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

Abstract

The invention discloses a customer service allocation method and device, equipment, a medium and a program product, and belongs to the field of customer service allocation. The method comprises the following steps: acquiring a consultation question of a service object and respective working state information of a plurality of candidate customer services; performing classification operation on the consultation questions through a first artificial intelligence model to obtain a first question category; screening out a first number of candidate customer services carrying a first question category label from the plurality of candidate customer services, wherein the first question category label corresponds to the first question category; based on the respective working state information of the plurality of candidate customer service staff, screening out a second number of candidate customer service staff of which the working states meet the working state condition from the plurality of candidate customer service staff; and based on the intersection of the first number of candidate customer service staff and the second number of candidate customer service staff, determining the customer service staff for processing the consultation question. According to the method, question classification is carried out through the first artificial intelligence model, and compared with keyword matching of related technologies, the classification result is more accurate.

Description

Technical Field

[0001] This application relates to the field of customer service assignment, and particularly to a customer service assignment method, apparatus, device, medium, and program product. Background Technology

[0002] In daily life, users can consult customer service for help when they encounter difficulties.

[0003] In related technologies, when a user submits an inquiry to the customer service center, the customer service center will determine the target category of the inquiry through keyword matching, and then select the customer service representative who belongs to the same target category from among multiple candidate customer service representatives as the target customer service representative, who will then answer the user's inquiry.

[0004] In related technologies, keyword matching results in poor accuracy in problem classification, leading to inaccurate matching of customer service representatives and preventing users from receiving timely and correct answers to their inquiries. Summary of the Invention

[0005] This application provides a customer service assignment method, apparatus, device, medium, and program product. This application uses a first artificial intelligence model for problem classification, which yields more accurate classification results compared to keyword matching in related technologies. The technical solution is as follows:

[0006] According to one aspect of this application, a customer service assignment method is provided, the method comprising:

[0007] Obtain the customer's inquiry and the work status information of multiple candidate customer service representatives;

[0008] The first artificial intelligence model is used to classify the consultation questions to obtain the first question category.

[0009] Select the first number of candidate customer service representatives from multiple candidate customer service representatives who carry the first problem category tag, where the first problem category tag corresponds to the first problem category;

[0010] In addition, based on the work status information of multiple candidate customer service representatives, a second number of candidate customer service representatives whose work status meets the work status conditions are selected from the multiple candidate customer service representatives.

[0011] Based on the intersection of the first and second number of candidate customer service representatives, the customer service representative to handle the inquiry is determined.

[0012] According to another aspect of this application, a customer service assignment device is provided, the device comprising:

[0013] The acquisition module is used to acquire the inquiry questions of the service recipients and the work status information of multiple candidate customer service representatives;

[0014] The classification module is used to perform classification operations on the consultation questions through the first artificial intelligence model to obtain the first question category;

[0015] The filtering module is used to filter out a first number of candidate customer service representatives from multiple candidate customer service representatives who carry the first problem category label, where the first problem category label corresponds to the first problem category;

[0016] The filtering module is also used to filter out a second number of candidate customer service representatives whose work status meets the work status conditions from multiple candidate customer service representatives based on their respective work status information.

[0017] The determination module is used to determine the customer service representative to handle the inquiry based on the intersection of a first number of candidate customer service representatives and a second number of candidate customer service representatives.

[0018] According to one aspect of this application, a computer device is provided, the computer device comprising: a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the customer service assignment method as described above.

[0019] According to another aspect of this application, a computer-readable storage medium is provided, the storage medium storing a computer program that is loaded and executed by a processor to implement the customer service assignment method as described above.

[0020] According to another aspect of this application, a computer program product is provided, the computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform the client allocation method provided in the above aspect.

[0021] The beneficial effects of the technical solutions provided in this application include at least the following:

[0022] In this application, a first artificial intelligence model is used to classify inquiries. Then, based on the classified first question category, candidate customer service representatives' question category tags are matched. The final target customer service representative (the one handling the inquiry) will carry the correct question category tag. The first artificial intelligence model performs question classification, resulting in more accurate results compared to keyword matching in related technologies. Furthermore, the customer service allocation scheme provided in this application avoids manual classification, speeding up the allocation process and reducing labor costs. Additionally, this application considers the work status information of candidate customer service representatives, avoiding assigning current inquiries to representatives with poor recent work performance. The assigned target customer service representative is better able to resolve inquiries and provide better service to users.

[0023] In addition, this application will screen out a first number of candidate customer service representatives and a second number of candidate customer service representatives in parallel. The first number of candidate customer service representatives carries the correct question category label, and the second number of candidate customer service representatives meets the working status conditions. Through two parallel screening processes, the final target customer service representatives can be screened out in a distributed manner. Distributed computing helps to shorten the overall customer service allocation time. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram of a customer service allocation system provided in an exemplary embodiment is shown;

[0026] Figure 2 A flowchart of a customer service assignment method provided by an exemplary embodiment is shown;

[0027] Figure 3 A flowchart of a problem classification method provided by an exemplary embodiment is shown;

[0028] Figure 4 A flowchart of a customer service assignment method provided by an exemplary embodiment is shown;

[0029] Figure 5 A schematic diagram of a customer service assignment method provided in an exemplary embodiment is shown;

[0030] Figure 6 A flowchart illustrating a method for assigning category labels to candidate customer service issues, as provided in an exemplary embodiment, is shown.

[0031] Figure 7 A schematic diagram illustrating a method for assigning category labels to candidate customer service issues, provided by an exemplary embodiment, is shown.

[0032] Figure 8 A schematic diagram of a method for retraining a second artificial intelligence model provided by an exemplary embodiment is shown;

[0033] Figure 9 A schematic diagram of a method for retraining a first artificial intelligence model provided by an exemplary embodiment is shown;

[0034] Figure 10 A schematic diagram illustrating a method for training a first artificial intelligence model and a second artificial intelligence model provided in an exemplary embodiment is shown.

[0035] Figure 11 A structural block diagram of a customer service allocation device provided in an exemplary embodiment is shown;

[0036] Figure 12 A structural block diagram of a computer device provided in an exemplary embodiment is shown. Detailed Implementation

[0037] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0038] It should be understood that "several" in this article refers to one or more, and "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0039] It should be noted that the information (including but not limited to device information, personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the subject or fully authorized by all parties, and the collection, use and processing of related data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0040] First, a brief introduction to the terms used in the embodiments of this application:

[0041] Customer service assignment: In various service scenarios on the internet, customer service agents are often set up to handle inquiries from users. When a user raises an inquiry, the backend will assign a customer service agent to handle that inquiry. The need for customer service assignment exists in various common service scenarios. For example, in shopping scenarios, users can request assistance from the seller's customer service or the backend customer service of the shopping application when shopping online; in travel scenarios, users can request assistance from the backend customer service when encountering difficulties using a travel application; and in gaming scenarios, users can request assistance from the backend customer service when encountering errors in a game application, and so on.

[0042] Figure 1A schematic diagram of a customer service allocation system provided in an exemplary embodiment of this application is shown.

[0043] Figure 1 In this system, the customer service allocation system includes a first terminal device 110, a first backend server 120, a second backend server 130, and a second terminal device 140. The first terminal device 110 runs a first instant messaging application. A service recipient submits a consultation question 101 on the first instant messaging application. The first terminal device 110 sends consultation question 101 to the first backend server 120, which is a backend server supporting the operation of the first instant messaging application. The first backend server 120 forwards the consultation question to the second backend server 130, which executes a customer service allocation process 100. Based on the customer service allocation process 100, the second backend server 130 allocates a target customer service representative 180 to handle consultation question 101. The second backend server 130 then sends consultation question 101 to the second terminal device 140, which runs a second instant messaging application. The customer service account of the target customer service representative 180 is logged into the second instant messaging application. The second backend server 130 is a backend server supporting the operation of the second instant messaging application. Optionally, the first instant messaging application is a personal instant messaging application, and the second instant messaging application is an enterprise instant messaging application.

[0044] like Figure 1 As shown, the customer service assignment process 100 includes two screening processes.

[0045] In the first screening process, a first number of candidate customer service representatives 160 will be selected from multiple candidate customer service representatives 150. Each of the multiple candidate customer service representatives 150 will carry one or more question category tags. In this application, the consultation question 101 initiated by the service recipient will be obtained. Then, the consultation question 101 will be input into the first artificial intelligence model 102, which will classify the consultation question 101 to obtain the first question category 103. From the multiple candidate customer service representatives 150, a first number of candidate customer service representatives 160 carrying the first question category tag will be selected. The first question category tag corresponds to the first question category. For example, if the first question category is a work order type question, the first number of candidate customer service representatives 160 will be selected from the multiple candidate customer service representatives 150, and each of the first number of candidate customer service representatives 160 will carry a work order type question tag.

[0046] The first screening process starts with the consultation question 101 raised by the service recipient and selects a first number of candidate customer service representatives 160 who have the ability to solve the consultation question 101. In this process, the consultation question 101 is classified by the first artificial intelligence model 102, which can effectively improve the accuracy of question classification. Finally, the target customer service representative 180 matched can also effectively solve the consultation question 101.

[0047] In the second screening process, a second number of candidate customer service representatives 170 will be selected from multiple candidate customer service representatives 150. In this application, the work status information 104 of each candidate customer service representative will be obtained. This work status information includes real-time idle status, workload, and the status of recently handled issues. From the multiple candidate customer service representatives 150, those whose work status meets the work status criteria will be selected, resulting in the second number of candidate customer service representatives 170. The second screening process will select candidate customer service representatives whose work status meets the work status criteria. For candidate customer service representatives with poor work status, consultation question 101 will not be assigned to them. For example, consultation question 101 will not be assigned to customer service representatives who have been temporarily absent from their posts, customer service representatives with many pending issues, or customer service representatives whose work status has recently been reported as poor.

[0048] After two screening processes, the intersection of the first number of candidate customer service representatives 160 selected through the first screening process and the second number of candidate customer service representatives 170 selected through the second screening process will be used to determine the target customer service representative 180. The target customer service representative 180 is the final customer service representative to handle the inquiry 101.

[0049] In this application, in addition to considering whether the assigned customer service representatives have the ability to resolve the inquiries raised by the service recipients, the working status of the customer service representatives is also taken into account. This avoids assigning inquiries to customer service representatives with poor recent performance, ensuring that the assigned target customer service representatives are better able to resolve the inquiries. Furthermore, through the aforementioned parallel first and second screening processes, the final target customer service representatives (180) can be selected in a distributed manner. Distributed computing helps to shorten the overall customer service representative allocation time and accelerate the resolution of inquiries.

[0050] In fact, for service scenarios with a large audience, a large number of customer service allocation requests may be received in a short period of time. The second-backup server needs to respond to a large number of customer service allocation requests in a short period of time, resulting in a long response time. This application adopts a distributed computing approach to respond to customer service allocation requests. For a single customer service allocation request, it is processed in parallel, which shortens the response time of that request. For a large number of customer service allocation requests, the overall parallel processing can significantly shorten the overall allocation time. For example, if the first screening process of customer service allocation request 1 has not been completed while the second screening process has been completed, the second screening process of customer service allocation request 2 can continue to be executed without waiting for customer service allocation request 1 to be fully processed before responding to customer service allocation request 2. Therefore, distributed computing has a significant speed improvement effect in scenarios with a large number of customer service allocation requests.

[0051] In one embodiment, the device type of either the first terminal device 110 or the second terminal device 140 includes at least one of the following: smartphone, smartwatch, smart TV, tablet computer, wearable device, in-vehicle terminal, e-book reader, MP3 player, MP4 player, laptop computer, and desktop computer. Either the first backend server 120 or the second backend server 130 includes at least one of the following: a server, multiple servers, a cloud computing platform, and a virtualization center.

[0052] In the above embodiments, the customer service allocation process 100 is only described as being executed by the second backend server 130. In fact, the customer service allocation process 100 can also be executed in a distributed manner by the first terminal device 110, the first backend server 120, the second backend server 130, and the second terminal device 140.

[0053] Figure 2 A flowchart of a customer service assignment method provided in an exemplary embodiment of this application is shown, illustrated by way of example, the method being performed by a computer device, the method including:

[0054] Step 210: Obtain the service recipient's inquiry and the work status information of each of the multiple candidate customer service representatives;

[0055] The service recipient, optionally, refers to the application's users, such as users logged into instant messaging applications or travel applications. Service recipients frequently encounter various problems while using the application, at which point they will contact the application's customer service for assistance. In this scenario, Figure 2 The customer service assignment method shown will be executed by the application's backend server.

[0056] Alternatively, the service recipient refers to the object served by a third party providing services based on an application. For example, a service recipient initiates a consultation with a third party on the first application, and the third party's customer service answers the consultation with the service recipient on the second application. Optionally, the first application and the second application may be the same application or different applications. In this scenario, Figure 2 The customer service assignment method shown will be executed by the backend server of the first application and / or the backend server of the second application.

[0057] In one embodiment, the first application is a service-providing application, and the second application is an instant messaging application with a customer service account logged in. In another embodiment, the first application is a first instant messaging application, and the second application is a second instant messaging application. Optionally, the first instant messaging application is a personal instant messaging application, and the second instant messaging application is an enterprise instant messaging application.

[0058] Consultation questions refer to questions raised by service recipients. For example, service recipients may ask questions such as "How long does it take to process a work order?", "When will order xxx be shipped?", or "Does the latest version of the application have the xxx function?" These are questions that service recipients ask the service provider offering the service.

[0059] In one embodiment, the consultation question includes at least one of text, emoticons, images, audio, and video. Optionally, the text in the consultation question can be in languages ​​such as Chinese, English, or German.

[0060] Work status information is used to describe the work status of customer service representatives. For example, among multiple candidate customer service representatives, customer service representative 1 has an excellent work status, customer service representative 2 has an excellent work status, customer service representative 3 has a moderate work status, and customer service representative 4 has a poor work status. In this application, the work status of customer service representatives will also be taken into account when assigning them.

[0061] Optionally, work status information includes at least one of the following: real-time idle status, workload status, and work status of recently handled issues. Real-time idle status indicates whether the current customer service representative is available. For example, if the current representative has pending inquiries, it indicates they are not idle; conversely, if there are no pending inquiries, they are idle. Workload status indicates the number of inquiries awaiting the current customer service representative's attention. For instance, if there are five inquiries, the representative's workload is high. Work status of recently handled issues refers to an evaluation of the representative's recent work performance. For example, if the representative has received many complaints recently, their work performance is poor; if they have received few or no complaints, their work performance is good.

[0062] Step 220: Using the first artificial intelligence model, perform a classification operation on the consultation questions to obtain the first question category;

[0063] After obtaining the consultation questions initiated by the service recipients, the consultation questions will be classified using a first artificial intelligence model to obtain the first question category corresponding to the consultation questions. Optionally, the first artificial intelligence model can be a machine learning model such as support vector machine, random forest, or logistic regression. Optionally, the first artificial intelligence model can also be a deep learning model, such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs).

[0064] Step 230: Select a first number of candidate customer service representatives from multiple candidate customer service representatives who carry the first problem category tag, where the first problem category tag corresponds to the first problem category;

[0065] In this application, each of the multiple candidate customer service representatives carries one or more problem category tags, representing the ability of each candidate customer service representative to solve one or more types of problems. After determining the first problem category to which the consultation question initiated by the service recipient belongs, candidate customer service representatives carrying the first problem category tag are selected from the multiple candidate customer service representatives to obtain a first number of candidate customer service representatives, where the first problem category tag corresponds to the first problem category.

[0066] For illustrative purposes, the issue category labels include labels for issues such as application update issues, application function issues, settings issues, and data security issues; for illustrative purposes, the issue category labels include labels for issues such as pricing issues, product information issues, express delivery information issues, and quality issues. The issue category labels mentioned above are merely examples, and in actual applications, other issue category labels may be included, as well as more granular or coarser issue category labels. This application does not limit such applications.

[0067] Step 240: Based on the work status information of each of the multiple candidate customer service representatives, select a second number of candidate customer service representatives whose work status meets the work status conditions from the multiple candidate customer service representatives.

[0068] After obtaining the work status information of multiple candidate customer service representatives, a second number of candidate customer service representatives will be selected from them based on their work status meeting the specified criteria. For example, the second number of candidate customer service representatives will be selected based on their excellent work status.

[0069] In one embodiment, the work status information includes at least two of the following: real-time idle status, workload status, and work status of recently handled issues. In this application, by comprehensively considering at least two of these three work statuses, candidate customer service representatives whose overall work status meets the work status criteria can be selected. By comprehensively considering multiple work status information, the work status of customer service representatives can be evaluated more accurately, leading to more accurate candidate customer service representatives selected subsequently based on the work status information, and facilitating the accurate allocation of customer service resources.

[0070] Step 250: Based on the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives, determine the customer service representative who will handle the inquiry.

[0071] In one embodiment, if the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives is exactly one customer service representative, then that customer service representative is identified as the customer service representative to handle the inquiry. If the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives includes multiple customer service representatives, then based on the time when the question was last assigned, the customer service representative who has not been assigned a question for the longest time is selected from the multiple customer service representatives and identified as the target customer service representative for handling the inquiry initiated by the service recipient.

[0072] In this application, based on the first problem category of the consultation question, customer service representatives 1, 2, and 3 with the ability to solve the first problem category are selected from multiple candidate customer service representatives. Based on the work status information of each of the multiple candidate customer service representatives, customer service representatives 1 and 2 whose work status meets the work status conditions are selected. Based on the intersection of these two ranges, customer service representative 1 is determined to be the customer service representative who handles the consultation question.

[0073] Step 230 above illustrates the first screening process, and step 240 illustrates the second screening process. In one embodiment, the first screening process is initiated upon receiving an inquiry from a service recipient, and the second screening process is executed automatically at regular intervals, for example, every 5 minutes. After the first screening process is completed, the second number of candidate customer service representatives selected from the most recently executed second screening process is chosen based on the completion time of the first screening process.

[0074] At this point, the second screening process will be executed automatically and repeatedly. This makes it easy to quickly extract the latest screening results of the second screening process whenever a service recipient's inquiry is received. At this point, the time spent on the entire customer service allocation only needs to take into account the time spent on the first screening process, without having to consider the time spent on the second screening process.

[0075] Optionally, the time interval for executing the second screening process can be set. For example, during peak hours, more service recipients will initiate inquiries. In this case, setting the time interval for executing the second screening process to be shorter and executing the second screening process more frequently is beneficial for timely screening of the second number of candidate customer service representatives based on the latest work status, and timely and accurate customer service allocation.

[0076] For example, during off-peak hours, fewer service recipients will initiate inquiries. In this case, setting a longer interval for executing the second screening process and executing the second screening process infrequently helps save computing and storage resources used for executing the second screening process and reduces idle time on the backend server.

[0077] In another embodiment, both the first and second screening processes are initiated upon receiving a consultation question from a service recipient. After either the first or second screening process is completed, the process waits for the other screening process to complete before performing the intersection operation.

[0078] At this point, the second screening process is only executed after the inquiry is received. This avoids the backend server running idle. During some off-peak periods (such as the early morning), only a very small number of service recipients will initiate inquiry questions. If the backend server repeatedly executes the second screening process, it will seriously waste computing and storage resources. Executing the second screening process only after the inquiry is received can greatly reduce resource waste.

[0079] based on Figure 2 In the alternative embodiments shown, Figure 3 A flowchart illustrating a method for classifying consultation questions according to an exemplary embodiment of this application is shown, wherein step 220 includes... Figure 3 Steps 320 to 360 are shown. Figure 3 The methods shown include:

[0080] Step 320: Extract key information from the consultation question using natural language processing technology;

[0081] Optionally, natural language processing techniques include at least one of word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, and semantic analysis. By performing at least one of word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, and semantic analysis on the consultation question, a processed consultation question is obtained; then, key information is extracted from the processed consultation question, including at least one of the following: the topic of the consultation question, the actions in the consultation question, and the objects in the consultation question.

[0082] Once the consultation question is obtained, the question text is broken down into words or phrases. Then, part-of-speech tagging and named entity recognition technologies are used to identify nouns, verbs, and proper nouns in the question. Next, syntactic and semantic analysis is performed to understand the question's structure and meaning. After this processing, the processed consultation question is obtained.

[0083] Extract key information from the processed inquiries. Key information includes at least one of the following: the topic of the inquiry, the actions involved in the inquiry, and the object of the inquiry. The topic of the inquiry may be, for example, an inquiry about product information, an inquiry about express delivery information, or an inquiry about discounts. The actions involved in the inquiry refer to verbs such as "buy," "apply," "ship," or "discount." The object of the inquiry refers to nouns such as fabric, brand, belt, or mobile phone.

[0084] In this embodiment of the application, natural language processing technology will be used to gain a deeper understanding of the consultation questions raised by the service recipients, thereby more accurately understanding the semantics of the consultation questions, knowing the needs of the service recipients, and laying a foundation for subsequent question classification and customer service allocation.

[0085] Step 340: Based on key information, extract the problem characteristics of the consultation question;

[0086] Based on the key information extracted from the consultation questions above, the characteristics of the consultation questions are extracted. The characteristics include at least one of the following: keyword characteristics, theme characteristics, and emotional characteristics. Keyword characteristics are the representation of keywords in the consultation questions, such as "express delivery" and "return". Theme characteristics are the representation of the theme in the consultation questions, such as the theme of product information or the theme of express delivery information. Emotional characteristics are the representation of the user's emotions when asking the consultation questions, such as the user asking the consultation questions with anger or with a calm emotion. Emotional characteristics are anger and calmness.

[0087] In this embodiment of the application, the key information extracted from the consultation question will be used to further summarize the characteristics of the consultation question, so as to classify the consultation question in the future.

[0088] Step 360: Input the problem features into the first artificial intelligence model to obtain the first problem category.

[0089] The problem features are input into the pre-trained first artificial intelligence model to obtain the first problem category.

[0090] In summary, in addition to using the first artificial intelligence model for classification, natural language processing technology was also used to process the consultation questions, ensuring a correct understanding of the semantics of the consultation questions, which in turn enabled accurate classification of the consultation questions.

[0091] In one embodiment, in Figure 3 Before step 320, i.e., before performing natural language processing on the consultation question, data preprocessing is first performed on the consultation question. In one embodiment, the data preprocessing method used in this application includes at least one of the following two:

[0092] 1. Use the RoBERTa pre-trained language model to generate word vectors for consultation questions;

[0093] For example, if the consultation question is "What is the quality of this product?", input the words in this consultation question into the RoBERTa model. The RoBERTa model will generate a word vector of a specific length for each word. For example, "this model" will be represented as a vector of length 768 [0.12, -0.34, 0.56,...], and "product" will be represented as another vector [0.23, 0.11, -0.45,...], and so on.

[0094] In the RoBERTa model, byte-level BPE is used as the text encoding method, replacing the character-level BPE in the BERT model. Byte-level BPE can learn a medium-sized (50k units) sub-word vocabulary, and thus can encode any input text without introducing additional tokens to replace words not appearing in the vocabulary, thereby improving the model's text processing ability and efficiency to a certain extent.

[0095] In the embodiments of this application, for the consultation questions raised by the service object, each word in the consultation question is input into the RoBERTa pre-trained model, and the corresponding word vector representation will be obtained. The word vectors encoded by the RoBERTa pre-trained model not only contain the basic meanings of the words, but also can reflect the semantic changes in different contexts, thereby improving the accuracy of question classification.

[0096] 2. Clean and standardize the consultation questions;

[0097] Cleaning and standardizing the consultation questions includes at least one of the following:

[0098] (1) Identify and remove the noise and irrelevant content in the consultation question;

[0099] For example, remove HTML tags, special characters, punctuation marks, etc. in the consultation question. These contents may interfere with the generation of word vectors and the accuracy of question classification.

[0100] (2) Unify case and perform stemming;

[0101] For example, unify the words in the text to lowercase to reduce the diversity of vocabulary. For example, unify "Hello" and "hello" to "hello". For stemming, a similar operation can be performed in Chinese. For example, reduce "running", "running", "running away", etc. to "run", which can unify different expressions, reduce the number of vocabulary, and improve processing efficiency.

[0102] (3) Remove stop words;

[0103] For stop word removal, in Chinese, common words that contribute less to semantic expression, such as "的", "是", "在", "了", etc., can be removed. For example, for a consulting question "我的手机是黑色的,现在出现了一些问题", after removing stop words, it becomes "手机黑色出现问题", which reduces the interference of some meaningless words and is more conducive to subsequent word vector generation and question classification. Also, existing stop word lists can be used, or a custom stop word list can be created according to the specific problem domain. In this application, removing stop words can reduce the noise in the text, improve the quality of word vectors, and the accuracy of question classification.

[0104] (4) Spelling correction and synonym replacement;

[0105] For spelling correction, a spelling check tool can be used to correct spelling mistakes in the consulting question. For synonym replacement, synonyms in the text can be unified into a standard vocabulary. For example, "美丽", "漂亮", and "好看" can be unified into "美丽", which can reduce the diversity of vocabulary, improve the accuracy of word vectors, and the efficiency of question classification.

[0106] (5) Text normalization and standardization;

[0107] In the embodiments of this application, the consulting question is normalized and standardized to conform to certain formats and specifications. For example, the date format is unified into "YYYY-MM-DD", and the currency amount format is unified into a specific symbol and number combination, etc. Also, regular expressions or specialized text processing tools can be used to achieve text normalization and standardization, which can make the text data more tidy and easy to process, and improve the accuracy of word vector generation and question classification.

[0108] Select the second number of candidate customer service representatives according to the working status

[0109] Based on Figure 2 In the optional embodiment shown, the working status information includes the real-time idle status, workload situation, and working status of handling recent problems. Figure 4 shows a flowchart of a method for selecting the second number of candidate customer service representatives according to the working status information provided by an embodiment of this application. Step 240 includes Figure 4 Steps 420 and 440 in Figure 4 The method shown includes:

[0110] Step 420, for the first candidate customer service representative among multiple candidate customer service representatives, based on the weighted sum of the first working status data, second working status data, and third working status data of the first candidate customer service representative, obtain the working status score of the first candidate customer service representative;

[0111] The first candidate customer service representative is any one of several candidate customer service representatives. (Illustrative example, such as...) Figure 5 As shown, for the first candidate customer service representative among multiple candidate customer service representatives, the first status score 503 is obtained by multiplying the first weight 501 and the first work status data 502 of the first candidate customer service representative; the second status score 506 is obtained by multiplying the second weight 504 and the second work status data 505 of the first candidate customer service representative; the third status score 509 is obtained by multiplying the third weight 507 and the third work status data 508 of the first candidate customer service representative; and the work status score 510 of the first candidate customer service representative is obtained by summing the first status score 505, the second status score 506, and the third status score 509.

[0112] The first working status data is used to indicate the real-time idle status of the first candidate customer service representative. For example, a first value indicates that the first candidate customer service representative is in an idle state, and a second value indicates that the first candidate customer service representative is in a non-idle state. Further, more values ​​can be used for finer-grained distinctions. For example, the first working status data can represent the duration of the first candidate customer service representative's idle state. For non-idle states, the first working status data is represented by a negative number. For instance, if the duration of the first candidate customer service representative's non-idle state is 300 minutes, the first working status data is represented by the value -300; if the duration of the first candidate customer service representative's idle state is 50 minutes, the first working status data is represented by the value 50. Optionally, after determining the first working status data, a normalization operation is performed. Based on the product of the normalized first working status data and the first weight, a first status score is obtained. For example, the first working status data of the first candidate customer service representative is divided by the sum of the first working status data of multiple candidate customer services to obtain the normalized first working status data of the first candidate customer service representative.

[0113] The second work status data is used to indicate the workload of the first candidate customer service representative. For example, the second work status data might be the number of pending inquiries currently being handled by the first candidate customer service representative, say, 5. Optionally, after determining the second work status data, a normalization operation is performed. Based on the product of the normalized second work status data and the second weight, a second status score is obtained. For example, the second work status data of the second candidate customer service representative is divided by the sum of the second work status data of multiple candidate customer service representatives to obtain the normalized second work status data of the second candidate customer service representative.

[0114] The third work status data is used to indicate the work status of the first candidate customer service representative in handling recent issues. Optionally, the third work status data is the negative of the number of complaints received by the first candidate customer service representative in handling recent issues. "Recent" can refer to a preset time range. For example, if the first candidate customer service representative was complained about 3 times in handling recent issues, then the third work status data is determined to be -3. Optionally, after determining the third work status data, a normalization operation is performed. Based on the product of the normalized third work status data and the third weight, a third status score is obtained. For example, the third work status data of the third candidate customer service representative is divided by the sum of the third work status data of multiple candidate customer service representatives to obtain the normalized third work status data of the third candidate customer service representative.

[0115] In one embodiment, the first weight, the second weight, and the third weight are dynamically adjustable. For example, if more consideration is given to the workload of customer service staff, a larger second weight is set; if more consideration is given to the recent work status of customer service staff in handling issues, a larger third weight is set.

[0116] Step 440: Select the candidate customer service representatives whose work status scores meet the score conditions from multiple candidate customer service representatives to obtain the second number of candidate customer service representatives.

[0117] In one embodiment, customer service representatives with work status scores greater than a threshold are selected from a pool of candidate customer service representatives and added to a second pool of candidate customer service representatives. In another embodiment, the top K customer service representatives with the highest work status scores are selected from a pool of candidate customer service representatives and added to a second pool of candidate customer service representatives. In this case, the second pool is a preset value, namely K, where K is a positive integer.

[0118] In one embodiment, if the work status scores of multiple candidate customer service representatives are all less than the score threshold, the work status scores of multiple candidate customer service representatives are recalculated after waiting for a target preset time t.

[0119] It should be noted that the process of selecting the first number of candidate customer service representatives by reasoning from the first artificial intelligence model takes a relatively long time, while the process of selecting the second number of candidate customer service representatives by selecting the second number of candidate customer service representatives (i.e., the second screening process) takes a relatively short time. Therefore, if a satisfactory number of candidate customer service representatives cannot be obtained after executing the second screening process once, the second screening process will be executed again after a period of time until a satisfactory result is obtained. Since the first screening process and the second screening process are parallel, re-executing the second screening process will not take up additional time.

[0120] In this embodiment, the real-time idle status, workload, and recent problem-solving status of customer service representatives will be comprehensively considered to comprehensively evaluate their work status. Customer service representatives with better work status will handle the inquiries initiated by the current service recipients, and the problems can be resolved in a timely, effective, and high-quality manner.

[0121] Assign category labels to candidate customer service issues

[0122] Based on the above introduction, step 240 will select candidate customer service representatives carrying the first problem category tag from multiple candidate customer service representatives. Before the selection, each candidate customer service representative will be assigned a problem category tag. Assigning problem category tags to customer service representatives can be done in real time after receiving the customer service representative assignment request, or it can be done offline. The backend pre-stores the problem category tags of each candidate customer service representative. Figure 6 This application illustrates a method for assigning category labels to customer service issues, the method comprising:

[0123] Step 620: Based on the job skill information of each of the multiple candidate customer service representatives, a second artificial intelligence model is used to assign a problem category label to each of the multiple candidate customer service representatives. The job skill information includes at least one of historical training, work experience, and historical problem handling.

[0124] The second artificial intelligence model can be a machine learning model such as support vector machine, random forest, logistic regression, or a deep learning model such as convolutional neural network and long short-term memory network.

[0125] Job skills information includes at least one of the following: historical training, work experience, and past problem-solving experience. Historical training includes information such as the training courses the candidate has participated in, their performance during those courses, and the dates of those courses. Work experience includes the candidate's previous work history, such as their job positions and previous employers. Past problem-solving experience includes issues the candidate has handled, such as difficult or noteworthy problems.

[0126] like Figure 7 As shown, for the first candidate customer service representative among multiple candidate customer service representatives (where the first candidate customer service representative is any one of the multiple candidate customer service representatives), the job skill information 701 of the first candidate customer service representative is input into the second artificial intelligence model 702 to obtain at least one problem category label 703. At least one problem category label 703 indicates that the first candidate customer service representative has the ability to solve one or more types of problems. Figure 7 The image shows issue category label 1, issue category label 2, and issue category label 3.

[0127] In this embodiment of the application, by assigning question category tags to candidate customer service representatives based on their job skill information, it is possible to accurately distinguish customer service representatives with various professional capabilities, thereby ensuring that the consultation questions raised by the service recipients are ultimately answered by professional customer service representatives, thus providing professional services to users.

[0128] In one embodiment, after assigning multiple candidate customer service problem category labels through the second artificial intelligence model, the problem category labels carried by each candidate customer service representative will be adjusted according to the actual handling effect of each candidate customer service representative on historically handled problems, resulting in updated problem category labels carried by each candidate customer service representative.

[0129] For example, the second AI model might assign a "post-sales issue" tag to the first candidate customer service representative, but if that representative has a history of slow processing of such issues, the tag would be removed. Conversely, even if the second AI model doesn't assign a "product information issue" tag to the first candidate customer service representative, but they have received positive feedback from users regarding product information issues, the tag would be added to them.

[0130] In this embodiment of the application, the problem category label of each candidate customer service representative will be dynamically adjusted based on the actual handling effect of historical problems, so as to more accurately reflect the skills and strengths of each candidate customer service representative.

[0131] Training a second artificial intelligence model

[0132] In the above Figure 6 The method shown for assigning category labels to candidate customer service questions utilizes a second artificial intelligence model, and the training method for the second artificial intelligence model will be introduced next.

[0133] In one embodiment, when training the second AI model, the job skill information and multiple problem category labels of multiple sample customer service representatives are obtained. For a sample customer service representative, the job skill information of that representative is input into the second AI model to predict the problem categories that the representative can solve. Based on the gap between the problem category labels possessed by the representative and the predicted problem categories, the model parameters of the second AI model are adjusted. That is, the second AI model needs to rely on problem category labels for training.

[0134] In one embodiment, after each first preset time period, updated problem category tags are generated based on feedback information provided by historical service recipients and multiple candidate customer service representatives within the first preset time period; and updated job skill information of each of the multiple candidate customer service representatives is obtained; and a second artificial intelligence model is retrained based on the updated problem category tags and the updated job skill information of each of the multiple candidate customer service representatives.

[0135] Within the first preset time period, the job skills information of the candidate customer service representatives may be updated. For example, if the first preset time period is one week, the candidate customer service representatives may participate in new training courses or solve new problems within one week. This application will use the updated job skills information of the candidate customer service representatives to retrain the second artificial intelligence model.

[0136] Furthermore, in this embodiment, the problem category label will be updated based on the feedback information provided by the service recipient within the first preset time period by the candidate customer service. For example, if a historical service recipient believes that the candidate customer service assigned by the system is not a good match and cannot accurately solve the consultation problem, the historical service recipient will provide the problem category of the consultation problem in the service feedback table. If the problem category was not used in the past, it will be used as the updated problem category label this time, and then the second artificial intelligence model will be retrained.

[0137] Furthermore, in this embodiment, the problem category label will be updated based on the feedback information provided by multiple candidate customer service representatives within a first preset time period. For example, if a candidate customer service representative handles several problems within the first preset time period and believes that many of the problems belong to the same problem category, and that the problem category has not been used in the past, then the problem category will be used as the updated problem category label, and the second artificial intelligence model will be retrained.

[0138] like Figure 8 As shown, Figure 8 Three scenarios for retraining the second artificial intelligence model are shown.

[0139] exist Figure 8 In part (A), for the first candidate customer service representative among multiple candidate customer service representatives, the first candidate customer service representative is any one of the multiple candidate customer service representatives. The new job skill information of the first candidate customer service representative (i.e., the job skill information of the first candidate customer service representative has been updated within the first preset time period) 801 is input into the second artificial intelligence model 802 to predict the problem category that the first candidate customer service representative can solve, and obtain the first predicted problem category 803. Based on the difference between the old problem category label carried by the first candidate customer service representative (i.e., the problem category label corresponding to the first candidate customer service representative has not been updated within the first preset time period) 804 and the first predicted problem category 803, the second artificial intelligence model 802 is trained.

[0140] exist Figure 8 In part (B), the old job skill information of the first candidate customer service representative (i.e., the job skill information of the first candidate customer service representative has not been updated within the first preset time period) 805 is input into the second artificial intelligence model 802 to predict the problem categories that the first candidate customer service representative can solve, and obtain the second predicted problem category 806. Based on the new problem category label carried by the first candidate customer service representative (i.e. the problem category label corresponding to the first candidate customer service representative has been updated within the first preset time period) 807 and the difference between the second predicted problem category 806, the second artificial intelligence model 802 is trained.

[0141] exist Figure 8In part (C), the new job skills information 801 of the first candidate customer service representative is input into the second artificial intelligence model 802 to predict the problem categories that the first candidate customer service representative can solve, and the first predicted problem category 803 is obtained. Based on the gap between the new problem category label 807 carried by the first candidate customer service representative and the first predicted problem category 803, the second artificial intelligence model 802 is trained.

[0142] It is worth noting that, in Figure 8 In the three scenarios shown, the retraining of the second artificial intelligence model in this application involves updating the labels, rather than just updating the job skills information of the candidate customer service representatives. In this case, the updating of the labels enables the retrained second artificial intelligence model to assign more accurate problem category labels to multiple candidate customer service representatives in a timely manner, which helps to improve the accuracy of customer service assignment.

[0143] In summary, in this embodiment, by utilizing feedback information provided by historical service recipients and multiple candidate customer service representatives within the first preset time interval, updated problem category labels are generated, and updated work skill information of each candidate customer service representative is obtained. Based on the updated problem category labels and updated work skill information, the second artificial intelligence model is retrained, which helps to achieve the sustainable development and continuous optimization of the customer service allocation system.

[0144] Training the first artificial intelligence model

[0145] exist Figure 2 The classification method for consultation questions shown utilizes the first artificial intelligence model, and the training method of the first artificial intelligence model will be introduced next.

[0146] In one embodiment, when training the first artificial intelligence model, multiple sample consultation questions are obtained from a question database. These sample consultation questions each have their own question category labels. For a given sample consultation question, the sample consultation question is input into the first artificial intelligence model to predict its question category. Based on the difference between the sample consultation question's question category label and the predicted question category, the model parameters of the first artificial intelligence model are adjusted. That is, the first artificial intelligence model relies on the question category labels for training.

[0147] In one embodiment, after each second preset time period, updated multiple question category tags are generated based on feedback information provided by historical service recipients and multiple candidate customer service representatives within the second preset time period; and an updated question library is generated based on consultation questions raised by historical service recipients within the second preset time period; and the first artificial intelligence model is retrained based on the updated multiple question category tags and the updated question library.

[0148] Within the second preset time period, new questions may arise. For example, if the second preset time period is one week, users may raise new inquiries within that week. This application will retrain the second artificial intelligence model using the updated question database. Similarly, new question category labels may also arise within the second preset time period. This application will also update the question category labels based on feedback information provided by the service recipients of the candidate customer service representatives within the second preset time period, and further update the question category labels based on feedback information provided by the candidate customer service representatives, thereby retraining the first artificial intelligence model.

[0149] Optionally, the second preset duration is the same as the first preset duration, that is, the first artificial intelligence model and the second artificial intelligence model are retrained at the same interval. In this case, the question category of the consultation question predicted by the first artificial intelligence model and the question category label carried by the candidate customer service predicted by the second artificial intelligence model can perfectly match, and there will be no situation where one party has a specific question category while the other party does not.

[0150] like Figure 9 As shown, Figure 9 Three scenarios for retraining the first AI model are shown.

[0151] exist Figure 9 In part (A), the new sample consultation question (i.e., the question that does not exist in the original question library) 901 in the updated question library is input into the first artificial intelligence model 902. The first artificial intelligence model 902 predicts the question category of the new sample consultation question 901 and obtains the third predicted question category 903. Based on the error between the third predicted question category 903 and the old question category label of the sample consultation question (i.e. the question category label that existed before the update) 904, the first artificial intelligence model 902 is retrained.

[0152] exist Figure 9 In part (B), the old sample consultation question (i.e. the question that existed in the original question library) 905 in the updated question library is input into the first artificial intelligence model 902. The first artificial intelligence model 902 predicts the question category of the old sample consultation question 901 and obtains the fourth predicted question category 906. Based on the error between the fourth predicted question category 906 and the new question category label of the sample consultation question (i.e. the question category label that did not exist before the update) 907, the first artificial intelligence model 902 is retrained.

[0153] exist Figure 9In part (C), the new sample consultation question 901 in the updated question library is input into the first artificial intelligence model 902, and the first artificial intelligence model 902 predicts the question category of the new sample consultation question 901 to obtain the third predicted question category 903; based on the error between the third predicted question category 903 and the new question category label 907 of the sample consultation question, the first artificial intelligence model 902 is retrained.

[0154] It is worth noting that, in Figure 9 In the three scenarios shown, the retraining of the first artificial intelligence model in this application involves updating the labels, rather than just updating the question database. In this case, the updated labels enable the retrained first artificial intelligence model to assign more accurate question category labels to the consultation questions in a timely manner, which helps to improve the accuracy of customer service assignment.

[0155] In summary, in this embodiment, every second preset time interval, feedback information provided by historical service objects and multiple candidate customer service representatives within the second preset time interval is used to generate updated problem category tags and obtain an updated problem library. Based on the updated problem category tags and the updated problem library, the first artificial intelligence model is retrained, which helps to achieve the sustainable development and continuous optimization of the customer service allocation system.

[0156] In one embodiment, a first AI model and a second AI model are trained simultaneously in an end-to-end manner. A sample inquiry question from a question database, a sample candidate customer service representative from a pool of candidate customer service representatives, and the question category label of the sample inquiry question, along with the question category label carried by the candidate customer service representative, constitute a four-tuple of data. This data is then used to train both the first and second AI models.

[0157] like Figure 10 As shown, a sample consultation question 1001 is obtained, and the sample consultation question 1001 is input into the first artificial intelligence model 1002 to predict the question category of the sample consultation question, thereby obtaining the target first question category 1003. Based on the difference between the target first question category 1003 and the question category label 1004 of the sample consultation question, a first loss 1005 is constructed.

[0158] Obtain the job skills information 1006 of the sample candidate customer service representatives, input the job skills information 1006 of the sample candidate customer service representatives into the second artificial intelligence model 1007 to predict the problem category corresponding to the sample candidate customer service representatives, and obtain the target second problem category 1008. Based on the difference between the target second problem category 1008 and the problem category label 1009 carried by the sample candidate customer service representatives, construct the second loss 1010.

[0159] The target loss 1011 is obtained by summing the first loss 1005 and the second loss 1010; the first artificial intelligence model 1002 and the second artificial intelligence model 1007 are trained simultaneously based on the target loss 1011.

[0160] Optionally, the question category label 1004 of the sample consultation question and the question category label 1009 carried by the sample candidate customer service are the same question category label.

[0161] In this embodiment of the application, the first artificial intelligence model and the second artificial intelligence model are trained simultaneously in an end-to-end manner. This ensures the consistency of the question categories output by the first and second artificial intelligence models. The training of the first artificial intelligence model utilizes relevant information on customer service classification, while the training of the second artificial intelligence model utilizes relevant information on consultation question classification. The question categories predicted by the two models during inference can be better matched. As a result, the two artificial intelligence models trained are better applicable to the entire customer service allocation process, making the final customer service allocation result more accurate.

[0162] Batch read and write operations

[0163] based on Figure 2 In the optional embodiment shown, step 210 is further included as follows: Step 1.

[0164] Step 1: Read multiple customer service assignment requests in batches, determine which customer service representatives will be assigned to which service objects, and ensure that each service object corresponds to a specific customer service assignment request.

[0165] In this embodiment, the customer service allocation request is not read immediately upon receipt of each request. Instead, after receiving multiple requests, they are merged into one and the merged request is read. This reduces the number of read operations on the customer service allocation requests and allows for a one-time response to a batch of requests, greatly shortening the overall customer service allocation time.

[0166] based on Figure 2 In the optional embodiment shown, step 250 is followed by step two.

[0167] Step 2: Batch output multiple customer service assignment results, with each result corresponding to a different customer service assignment request.

[0168] The customer service allocation result includes the assigned target customer service representatives. In this embodiment, the customer service allocation result is not written out immediately after each one is generated. Instead, multiple customer service allocation results are generated and then written out all at once. This reduces the number of writing operations for the customer service allocation results and allows a batch of customer service allocation results to be written out at once, greatly shortening the overall customer service allocation time.

[0169] In summary, the batch read / write operations described in the above embodiments can significantly shorten customer service allocation time for service scenarios with a wide audience. For service scenarios with a large audience, a large number of customer service allocation requests may be received in a short period. The backend server would need considerable time to read each request and write the allocation result individually. This application, however, performs batch read / write operations, significantly reducing the overall allocation time and I / O operations for a large number of customer service allocation requests.

[0170] Figure 11 This application shows a structural block diagram of a customer service allocation device provided in an exemplary embodiment. The device includes:

[0171] The acquisition module 1101 is used to acquire the inquiry questions of the service recipients and the work status information of multiple candidate customer service representatives;

[0172] The classification module 1102 is used to perform a classification operation on the consultation questions through the first artificial intelligence model to obtain the first question category;

[0173] The filtering module 1103 is used to filter out a first number of candidate customer service representatives from multiple candidate customer service representatives, each carrying a first problem category label, where the first problem category label corresponds to the first problem category.

[0174] The filtering module 1103 is also used to filter out a second number of candidate customer service representatives whose work status meets the work status conditions from multiple candidate customer service representatives based on their respective work status information.

[0175] The determination module 1104 is used to determine the customer service representative who will handle the inquiry based on the intersection of a first number of candidate customer service representatives and a second number of candidate customer service representatives.

[0176] In an optional embodiment, the classification module 1102 is further configured to extract key information from the consultation question using natural language processing technology;

[0177] Based on key information, the problem characteristics of the consultation questions are extracted;

[0178] Inputting the problem features into the first artificial intelligence model yields the first problem category.

[0179] In an optional embodiment, the classification module 1102 is further configured to perform at least one of word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, and semantic analysis on the consultation question to obtain the processed consultation question;

[0180] Extract key information from the processed consultation problem. Key information includes at least one of the following: the topic of the consultation problem, the actions in the consultation problem, and the object of the consultation problem.

[0181] In an optional embodiment, the work status information includes at least one of: real-time idle status, workload status, and work status of recently processed issues.

[0182] In an optional embodiment, the filtering module 1103 is further configured to obtain a work status score for the first candidate customer service representative among multiple candidate customer service representatives, based on a weighted sum of the first work status data, the second work status data, and the third work status data of the first candidate customer service representative; the first work status data is used to indicate the real-time idle status, the second work status data is used to indicate the workload, and the third work status data is used to indicate the work status of recently handled issues.

[0183] From multiple candidate customer service representatives, select those whose work status scores meet the score criteria to obtain the second number of candidate customer service representatives.

[0184] In an optional embodiment, the classification module 1102 is further configured to assign a problem category label to each of the multiple candidate customer service representatives based on their respective job skill information using a second artificial intelligence model. The job skill information includes at least one of historical training, work experience, and historical problem handling.

[0185] In an optional embodiment, the device further includes a training module 1105. The training module 1105 is configured to, after each first preset time interval, generate updated problem category labels based on feedback information provided by historical service recipients and multiple candidate customer service representatives within the first preset time interval; and to obtain updated work skill information for each of the multiple candidate customer service representatives.

[0186] Based on the updated question category labels and the updated job skill information of each of the multiple candidate customer service representatives, the second artificial intelligence model was retrained.

[0187] In an optional embodiment, the acquisition module 1101 is further configured to acquire the question category label carried by each of the multiple candidate customer service representatives;

[0188] The classification module 1102 is also used to adjust the problem category label carried by each candidate customer service representative based on the actual handling effect of each candidate customer service representative on historical issues, so as to obtain the updated problem category label carried by each candidate customer service representative.

[0189] In an optional embodiment, the training module 1105 is further configured to generate updated multiple question category labels based on feedback information provided by historical service recipients and multiple candidate customer service representatives within the second preset time period after each second preset time period; and to generate an updated question library based on the consultation questions raised by historical service recipients within the second preset time period.

[0190] The first AI model is retrained based on the updated multiple question category labels and the updated question library.

[0191] In an optional embodiment, the acquisition module 1101 is further configured to read multiple customer service allocation requests in batches, determine to allocate customer service to multiple service objects, and the multiple service objects correspond one-to-one with the multiple customer service allocation requests.

[0192] In an optional embodiment, the device further includes an output module 1106. The output module 1106 is used to batch write out multiple customer service assignment results, with each customer service assignment result corresponding to a different customer service assignment request.

[0193] In summary, this application uses a first artificial intelligence model to classify inquiries, then matches the first question category obtained from the classification with the question category tags carried by candidate customer service representatives. The final target customer service representative (the one handling the inquiry) will carry the correct question category tag. The first artificial intelligence model, used for question classification, provides more accurate results compared to keyword matching in related technologies. Furthermore, the customer service allocation scheme provided in this application avoids manual classification, speeding up the allocation process and reducing labor costs. Additionally, this application considers the work status information of candidate customer service representatives, avoiding assigning current inquiries to representatives with poor recent work performance. This ensures that the assigned target customer service representative can better resolve inquiries and better serve users.

[0194] In addition, this application will screen out a first number of candidate customer service representatives and a second number of candidate customer service representatives in parallel. The first number of candidate customer service representatives carries the correct question category label, and the second number of candidate customer service representatives meets the working status conditions. Through two parallel screening processes, the final target customer service representatives can be screened out in a distributed manner. Distributed computing helps to shorten the overall customer service allocation time.

[0195] Figure 12 This is a schematic diagram illustrating the structure of a computer device according to an exemplary embodiment. The computer device includes... Figure 1The computer device 1200 includes a second backend server 130. The computer device 1200 includes a Central Processing Unit (CPU) 1201, a system memory 1204 including Random Access Memory (RAM) 1202 and Read-Only Memory (ROM) 1203, and a system bus 1205 connecting the system memory 1204 and the CPU 1201. The computer device 1200 also includes a basic input / output system (I / O system) 1206 to facilitate information transfer between various devices within the computer device, and a mass storage device 1207 for storing the operating system 1213, application programs 1214, and other program modules 1215.

[0196] The basic input / output system 1206 includes a display 1208 for displaying information and an input device 1209 for user input, such as a mouse or keyboard. Both the display 1208 and the input device 1209 are connected to the central processing unit 1201 via an input / output controller 1210 connected to the system bus 1205. The basic input / output system 1206 may also include the input / output controller 1210 for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1210 also provides output to a display screen, printer, or other types of output devices.

[0197] The mass storage device 1207 is connected to the central processing unit 1201 via a mass storage controller (not shown) connected to the system bus 1205. The mass storage device 1207 and its associated computer device-readable media provide non-volatile storage for the computer device 1200. That is, the mass storage device 1207 may include computer device-readable media (not shown), such as a hard disk or a compact disc read-only memory (CD-ROM) drive.

[0198] Without loss of generality, the computer device readable medium may include computer device storage media and communication media. Computer device storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer device readable instructions, data structures, program modules, or other data. Computer device storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD-ROM, digital video disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer device storage media are not limited to the above-mentioned types. The system memory 1204 and mass storage device 1207 described above can be collectively referred to as memory.

[0199] According to various embodiments of this disclosure, the computer device 1200 can also be connected to a remote computer device on a network, such as the Internet. That is, the computer device 1200 can be connected to the network 1211 via a network interface unit 1212 connected to the system bus 1205, or the network interface unit 1212 can be used to connect to other types of networks or remote computer device systems (not shown).

[0200] The memory also includes one or more programs stored in the memory, and the central processing unit 1201 executes the one or more programs to implement all or part of the steps applied to the customer service allocation method.

[0201] This application also provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the customer service allocation method provided in the above method embodiments.

[0202] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the customer service allocation method provided in the above-described method embodiments.

[0203] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0204] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0205] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A customer service assignment method, characterized in that, The method includes: Obtain the customer's inquiry and the work status information of multiple candidate customer service representatives; The consultation question is classified using a first artificial intelligence model to obtain a first question category. From the plurality of candidate customer service representatives, a first number of candidate customer service representatives carrying a first problem category label are selected, wherein the first problem category label corresponds to the first problem category; And, based on the work status information of each of the multiple candidate customer service representatives, a second number of candidate customer service representatives whose work status meets the work status conditions are selected from the multiple candidate customer service representatives. Based on the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives, the customer service representative who will handle the inquiry is determined.

2. The method according to claim 1, characterized in that, The first artificial intelligence model is used to perform a classification operation on the consultation question to obtain a first question category, including: Key information in the consultation question was extracted using natural language processing technology. Based on the key information, the problem characteristics of the consultation question are extracted; The problem features are input into the first artificial intelligence model to obtain the first problem category.

3. The method according to claim 2, characterized in that, The process of extracting key information from the consultation question using natural language processing technology includes: Perform at least one of word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, and semantic analysis on the consultation question to obtain the processed consultation question; Extract the key information from the processed consultation question. The key information includes at least one of the topic of the consultation question, the action in the consultation question, and the object in the consultation question.

4. The method according to any one of claims 1 to 3, characterized in that, The work status information includes at least one of the following: real-time idle status, workload status, and work status of recently processed issues.

5. The method according to claim 4, characterized in that, The step of selecting a second number of candidate customer service representatives whose work status meets the work status conditions from the plurality of candidate customer service representatives based on their respective work status information includes: For the first candidate customer service representative among the plurality of candidate customer service representatives, a work status score is obtained by weighted summation of the first candidate customer service representative's first work status data, second work status data, and third work status data; the first work status data is used to indicate the real-time idle status of the first candidate customer service representative, the second work status data is used to indicate the workload of the first candidate customer service representative, and the third work status data is used to indicate the work status of the first candidate customer service representative in handling recent issues. From the plurality of candidate customer service representatives, those whose work status scores meet the score criteria are selected to obtain the second number of candidate customer service representatives.

6. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Based on the job skill information of each of the multiple candidate customer service representatives, a second artificial intelligence model is used to assign a problem category label to each of the multiple candidate customer service representatives. The job skill information includes at least one of historical training, work experience, and historical problem handling.

7. The method according to claim 6, characterized in that, The method further includes: After each first preset time period, based on the feedback information provided by historical service recipients and the multiple candidate customer service representatives within the first preset time period, updated multiple problem category tags are generated; and updated work skill information of each of the multiple candidate customer service representatives is obtained. The second artificial intelligence model is retrained based on the updated question category labels and the updated job skill information of the multiple candidate customer service representatives.

8. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain the question category tag carried by each of the multiple candidate customer service representatives; Based on the actual handling effect of each candidate customer service representative on historical issues, the issue category label carried by each candidate customer service representative is adjusted to obtain the updated issue category label carried by each candidate customer service representative.

9. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Every second preset time interval, based on the feedback information provided by historical service recipients and the multiple candidate customer service representatives within the second preset time interval, updated multiple question category tags are generated; and based on the consultation questions raised by the historical service recipients within the second preset time interval, an updated question database is generated. The first artificial intelligence model is retrained based on the updated multiple question category labels and the updated question library.

10. The method according to any one of claims 1 to 3, characterized in that, Before obtaining the service recipient's inquiry and the work status information of multiple candidate customer service representatives, the process also includes: Multiple customer service allocation requests are read in batches, and customer service representatives are assigned to multiple service objects. Each service object corresponds to one customer service allocation request.

11. The method according to claim 10, characterized in that, After determining the customer service representative to handle the inquiry based on the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives, the process further includes: Multiple customer service assignment results are written in batches, and each of the multiple customer service assignment results corresponds one-to-one with a multiple customer service assignment request.

12. A customer service distribution device, characterized in that, The device includes: The acquisition module is used to acquire the inquiry questions of the service recipients and the work status information of multiple candidate customer service representatives; The classification module is used to perform a classification operation on the consultation question using a first artificial intelligence model to obtain a first question category; The filtering module is used to filter out a first number of candidate customer service representatives carrying a first problem category label from the plurality of candidate customer service representatives, wherein the first problem category label corresponds to the first problem category; The filtering module is also used to filter out a second number of candidate customer service representatives whose work status meets the work status conditions from the multiple candidate customer service representatives based on their respective work status information. The determining module is used to determine the customer service representative who will handle the inquiry based on the intersection of the first number of candidate customer service representatives and the second number of candidate customer service representatives.

13. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program that is loaded and executed by the processor to implement the customer service assignment method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is loaded and executed by a processor to implement the customer service assignment method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that, The computer program product stores a computer program that is loaded and executed by a processor to implement the customer service assignment method as described in any one of claims 1 to 11.

Images