if the AI ​​customer service representative's knowledge base completely covers all keywords and entities in the first question information, the first score is 1; if the AI ​​customer service representative's knowledge base partially covers (containing at least one keyword or entity), the first score is 0.5; if the AI ​​customer service representative's knowledge base does not cover any keywords or entities, the first score is 0. For example, for the keywords "product model X100" and "price inquiry," AI customer service representative A receives a first score of 1, while AI customer service representative B receives a first score of 0.5. Keyword matching ensures that the assigned AI customer service representative is capable of handling the current problem, significantly improving the first-time resolution rate. This knowledge-matching-based scoring mechanism effectively avoids service failures due to insufficient customer service knowledge, improving user satisfaction. The use of an inverted index makes the keyword matching process highly efficient; the system can complete the knowledge matching scoring of all AI customer service representatives within milliseconds, significantly improving the efficiency of customer service representative assignment. This method does not rely on a fixed knowledge base structure and can adapt to different types of AI customer service knowledge bases. Furthermore, by dynamically updating the inverted index, the system can quickly adapt to knowledge base updates and expansions, maintaining the timeliness and accuracy of the scoring mechanism. By ensuring that the assigned AI customer service representative is capable of handling the problem, user waiting and repeated consultation time are reduced, improving user experience. Simultaneously, this mechanism also reduces user dissatisfaction caused by insufficient customer service knowledge, further improving service quality.

[0064] Optionally, based on the first session's demand information, the system analyzes whether the target user expresses any intention to refuse to use or express dissatisfaction with the provided customer service from a specific AI customer service representative. Based on this intention, a second score is generated for each AI customer service representative. This includes: performing semantic analysis and sentiment recognition on the text content of the first session's demand information; using natural language processing techniques (such as NLP), especially sentiment analysis models, to identify the emotional tendencies and intentions in the user's expressions. For example, by analyzing the user's word choice, tone, and dialogue content, it determines whether the user is expressing dissatisfaction or refusal; extracting keywords and patterns from the user's text to identify whether it contains expressions of refusal or dissatisfaction, such as keywords like "dissatisfied," "refuse," or "don't want to use," or expression patterns like "I didn't like the last customer service representative." These keywords and patterns can be matched using a predefined rule base; combining the user's historical feedback and specific expressions in the current session, it assesses the user's attitude towards a specific AI customer service representative. For example, if the user has expressed dissatisfaction with a particular AI customer service representative in a previous conversation, the system records this information and considers it in subsequent assignments; and generating a second score for each AI customer service representative based on the above analysis results. Specifically, if a user explicitly expresses refusal or dissatisfaction with a certain AI customer service representative, the AI ​​customer service representative's second score is 0; if the user does not express any refusal or dissatisfaction, the AI ​​customer service representative's second score is 1.

[0065] In one optional embodiment, the first multi-dimensional demand information includes first current session demand information and first profile information. Based on the first multi-dimensional demand information and the attribute information of whether each AI customer service representative has provided customer service to the target user, the suitability score of each AI customer service representative is calculated, including: determining a third score for each AI customer service representative based on the first current session demand information, the first profile information, and the first target service domain to which each AI customer service representative belongs, whereby the third score represents the initial suitability of the AI ​​customer service representative in providing customer service to the target user in the current session; determining the calculation weight of each AI customer service representative based on the attribute information of whether each AI customer service representative has provided customer service to the target user, whereby the calculation weight of AI customer service representatives that have provided customer service to the target user is greater than that of AI customer service representatives that have not provided customer service to the target user; and adjusting the third score of each AI customer service representative based on their calculation weight to obtain the suitability score of each AI customer service representative. Based on the needs of this conversation, user profile, and the service area of ​​the customer service representative, an initial suitability score (third score) is determined to provide a basic assessment for customer service representative matching. Next, different weights are assigned to customer service representatives based on whether they have previously served the target user; those with prior experience receive higher weights, highlighting the value of past service experience. Finally, the third score is adjusted using these weights to arrive at the final suitability score. This process not only ensures an accurate match between the customer service representative's knowledge and skills and the user's needs but also prioritizes the use of customer service representatives with prior experience, improving service consistency and user satisfaction. Simultaneously, it optimizes the efficiency of customer service resource allocation, achieving a dual improvement in service quality and user experience.

[0066] More specifically, key information can be extracted from the initial conversation's demand information, including intent category, sentiment category, and target entity information; long-term demand weights, value level tags, and historical behavioral patterns can be extracted from the initial profile information; the primary target service area to which each AI customer service representative belongs can be determined, for example, AI customer service representative A focuses on pre-sales consultation, AI customer service representative B focuses on after-sales service, etc.; and it can be checked whether the user's demand information is consistent with the service area to which the AI ​​customer service representative belongs. For example, if the user's demand is pre-sales consultation, and AI customer service representative A focuses on pre-sales consultation, then the AI ​​customer service representative has a high degree of matching with the user's demand; furthermore, the matching degree between the user's intent and the AI ​​customer service representative's service area can be calculated. If the user's intent and the AI ​​customer service representative's service area are completely consistent, a high score (e.g., 0.8) is assigned; a partial match is assigned a medium score. A score of 0.5 is assigned for a match, and a low score (e.g., 0.2) is assigned for a non-match. The score is adjusted based on the user's emotional category. If the user's emotion is positive and the AI ​​customer service performs well in that area, a score is added; if the user's emotion is negative, a score is deducted. The system also checks whether the key entities mentioned by the target user are in the AI ​​customer service's knowledge base. If the key entities match perfectly, a score is added; if they match partially or not, no score is added or a score is deducted. Furthermore, user profile information is used to assess whether the AI ​​customer service can meet the target user's long-term needs. For example, if the user is a high-value user and the AI ​​customer service has a good service record in that area, a score is added; otherwise, a score is deducted. Finally, the above matching scores are weighted and summed to obtain a third score for each AI customer service representative. The weights can be adjusted according to actual business needs. For example, the weighted summation formula is as follows:

[0067] Third_score = w1·intent_match + w2·emotion_match + w3·entity_match + w4·profile_match, where w1, w2, w3, and w4 are weights that can be adjusted according to business needs, for example, w1 = 0.4, w2 = 0.2, w3 = 0.2, and w4 = 0.2.

[0068] Furthermore, to enhance the user experience for target users, during the process of the first target AI customer service representative providing customer service to the target user, the target user's satisfaction information with the first target AI customer service representative is analyzed from multiple dimensions to obtain multi-dimensional satisfaction information. If it is determined based on the multi-dimensional satisfaction information that a change of customer service representative is needed, then based on the target user's user information, business scenario information, and the second question information most recently submitted by the target user in this session, second multi-dimensional demand information corresponding to the target user is generated. The second question information may include multiple questions and corresponding responses in subsequent communication processes. The implementation method for generating the second multi-dimensional demand information can be found in the relevant section of the first multi-dimensional demand information in the above embodiments. The description is omitted here. Based on the second multi-dimensional demand information, a second target service area is determined from multiple service areas. This second target service area contains multiple AI customer service representatives, and a second target AI customer service representative is selected from these multiple AI customer service representatives. The second target AI customer service representative is different from the first target AI customer service representative. The implementation method for determining the second target service area can be found in the relevant description of the first target service area in the above embodiments, and will not be repeated here. The dialogue content between the target user and the first target AI customer service representative is used as the dialogue context and input along with the second question information into the second target AI customer service representative so that the second target AI customer service representative can take over from the first target AI customer service representative to continue providing customer service to the target user. By continuously matching the target user with a more suitable target customer service representative in real time based on the user's needs and real-time satisfaction information during the conversation, the service quality and user experience can be continuously improved during the conversation.

[0069] In one optional embodiment, the target user's satisfaction information with the first target AI customer service is analyzed from multiple dimensions to obtain satisfaction information in multiple dimensions, including: based on the dialogue content between the first target AI customer service and the target user regarding the first question information, a satisfaction analysis of the dialogue dimension is performed to obtain satisfaction information in the dialogue dimension; or, based on the opinion information submitted by the target user, a satisfaction analysis of the subjective evaluation dimension is performed to obtain satisfaction information in the subjective evaluation dimension, wherein the opinion information is submitted by the target user through the satisfaction information collection portal.

[0070] Further optional, such as Figure 2a As shown, this embodiment also provides a human customer service resource pool, which stores human customer service representatives. If neither the first nor the second target AI customer service representative is selected, the user can be transferred to a human customer service representative to provide customer service to the target user. By combining human and AI customer service representatives and alternately providing customer service to the target user, sufficient customer service resources can be provided to ensure the quality of customer service and improve the user experience.

[0071] At this point, the customer service assignment method has been completed. This improves the matching degree between the services provided by the assigned customer service representatives and the actual needs of users, enhances the adaptability of customer service to business requirements, thereby improving the quality and efficiency of customer service and enhancing the user experience.

[0072] Figure 3 A schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. For example... Figure 3 As shown, it includes: a memory 30a and a processor 30b; the memory 30a is used to store computer programs; the processor 30b is coupled to the memory 30a and is used to execute the computer programs to implement the steps in the above method.

[0073] Furthermore, such as Figure 3 As shown, the server also includes other components such as a communication component 30c, a display 30d, a power supply component 30e, and an audio component 30f. Figure 3 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 3 The components shown.

[0074] The aforementioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0075] The aforementioned communication components are configured to facilitate wired or wireless communication between the device containing the communication components and other devices. The device containing the communication components can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication components receive broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication components also include a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wide Band (UWB), Bluetooth (BT), and other technologies.

[0076] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0077] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0078] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0079] The detailed implementation methods and beneficial effects of the electronic devices provided in this application have been described in detail in the foregoing embodiments, and will not be elaborated further here.

[0080] Exemplary embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps in the above-described method embodiments.

[0081] An exemplary embodiment of this application also provides a computer program product comprising a computer program / instructions that, when executed by a processor, enable the processor to perform the steps described in the above method embodiments.

[0082] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code.

[0083] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interfaces, network interfaces, and memory.

[0087] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0088] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. 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 random access 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 technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0089] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0090] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A customer service assignment method, characterized in that, A server-side application for a target application, wherein the server maintains multiple service domains, the method includes: Receive the first question information submitted by the target user in this session, and obtain the business scenario information to which the current session belongs and the user information of the target user; Based on the user information of the target user, the business scenario information, and the first problem information, generate the first multi-dimensional demand information corresponding to the target user; Based on the first multi-dimensional demand information, a first target service area that matches the first problem information is determined from the multiple service areas. The first target service area includes multiple AI customer service representatives. Based on the first multi-dimensional demand information, the knowledge information of the multiple AI customer service representatives, and the attribute information of whether or not customer service has been provided to the target user, the service score corresponding to each AI customer service representative is calculated. The service score reflects the suitability of each AI customer service representative to provide customer service to the target user. Based on the service score of each AI customer service representative, select the first target AI customer service representative from among the multiple AI customer service representatives; The first question information and subsequent question information are input into the first target AI customer service, so that the first target AI customer service can provide customer service to the target user.

2. The method according to claim 1, characterized in that, Based on the service scores of each AI customer service representative, a first target AI customer service representative is selected from the plurality of AI customer service representatives, including: Identify whether among the multiple AI customer service representatives included in the first target service area there is at least one historical AI customer service representative who has provided services to the target user; If included, attempt to select an AI customer service representative whose service score meets the first set condition from the at least one historical AI customer service representative as the first target AI customer service representative; and if no historical AI customer service representative whose service score meets the first set condition is selected from the at least one historical AI customer service representative, select a non-historical AI customer service representative whose service score meets the second set condition from the non-historical AI customer service representatives included in the plurality of AI customer service representatives as the first target AI customer service representative. If not included, select the AI ​​customer service representative whose service score meets the third set condition from the plurality of AI customer service representatives as the first target AI customer service representative.

3. The method according to claim 1, characterized in that, Based on the target user's user information, the business scenario information, and the first problem information, generate first multi-dimensional demand information corresponding to the target user, including: Semantic understanding is performed on the user information of the target user, the business scenario information, and the first question information to obtain semantic information; The semantic information is input into a multi-dimensional demand analysis model to analyze the target user's demands from the dimensions of the current session and routine demands, so as to obtain the first current session demand information and the first profile information corresponding to the target user, which serve as the first multi-dimensional demand information. The first profile information reflects the target user's routine demand information.

4. The method according to claim 3, characterized in that, The multi-dimensional demand analysis model includes: an intent recognition model, a sentiment analysis model, an entity extraction model, and a user profile analysis model. The semantic information is then input into the multi-dimensional demand analysis model to perform demand analysis on the target user from both the current conversation dimension and the normalized demand dimension, to obtain the first current conversation demand information and the first user profile information corresponding to the target user, including: The semantic information is input into the intent recognition model, and the intent category of the current session is identified based on the semantic information to obtain the target intent information, which serves as the first requirement information for the current session. The semantic information is input into the sentiment analysis model, and the sentiment category of the target user in this conversation is analyzed based on the semantic information to obtain target sentiment information, which serves as a first requirement information for this conversation. The semantic information is input into the entity extraction model, and the target entity information associated with this session is extracted from the question information based on the semantic information and the entity information template, as a first requirement information for this session. The semantic information is input into the user profile analysis model, and the user profile of the target user is analyzed based on the semantic information to obtain the first profile information of the target user.

5. The method according to claim 1, characterized in that, Based on the first multi-dimensional demand information, the knowledge information of the multiple AI customer service representatives, and the attribute information of whether customer service has been provided to the target user, the service score corresponding to each AI customer service representative is calculated, including: Based on the first multi-dimensional demand information and the knowledge information of each AI customer service representative, the confidence score of each AI customer service representative is calculated. The confidence score reflects whether the AI ​​customer service representative is suitable to provide customer service to the target user. Based on the first multi-dimensional demand information and the attribute information of whether each AI customer service representative has provided customer service to the target user, the suitability score of each AI customer service representative is calculated. The suitability score reflects the suitability of the AI ​​customer service representative to provide customer service to the target user. Based on the confidence score and suitability score of each AI customer service representative, the service score of each AI customer service representative is calculated.

6. The method according to claim 5, characterized in that, The first multi-dimensional demand information includes the first demand information for this session and the first profile information. Based on the first multi-dimensional demand information and the knowledge information of each AI customer service representative, the confidence score of each AI customer service representative is calculated, including: Based on whether the knowledge information of each AI customer service representative contains the first question information or information related to the first question information, a first score is determined for each AI customer service representative; the first score for the first type of AI customer service representative is greater than the first score for the second type of AI customer service representative. The first type of AI customer service representative refers to an AI customer service representative whose knowledge information contains the first question information or information related to the first question information, while the second type of AI customer service representative does not contain the first question information or information related to the first question information in its knowledge information. Based on the analysis of the first session's demand information, it is determined whether the target user has expressed any intention to refuse to use or express dissatisfaction with the provided customer service from a specific AI customer service representative. A second score is generated for each AI customer service representative based on this expression. The second score for the third type of AI customer service representative is greater than the second score for the fourth type. The third type of AI customer service representative refers to a specific AI customer service representative that has not been refused use or expressed dissatisfaction with the provided customer service by the target user. The fourth type of AI customer service representative refers to a specific AI customer service representative that has been refused use or expressed dissatisfaction with the provided customer service by the target user. The specific AI customer service representative may be some or all of the multiple AI customer service representatives. Based on the first and second scores corresponding to each AI customer service representative, a confidence score is generated for each AI customer service representative.

7. The method according to claim 5, characterized in that, The first multi-dimensional demand information includes the first current session demand information and the first profile information. Based on the first multi-dimensional demand information and the attribute information of whether each AI customer service representative has provided customer service to the target user, the suitability score of each AI customer service representative is calculated, including: Based on the first current session requirement information, the first profile information, and the first target service area to which each AI customer service belongs, a third score is determined for each AI customer service representative. The third score represents the initial suitability of the AI ​​customer service representative in providing customer service to the target user in the current session. The calculation weight of each AI customer service representative is determined based on the attribute information of whether each AI customer service representative has provided customer service to the target user. The calculation weight of the AI ​​customer service representative who has provided customer service to the target user is greater than the calculation weight of the AI ​​customer service representative who has not provided customer service to the target user. Based on the calculated weights of each AI customer service representative, the third score corresponding to each AI customer service representative is adjusted to obtain the suitability score of each AI customer service representative.

8. The method according to any one of claims 1-7, characterized in that, Also includes: During the process of the first target AI customer service providing customer service to the target user, the satisfaction information of the target user with the first target AI customer service is analyzed from multiple dimensions to obtain multi-dimensional satisfaction information; If it is determined that a customer service representative needs to be replaced based on the multi-dimensional satisfaction information, then based on the target user's user information, the business scenario information, and the target user's latest submitted second question information in the current session, a second multi-dimensional demand information corresponding to the target user is generated. Based on the second multi-dimensional demand information, a second target service area is determined from the multiple service areas. The second target service area contains multiple AI customer service representatives. A second target AI customer service representative is selected from the multiple AI customer service representatives in the second target service area. The second target AI customer service representative is different from the first target AI customer service representative. The dialogue content between the target user and the first target AI customer service representative is used as the dialogue context and input together with the second question information into the second target AI customer service representative, so that the second target AI customer service representative can take over from the first target AI customer service representative to continue providing customer service to the target user.

9. The method according to claim 8, characterized in that, The satisfaction information of the target user with the first target AI customer service is analyzed from multiple dimensions to obtain satisfaction information from multiple dimensions, including: Based on the dialogue content between the first target AI customer service representative and the target user regarding the first question information, a satisfaction analysis of the dialogue dimension is performed to obtain satisfaction information of the dialogue dimension. or Based on the opinions submitted by the target users, a satisfaction analysis of the subjective evaluation dimension is performed to obtain the satisfaction information of the subjective evaluation dimension. The opinions submitted by the target users are submitted through the satisfaction information collection portal.

10. The method according to claim 8, characterized in that, Also includes: If neither the first nor the second target AI customer service representative is selected, the user will be transferred to a human customer service representative who will provide customer service to the target user.

11. The method according to any one of claims 1-7, characterized in that, Before receiving the first question information submitted by the target user in this session, the method further includes: Receive a session request sent by the target application, the session request including the user information of the target application and the business scenario information to which the current session belongs; The target application establishes a session interface with the target user for this session, and associates the session interface with the business scenario information and the target user's user information.

12. An electronic device, characterized in that, include: Memory and processor; The memory is used to store a computer program; the processor, coupled to the memory, is used to execute the computer program to implement the steps of the method according to any one of claims 1-11.

13. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method according to any one of claims 1-11.

14. A computer program product, characterized in that, The computer program product includes a computer program / instruction that, when executed by a processor, causes the processor to perform the steps of any one of the methods of claims 1-11.

Images