Question guiding method, server and storage medium

Abstract

The invention belongs to the field of artificial intelligence, and particularly relates to a question guiding method, a server and a storage medium. The question guiding method comprises the following steps: acquiring a recent consultation question set and a recent behavior data set of each user side; based on the recent consultation question set, determining a target question type concerned by each user side; based on the recent behavior data set, determining a target commodity type concerned by each user side; on the basis of the target commodity type and the target question type, determining commodity questions concerned by the user sides, and storing the user side identifiers and the commodity questions concerned by the user sides in a cache database in an associated manner; and after it is detected that the user side enters the customer service system, the commodity problem concerned by the user side is sent to the user side. According to the method, before the user asks questions, the commodity problems concerned by the user are analyzed based on the big data technology and are guided, so that the customer service efficiency and satisfaction are improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence, and specifically relates to a question guiding method, a server, and a storage medium. Background Art

[0002] Currently, many product sales platforms have an online customer service system, and users can consult products by accessing the online customer service system. Since the same question can be expressed in multiple ways in language and text, when the user's expression description deviates greatly from the knowledge in the customer service knowledge base in terms of text description, the similarity judgment will be relatively low, resulting in the customer service knowledge base being unable to recall the correct answer, thus replying incorrect content to the user and greatly reducing the user experience. In addition, many users do not consider the description of the question carefully when asking questions and directly ask with only a few short words. For example, the short question "Unable to charge" does not specify which product, what model, and at which charging link the problem occurs. If the customer service system directly replies based on the current question, there may be 100 potential answers, and it is very difficult to exactly select the answer that the user needs. If hints and guidance can be provided based on the user's vague expression to let the user supplement more conditional information, there is a greater possibility of screening out the answer that the user really wants from these 100 answers.

[0003] To solve the above problems, most existing customer service systems optimize after the user submits the question, which is completely uncontrollable for the user's question and will increase additional time costs and satisfaction costs to clarify the user's intention. Summary of the Invention

[0004] Embodiments of this application provide a question guiding method, a server, and a storage medium, which can improve the efficiency and satisfaction of customer service.

[0005] In a first aspect, embodiments of this application provide a question guiding method, and the method includes: Obtain the recent consultation question sets and recent behavior data sets of each user terminal; Based on the recent consultation question sets, determine the target question types that each user terminal is concerned about; Based on the recent behavior data sets, determine the target product types that each user terminal is concerned about; Based on the target product types and the target question types, determine the product questions that each user terminal is concerned about, and associate and store each user terminal identifier and the product questions that each user terminal is concerned about in a cache database; After detecting that the user terminal enters the customer service system, send the product questions that the user terminal is concerned about to the user terminal, so that the user terminal generates a predicted question list for display according to the product questions that the user terminal is concerned about.

[0006] Optionally, the method further includes: Input the consultation questions collected by the customer service system into the trained question bias classification model for type annotation to obtain the question types of the consultation questions; The determining of the target question types concerned by the user terminal based on the recent consultation question set includes: Statistically analyze each consultation question in the recent consultation question set according to the question type, and select the top N question types with the largest number of questions as the target question types concerned by the user terminal, where N is an integer greater than or equal to 1.

[0007] Optionally, the determining of the commodity questions concerned by each user terminal based on the target commodity type and the target question type includes: Statistically analyze the proportion of the number of questions of each target question type in the recent consultation question set; Obtain the commodity consultation question set corresponding to the target commodity type; Screen out a preset first number of commodity consultation questions from the commodity consultation question set according to the N target question types and the proportion of the number of questions corresponding to each target question type as the commodity questions concerned by each user terminal.

[0008] Optionally, the method further includes: Obtain customer service Q&A pair data and commodity consultation data, merge the customer service Q&A pair data and the commodity consultation data based on the questions to generate all Q&A data, and store the all Q&A data into the cache database and the vector database respectively. The customer service Q&A pair data includes multiple custom questions and corresponding custom answers, and the commodity consultation data includes the consultation questions of each user terminal based on each commodity type and the corresponding reply answers; Perform clustering processing on the consultation questions in the commodity consultation data to generate common commodity questions, and form common commodity Q&A data with the corresponding reply answers and store them in the cache database.

[0009] Optionally, the method further includes: When it is detected that the user terminal manually inputs a question in the customer service system, determine the completion mode based on the manual input method and / or the manual input content; If the completion mode is first-level completion, extract keywords from the manual input content, match the keywords with the question list data in the cache database, and generate a first-level completion question list based on several successfully matched questions and send it to the user terminal; If the completion mode is second-level completion, perform similarity matching on the manual input content and the question list data in the vector database, and generate a second-level completion question list based on several successfully matched questions and send it to the user terminal; If the completion mode is three - level completion, select a number of questions ranked at the top from the common question - and - answer data of products in the cache database to generate a three - level completion question list and send it to the client.

[0010] Optionally, the determining the completion mode based on the manual input method and / or the manual input content includes: When it is detected that the pause duration of the manual input is greater than the first preset pause duration, determine that the completion mode is first - level completion; When it is detected that the pause duration of the manual input is greater than the second preset pause duration and the first - level completion question list is empty, determine that the completion mode is second - level completion, where the second preset pause duration is greater than the first preset pause duration; When it is detected that the manual input content does not include the product type and the second - level completion question list is empty, determine that the completion mode is third - level completion.

[0011] Optionally, the performing a similarity match between the manual input content and the question list data in the vector database, and generating a second - level completion question list based on a number of successfully - matched questions and sending it to the client includes: Perform a similarity match between the manual input content and the question list data in the vector database, and select a number of questions with similarity values exceeding the preset similarity threshold and ranked at the top to generate a second - level completion question list.

[0012] Optionally, the method further includes: Count the number of conversations and the number of messages before and after executing the question - guiding strategy; Calculate the efficiency metrics before and after executing the question - guiding strategy based on the number of conversations and the number of messages before and after executing the question - guiding strategy; Evaluate whether the question - guiding strategy is effective based on the change in the efficiency metrics before and after executing the question - guiding strategy.

[0013] In a second aspect, an embodiment of the present application provides a customer service server, including at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method described above.

[0014] In a third aspect, an embodiment of the present application provides a computer storage medium. The computer storage medium stores instructions or programs. When the instructions or programs are executed by at least one processor, the at least one processor is enabled to execute the method described above.

[0015] In an embodiment of the present application, a question guiding method is provided. First, based on the recent consultation question set of the user terminal, the target question types concerned by the user terminal are determined, and based on the recent behavior data set of the user terminal, the target commodity types concerned by the user terminal are determined. Then, based on the target commodity types and target question types, the commodity questions concerned by each user terminal are determined, and each user terminal identifier and the commodity questions concerned by each user terminal are associated and stored in the cache database. Finally, when it is detected that the user terminal enters the customer service system, the commodity questions concerned by the user terminal are sent to the user terminal, so that the user terminal generates a pre-judgment question list for display according to the commodity questions concerned by the user terminal. The method of the present application analyzes the commodity questions concerned by the user based on big data technology for guidance before the user asks questions, intercepts the questions at the beginning of the upcoming question, saves the time for the user to organize questions, and improves the customer service efficiency and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Exemplarily shows an overall process architecture diagram of question guiding in a customer service system; Figure 2 Exemplarily shows a flowchart of a question guiding method before inputting a question; Figure 3 Exemplarily shows a flowchart of a question guiding method during inputting a question; Figure 4 Exemplarily shows a schematic diagram of question completion in an input box; Figure 5 Exemplarily shows a hardware structure diagram of a customer service server. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0018] The technical terms used in the present application are explained as follows: 1. Before asking a question: It refers to all operations before the user clicks the submit question button.

[0019] 2. After asking a question: It refers to the background processing operations after the user clicks the submit question button.

[0020] 3. Redis: An open-source, in-memory data storage system commonly used as a cache, message queue, database, etc. It is a NoSQL database that supports key-value data structures and also provides various complex data structures such as strings, hashes, lists, sets, sorted sets, bitmaps, etc.

[0021] 4. Vector database: A database system specifically designed for storing, managing, and retrieving high-dimensional vector data. In many applications, especially in the fields of machine learning and artificial intelligence, vectors (usually arrays composed of floating-point numbers) are used to represent the features or semantic information of data such as text, images, audio, and video.

[0022] 5. Energy storage device: A device or system used to store electrical energy, usually composed of battery units, inverters, control systems, etc.

[0023] Please refer to Figure 1 , Figure 1 which shows the customer service system architecture with a question-guiding function. The customer service system includes a data module 1, a pre-judgment module 2, a completion module 3, and a feedback module 4. Among them, the data module 1 is the core component of the customer service system, used to collect, store, and manage different types of data for efficient retrieval and quick response during subsequent customer service processes. The pre-judgment module 2 analyzes the user's behavior in real time through big data, pre-judges the content of the user's question before the user asks the question, and outputs a list of pre-judged questions, so that the user can get the desired answer after clicking. The completion module 3 is used to provide an intelligent auto-completion function when the user enters a question, helping the user to ask questions in a more standardized way, thereby improving the accuracy of question-answer matching and the efficiency of response. The feedback module 4 is used to compare and analyze the efficiency indicators before and after implementing the question-guiding strategy, so as to optimize the question-guiding strategy and improve the efficiency of the customer service system.

[0024] Specifically, the data sources of the data module 1 include product consultation data, user behavior data, and customer service question-and-answer pair data. Among them, the product consultation data includes the consultation questions and corresponding reply answers of each user terminal based on each product type. The user behavior data includes various operation data of each user terminal in the online mall, such as behaviors like purchasing products, adding products to the shopping cart, searching for products, browsing products, favoriting products, and clicking on products, which can be recorded through the method of data logging. The customer service question-and-answer pair data includes multiple custom questions and corresponding custom answers preset by the customer service team. Among them, the custom questions and corresponding custom answers are set by the customer service team based on business experience around the parameters, characteristics, faults, etc. of each product type. For example: Custom question 1: Does the AC180 support waterproofing? Custom reply 1: Yes.

[0025] In the prior art, after obtaining the problem content submitted by the user, it is generally directly retrieved based on the problem content in the product consultation data, user behavior data, and customer service Q&A pair data. After matching similar content, the corresponding answer is returned. This method not only has low retrieval efficiency but also cannot guarantee to give an answer that conforms to the user's expression. To solve this problem, this application guides the user to ask questions in various ways before and during the user's question asking, and processes the product consultation data, user behavior data, and customer service Q&A pair data to obtain various processed data, and stores each processed data in different databases respectively, so as to quickly provide the required data for various ways of guiding questions and improve the efficiency of question guidance.

[0026] In the first aspect, the data module 1 obtains the user-product related Q&A data based on the product consultation data and user behavior data, and stores the user-product related Q&A data in the Redis cache database. When the prediction module 2 detects that the user terminal enters the customer service system, it can quickly obtain the product questions associated with the user terminal from the Redis cache database and send them to the user terminal. Specifically, the data module 1 obtains the recent consultation question set (from the product consultation data) and the recent behavior data set (from the user behavior data) of each user terminal, determines the target question type concerned by the user terminal based on the recent consultation question set, and determines the target product type concerned by the user terminal based on the recent behavior data set; then, based on the target product type and the target question type, determines the product questions concerned by each user terminal, and associates and stores the identifiers of each user terminal, the product questions concerned by each user terminal, and the reply answers in the cache database. Among them, the user terminal identifier may include one or more pieces of information such as the user terminal ID, the IP address of the user terminal, and the geographical location where the user terminal is located. Associating and storing multiple user terminal information with the product questions and reply answers concerned by the user can achieve the extended recommendation of predicted questions. For example, when a user terminal first enters the customer service system for consultation, the product questions associated with other user terminals with the same geographical location as this user terminal can be pushed to this user terminal.

[0027] In the second aspect, the data module 1 uses a clustering algorithm to extract representative consultation questions from the consultation questions included in the product consultation data as common product questions, and then forms the common product Q&A data with the reply answers and stores it in the Redis cache database. When the completion module 3 enters the third-level completion mode, it can generate a third-level completion question list based on the common product Q&A data and send it to the user terminal to provide a question completion function when there is no specific product information.

[0028] In a third aspect, the data module 1 merges the customer service Q&A pair data and the product consultation data based on the questions to generate all Q&A data, and stores all the Q&A data in the Redis cache database and the vector database respectively. Specifically, the data module 1 extracts all the custom questions from the customer service Q&A pair data and all the consultation questions from the product consultation data, merges all the custom questions and all the consultation questions, and then performs deduplication, clustering and other processing to obtain all the question lists. Then, all the Q&A data is generated based on all the question lists and the answers corresponding to each question in all the question lists. When the completion module 3 enters the first-level completion mode, it can extract the keywords in the input content and quickly match them with all the Q&A data in the Redis cache database to generate a first-level completion question list and send it to the user side. When the completion module 3 enters the second-level completion mode, it can extract all the input content and perform similarity matching with all the Q&A data in the vector database to generate a second-level completion question list and send it to the user side. In an embodiment, in the vector database, the HNSW (Hierarchical Navigable Small World) algorithm is used to establish an index for the customer service Q&A pair data, so as to accelerate the nearest neighbor search in the multi-dimensional space.

[0029] In an embodiment, the data module 1 periodically updates the user-product associated Q&A data, the common product Q&A data, and all the Q&A data. For example, the data module 1 obtains various questions consulted by users on the same day at 11 pm every night and updates the above data.

[0030] In an embodiment, the prediction module 2 is used to, after detecting that the user side enters the customer service system, search for the product questions concerned by the user side in the user-product associated Q&A data in the Redis cache database based on the identifier of the user side, and send the product questions concerned by the user side to the user side, so that the user side can generate a prediction question list for display according to the product questions concerned by the user side.

[0031] The completion module 3 is used to provide an intelligent automatic completion function when the user enters a question. Specifically, the completion module 3 includes three levels of completion processes: (1) First-level completion When the user enters a question, the customer service system will query the questions matching the keywords entered by the user in all the Q&A data in the Redis database in real time to generate a first-level completion question list for the user to select. This completion method mainly relies on the fuzzy matching of keywords.

[0032] (2) Second-level completion When the keyword matching fails to find a suitable question, the customer service system will further search for questions with a high similarity to the input content in all the Q&A data of the vector database based on the user's input content, generate a secondary completion question list, and display it to the user. This process uses a vector model for matching to ensure that the system can still provide highly relevant candidate questions when the user enters a more complex or uncommon question.

[0033] (3)Tertiary Completion When the secondary completion fails to find relevant questions, the customer service system will generate a tertiary completion question list based on the common Q&A data of the products in the Redis cache database and display it to the user for selection to help the user quickly find relevant answers. This level of completion is mainly used in scenarios where the user's question is too vague or irrelevant to the product.

[0034] The feedback module 4 is used to compare and analyze the efficiency metrics before and after implementing the question guiding strategy to optimize the question guiding strategy and improve the efficiency of the customer service system. Specifically, first monitor and count the number of conversations and the number of messages before and after implementing the question guiding strategy; then calculate the efficiency metrics before and after implementing the question guiding strategy based on the number of conversations and the number of messages before and after implementing the question guiding strategy, and evaluate whether the question guiding strategy is effective based on the change in the efficiency metrics before and after implementing the question guiding strategy. Among them, the number of conversations refers to the number of dialogue rounds between the user side and the customer service system. For example, a conversation interval exceeding 1 hour is considered the end of one round of conversation. The number of messages refers to the total number of messages between the user side and the customer service system in each round of conversation.

[0035] In one embodiment, the efficiency metric E is defined as the number of messages / the number of conversations. Assume that the efficiency metric after implementing the question guiding strategy is and the efficiency metric before implementing the question guiding strategy is When , that is, the number of messages from the user side in each round of conversation after implementing the question guiding strategy is less than that before the question guiding, it indicates that the question guiding strategy is effective and the communication efficiency of the customer service system has been improved; when , that is, the number of messages from the user side in each round of conversation after implementing the question guiding strategy is more than that before the question guiding, it indicates that the question guiding strategy fails to effectively improve the efficiency and needs to be further optimized.

[0036] In other embodiments, the feedback module 4 is further configured to respectively count the number of times the predicted problem list, the first-level completion problem list, the second-level completion problem list, and the third-level completion problem list are triggered, and respectively count the number of times the problems in the predicted problem list, the first-level completion problem list, the second-level completion problem list, and the third-level completion problem list are selected. By calculating the ratio of the number of times the problems in each type of list are selected to the number of times the list is triggered within a preset time period (for example, one month), the screening strategy of the problems in each type of list is evaluated and optimized.

[0037] Generally, the consultation process of the customer service system includes the following steps: Step 1, the user has a consultation demand for a certain product; Step 2, the user enters a question in the dialogue window of the customer service system; Step 3, the user clicks the question submission button; Step 4, the customer service system displays the reply content in the dialogue window. The method of the present application is for guiding and assisting in the second step before the user clicks the submission button after entering the customer service system (including two stages before and during the input of the question), which can also be called "pre-question" guidance and optimization.

[0038] Please refer to Figure 2 , Figure 2 shows a flowchart of a question guidance method before entering a question. The question guidance method includes: Step S201, obtaining the recent consultation question set and the recent behavior data set of each user terminal.

[0039] In one embodiment, the customer service system collects and processes the consultation questions and reply answers of each user terminal based on various commodity types based on big data technology to obtain commodity consultation data, and collects and processes various operation data of each user terminal in the online mall (for example, records user behaviors such as purchasing commodities, adding commodities to the shopping cart, searching for commodities, browsing commodities, collecting commodities, and clicking on commodities through the method of data embedding) to obtain user behavior data.

[0040] In one embodiment, the commodity consultation data records the initiation time of each consultation question. According to the initiation time, the consultation questions initiated within the most recent preset time period can be screened out from the commodity consultation data to form a recent consultation question set. The user behavior data records the operation time of each behavior. According to the operation time, the user behaviors that occurred within the most recent preset time period can be screened out from the user behavior data to form a recent behavior data set. For example, the recent consultation question set includes the consultation questions initiated by the user terminal within the most recent 3 months, and the recent behavior data set includes the behavior data such as purchasing commodities, adding commodities to the shopping cart, searching for commodities, browsing commodities, collecting commodities, and clicking on commodities by the user terminal in the online mall within the most recent 3 months.

[0041] Step S202, based on the recent consultation question set, determining the target problem types concerned by each user terminal.

[0042] In one embodiment, the consultation questions collected by the customer service system are input into the trained question bias classification model for type annotation to obtain the question type of the consultation question. The stored content corresponding to each consultation question in the product consultation data may include question content, answer content, product type, client identifier, question type, initiation time, etc. The specific implementation process of step S202 is as follows: count each consultation question in the recent consultation question set according to the question type, and select the top N question types with the largest number of questions as the target question types concerned by the client, where N is an integer greater than or equal to 1.

[0043] Among them, the training method of the question bias classification model includes: Step 1, annotate the types of each question in the training samples. For example, the annotated question types include: 1) Technical preference type: Provide common questions such as technical parameters, product functions, and setting guides.

[0044] 2) Price-sensitive type: Display relevant questions such as discount information, promotional activities, and price comparisons.

[0045] 3) High-frequency purchase type: Provide questions such as after-sales service, product warranty, and replacement parts.

[0046] 4) Casual browsing type: Provide relevant questions such as product design, usage experience, and brand stories.

[0047] 5) Evaluation-driven type: Provide product evaluations, user feedback, and scoring questions.

[0048] 6) Health / environmental concern type: Provide questions such as material sources, environmental certifications, and harmlessness.

[0049] 7) Brand loyalty type: Provide questions such as brand history, user loyalty programs, and brand activities.

[0050] Annotation examples: 1) Question: "Does this headset support noise cancellation?" Annotated as "Technical preference type" 2) Question: "Is there any discount on the price of this mobile phone?" Annotated as "Price-sensitive type" 3) Question: "How is the wearing experience of the headset?" Annotated as "Evaluation-driven type" 4) Question: "Can the mobile phone be returned or exchanged?" Annotated as "High-frequency purchase type" Step 2, use the Naive Bayes algorithm to train the model with the annotated questions to obtain the trained question bias classification model.

[0051] Step S203, based on the recent behavior data set, determine the target product types concerned by each client.

[0052] In one embodiment, first, based on the commodity purchase data in the recent behavior dataset, determine the commodities recently purchased by the user terminal. If any exist, use such commodities as the target commodity types concerned by the user terminal; if none exist, search in the recent behavior dataset for the commodities most frequently searched by the user terminal. If any exist, use such commodities as the target commodity types concerned by the user terminal; if none exist, search in the recent behavior dataset for the commodities most frequently browsed by the user terminal. If any exist, use such commodities as the target commodity types concerned by the user terminal. If none exist, continue to search in the recent behavior dataset for the commodities added to the shopping cart, clicked on, or favorited by the user terminal, and so on. If the target commodity types concerned by the user terminal cannot be found in the recent behavior dataset, expand the search time range and continue to use the above search logic to search for the target commodity types concerned by the user terminal in an earlier behavior dataset (for example, if the default recent behavior dataset is the behavior dataset within 3 months, the earlier behavior dataset can be expanded to the behavior dataset within 6 months) until found. In one embodiment, the target commodity types concerned by the user terminal are one or more.

[0053] Step S204: Based on the target commodity types and the target question types, determine the commodity questions concerned by each user terminal, and store the associations between each user terminal identifier and the commodity questions concerned by each user terminal in the cache database.

[0054] Specifically, first, count the proportion of the number of questions of each target question type in the recent consultation question set, then search based on the target commodity types in the commodity consultation data to obtain the commodity consultation question set corresponding to the target commodity types, and then screen out a preset first number of commodity consultation questions from the commodity consultation question set according to the N target question types and the proportion of the number of questions corresponding to each target question type as the commodity questions concerned by each user terminal. The first number is the number of questions included in the pre-judged question list displayed on the user terminal. For example, assume the first number is 10, the target commodity types are A and B, and the target question types are X, Y, and Z respectively; after statistics, the proportion of the number of questions of the target question types X, Y, and Z in the recent consultation question set is 5:3:2; the commodity consultation question set corresponding to the target commodity type A or B contains a total of 100 consultation questions. Then, select 5 consultation questions of question type X, 3 consultation questions of question type Y, and 2 consultation questions of question type Z from the 100 consultation questions as the commodity questions concerned by the user terminal.

[0055] In one embodiment, each user terminal identifier is associated with the commodity questions that each user terminal is concerned about to form user-commodity related question and answer data and is stored in a cache database, such as a redis cache database. The user terminal identifier may include one or more information such as the user terminal ID, the user terminal IP address, and the user terminal's geographical location. By associating and storing a variety of user terminal information with the commodity questions that the user is concerned about and the replies, it is possible to achieve an expanded recommendation of the predicted questions.

[0056] In other embodiments, each user terminal identifier is associated with the product questions and replies of each user terminal to form user-product related question and answer data and stored in the cache database. The replies corresponding to the product questions of the user terminal are stored together, so that when the user clicks on the selected question in the predicted question list, the reply corresponding to the question can be quickly obtained from the cache database and returned to the user terminal.

[0057] Step S205, after detecting that the user terminal has entered the customer service system, the product issues that the user terminal is concerned about are sent to the user terminal, so that the user terminal generates a predicted question list for display according to the product issues that the user terminal is concerned about.

[0058] Specifically, when a user has a need to consult about a product and opens the customer service system dialogue window, the monitoring event of the front-end button will be triggered. The event will carry the user terminal identifier in the context (such as the user terminal ID, user terminal IP, user terminal geographic location information, etc.). The user terminal initiates a pre-judgment query request to the customer service system based on the user terminal identifier. The customer service system queries the user and product-related question and answer data in the Redis database based on the user terminal identifier, obtains the product questions that the user terminal is concerned about, and sends it to the user terminal. The user terminal generates a pre-judgment question list based on the product questions that the user terminal is concerned about and displays it in the drop-down list of the dialogue window. For example, the pre-judgment question list formed by the product questions that a user is concerned about is as follows: Question 1: Does phone A support wireless charging? Question 2: Can I replace the memory card after purchasing phone A? Question 3: How to transfer contacts from old phone to new phone A? Question 4: How long is the battery life of mobile phone A? Question 5: Does mobile phone A have dual SIM dual standby function? Question 6: Does earphone B support Bluetooth 5.0? Question 7: Are the B wireless earphones waterproof? Question 8: Does the charging box of earphone B support fast charging? Question 9: Does earphone B support noise reduction function? Question 10: How to reset the B wireless earphones? After the user sees the pre-judged question list, if there is a question in the pre-judged question list that the user wants to consult, the user can click on the question to display the question in the chat dialog box and give a prompt that the system is querying. Then, the customer service system obtains the corresponding reply answer from the user-product associated Q&A data in the cache database and returns it to the chat window. In this way, the pre-judgment and response of the user's question are completed, and the user's question is solved in an efficient manner.

[0059] When there is no question in the pre-judged question list that the user wants to consult, the user will continue to manually enter the question content in the input box. To further guide the customer's questions, the embodiment of the present application also provides a method for guiding questions in the input question, such as Figure 3 shown, the method for guiding questions includes: Step S301, when it is detected that the user enters a question manually in the customer service system, obtain the manual input method and / or the manual input content.

[0060] Among them, the manual input method mainly includes the pause duration of the manual input.

[0061] For the input question stage, the present application provides an intelligent auto-completion function. Based on the manual input method and manual input content of the user in the input box, the input content of the user is completed according to three levels of completion modes. To cope with the data required for different levels of completion modes, the method of the present application also performs the following data preparations: (1) Use the clustering algorithm to extract representative consultation questions from the consultation questions included in the product consultation data as common product questions, and then form the common product question and reply answer into common product Q&A data and store it in the cache database; (2) Merge the customer service Q&A pair data and the product consultation data based on the questions to generate all Q&A data, and store all Q&A data in the cache database and the vector database respectively. Specifically, all custom questions are extracted from the customer service Q&A pair data and all consultation questions are extracted from the product consultation data. After merging all custom questions and all consultation questions, de-duplication, clustering and other processing are performed to obtain all question lists, and all Q&A data are generated based on all question lists and the answers corresponding to each question in all question lists.

[0062] Step S302, when the pause duration of the manual input is greater than the first preset pause duration, enter the first-level completion mode.

[0063] When the manually input pause duration is greater than the first preset pause duration, it is determined that the current completion mode is first-level completion, and the first-level completion mode is entered. In the first-level completion mode, the customer service system extracts keywords from the manually input content, matches the keywords with all the Q&A data in the cache database, and generates a first-level completion question list based on several successfully matched questions. Assuming that the first-level completion question list contains at most a preset second quantity of questions, when matching through the keywords in the cache database, if the number of successfully matched questions is less than the second quantity, a first-level completion question list is generated based on all the successfully matched questions; otherwise, the second quantity of questions is selected from the successfully matched questions to generate a first-level completion question list.

[0064] For example, when the user starts to pause for 0.5 s after entering "EP800 charging" in the input box, the first-level completion question list generated based on the keyword "EP800 charging" is as follows: · How fast is the charging speed of EP800? · How to correctly use the EP800 charger for charging? · What safety precautions need to be taken when charging EP800? · Can the power supply be used simultaneously when charging EP800? · What voltage ranges does the EP800 charger support? · Will the battery life be affected when charging EP800? · Will it automatically power off after the EP800 is fully charged? · What should I do if overheating occurs during the charging process of EP800? · How long does it take to fully charge the EP800? · Does the EP800 charging support fast charging technology? Based on the above example, it can be seen that all the questions in the first-level completion question list contain the words "EP800 charging" input by the user.

[0065] Step S303, determine whether the first-level completion question list is empty. If it is, enter step S304; if not, send the first-level completion question list to the user terminal.

[0066] When the first-level completion question list is not empty, send the first-level completion question list to the user terminal, and the user terminal displays it in the pull-up list of the input box. As Figure 4As shown in the figure, after the customer enters "AC18" in the input box, the customer service system background matches the keyword "AC18" to obtain two completion questions: "How to charge AC180?" and "Can AC180 be used with B230?" The user end displays the two completion questions in the drop-down list of the input box. If the drop-down list contains the question the user wants to ask, the user can select and click the question. After clicking, the original content in the input box is cleared, and the selected question is filled in the input box. After the user clicks Submit, the customer service system finds the corresponding answer based on the question in all the question and answer data in the cache database and returns it to the user end. Since the questions in the first-level completion question list are all derived from the questions in all the question and answer data, the correct answers can usually be accurately recalled, and then the correct answers are displayed in the chat window with the user.

[0067] Step S304: When the manually input pause duration is greater than the second preset pause duration, the secondary completion mode is entered.

[0068] When there is a lack of questions matching the keywords in the customer service system, or the user inputs too much content at one time (lack of pauses, failure to trigger the first-level completion conditions), the first-level completion question list will be empty when the keywords are used in the cache database. In this case, the second-level completion is used. The conditions that must be met to enter the second-level completion mode include: (1) detecting that the user manually inputs a question in the customer service system (i.e., the input box contains input content); (2) the first-level completion question list is empty; (3) the manually input pause duration is greater than the second preset pause duration, and the second preset pause duration is greater than the first preset pause duration. For example, the first preset pause duration is 0.5s, and the second preset pause duration is 1s.

[0069] In the secondary completion mode, the customer service system obtains the manual input content in the input box of the user, matches the manual input content with all the question and answer data in the vector database for similarity, and selects several questions with similarity values ​​exceeding a preset similarity threshold (for example, the similarity threshold is 30%) and with high similarity value ranking to generate a secondary completion question list.

[0070] For example, when the user enters "EP800 charging time" in the input box and pauses for 1 second, the list of secondary completion questions generated based on the input content "EP800 charging time" is as follows: How long does it take to fully charge the EP800? What is the maximum charging time of EP800? How long does it take to fully charge the EP800? How long does it take to charge the EP800? How long does it take to fully charge the EP800 under standard conditions? · How long does the charging process of EP800 take? · How long does it take for EP800 to charge from 0% to 100%? · How long does it take to fully charge the EP800 battery? · What factors will affect the charging time of EP800? · What is the shortest charging time of EP800? Based on the above examples, it can be seen that the questions in the secondary completion question list and the user's input content may not be exactly the same in words, but they convey similar meanings.

[0071] Step S305, determine whether the secondary completion question list is empty. If it is, go to step S306; if not, send the secondary completion question list to the client.

[0072] When the secondary completion question list is not empty, send the secondary completion question list to the client and display it in the pull-up list of the input box. The user judges whether there are questions they want to consult based on the secondary completion question list. If so, the user selects the question with the mouse. After selection, the complete question is filled into the input box. After the user clicks submit, the customer service system searches for the corresponding answer in all the Q&A data in the vector database and returns it to the client. Since the questions in the secondary completion question list all come from the questions in all the Q&A data, the correct answer can usually be accurately recalled, and then the correct answer is displayed in the chat window with the user.

[0073] Step S306, when the manually input content does not contain the product type, enter the tertiary completion mode.

[0074] When neither the primary completion mode nor the secondary completion mode can find relevant questions, tertiary completion can be adopted. Among them, the conditions for entering the tertiary completion mode include: (1) detecting that the client manually enters a question in the customer service system (that is, the input box contains input content); (2) the secondary completion question list is empty; (3) the manually input content does not contain the product type.

[0075] In the tertiary completion mode, the customer service system queries the common Q&A data of the product in the cache database, obtains a number of questions with the highest rankings, and generates a tertiary completion question list to send to the client.

[0076] For example, the tertiary completion question list is as follows: · How to improve the service life of the energy storage battery? · What suggestions are there for the maintenance of the energy storage battery? · What are the common faults and troubleshooting methods of the energy storage system? · What models are there for energy storage devices? · What matters need attention when purchasing energy storage products? · What is the difference between energy storage batteries and traditional batteries? · How to determine whether the energy storage device needs to replace the battery? · How should the energy storage device be handled when it fails? · What are the safety precautions for using energy storage devices? · How does the energy storage device perform long-term storage? Based on the above examples, it can be seen that the questions in the three-level complement question list have nothing to do with the product type and are general questions applicable to all products. Users judge whether they have questions they want to consult based on the three-level complement question list. If so, they select the question with the mouse, and after selection, the complete question is filled into the input box. After the user clicks submit, the customer service system searches for the corresponding answer in the common question and answer data of the product in the cache database and returns it to the user side. Since the questions in the three-level complement question list all come from the questions in the common question and answer data of the product, the correct answer can usually be accurately recalled, and then the correct answer is displayed in the chat window with the user.

[0077] If the user still cannot find the question they want to consult in the three-level complement question list, they can ignore these question lists and continue to enter a custom question in the input box and then submit it.

[0078] In one embodiment, the question guiding method of the embodiment of the present application further includes: counting the number of conversations and the number of messages before and after executing the question guiding strategy, calculating the efficiency indicators before and after executing the question guiding strategy based on the number of conversations and the number of messages before and after executing the question guiding strategy, and evaluating whether the question guiding strategy is effective based on the change of the efficiency indicators before and after executing the question guiding strategy. Specifically, the efficiency indicator E is defined as the number of messages / the number of conversations. Assume that the efficiency indicator after executing the question guiding strategy is and the efficiency indicator before executing the question guiding strategy is When , that is, the number of messages of the user side in each round of conversation after executing the question guiding strategy is less than that before the question guiding, it means that the question guiding strategy is effective and the communication efficiency of the customer service system has been improved; when , that is, the number of messages of the user side in each round of conversation after executing the question guiding strategy is more than that before the question guiding, it means that the question guiding strategy fails to effectively improve the efficiency and needs to be further optimized.

[0079] In other embodiments, the question guiding method of the embodiments of the present application further includes: respectively counting the number of times the predicted question list, the first-level complement question list, the second-level complement question list, and the third-level complement question list are triggered, and respectively counting the number of times the questions in the predicted question list, the first-level complement question list, the second-level complement question list, and the third-level complement question list are selected. Based on the number of times the questions in each type of list are selected and the number of times the list is triggered within a preset time period (such as one month), the screening strategy of the questions in each type of list is evaluated and optimized.

[0080] The question guiding method provided by the embodiments of the present application first determines the target question type concerned by the user side based on the recent consultation question set of the user side, and determines the target commodity type concerned by the user side based on the recent behavior data set of the user side; then determines the commodity questions concerned by each user side based on the target commodity type and the target question type, and stores the association between each user side identifier and the commodity questions concerned by each user side in the cache database; finally, when it is detected that the user side enters the customer service system, the commodity questions concerned by the user side are sent to the user side, so that the user side generates a predicted question list for display according to the commodity questions concerned by the user side. The method of the present application analyzes the commodity questions concerned by the user based on big data technology and guides them before the user asks questions, intercepts the questions at the beginning of the upcoming question, saves the time for the user to organize questions, and improves the customer service efficiency and satisfaction.

[0081] According to the embodiments of the present application, a customer service server is provided, as Figure 5 shown, which is a schematic hardware structure diagram of a customer service server provided by the embodiments of the present application. The customer service server 100 includes a processor 10, a memory 20, and a communication interface 30. The processor 10, the memory 20, and the communication interface 30 are connected by lines. In Figure 5 the embodiment shown, the processor 10, the memory 20, and the communication interface 30 are communicatively connected to each other through a bus.

[0082] The memory 20 is used to store software programs, computer-executable program instructions, etc. The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the customer service server, etc.

[0083] The memory 20 can be a read-only memory (ROM), or other types of static storage devices that can store static information and instructions, or a random access memory (RAM), or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM). Specifically, it is not limited here.

[0084] Exemplarily, the aforementioned memory 20 can be a double data rate synchronous dynamic random access memory DDR SDRAM (abbreviation: DDR). This memory 20 can exist independently, but is connected to the processor 10. Optionally, this memory 20 can also be integrated with the processor 10. For example, integrated within one or more chips.

[0085] In some embodiments, the memory 20 optionally includes a memory remotely set relative to the processor 10, and these remote memories can be connected to the customer service server through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and their combinations.

[0086] The processor 10 connects various parts of the entire customer service server 100 using various interfaces and lines. By running or executing software programs stored in the memory 20, and by invoking data stored in the memory 20, it executes various functions of the customer service server and processes data, such as implementing the method described in any embodiment of the present application.

[0087] The processor 10 can be a field programmable gate array (FPGA), a digital signal processor (DSP), a central processing unit (CPU), etc.

[0088] The processor 10 can be a single-core processor or a multi-core processor. For example, the processor 10 can be composed of multiple FPGAs or multiple DSPs. In addition, the processor 10 can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions). The processor 10 can be a separate semiconductor chip or integrated with other circuits into a semiconductor chip. For example, it can form a system on a chip (SoC) with other circuits (such as codec circuits, hardware acceleration circuits, or various bus and interface circuits), or can be integrated as a built-in processor of an application specific integrated circuit (ASIC) in the ASIC. The ASIC integrated with the processor can be separately packaged or packaged together with other circuits.

[0089] The communication interface 30 can use a transceiver device such as a transceiver to achieve communication between the customer service server and other devices or communication networks.

[0090] The embodiments of the present application also provide a non-volatile computer-readable storage medium storing computer-executable instructions, which are executed by one or more processors. For example, the steps of the question guiding method described above are executed.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them; under the idea of the present application, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of the present application as described above. For the sake of brevity, they are not provided in detail; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A question guiding method, characterized in that: The method comprises: Obtain the recent consultation question set and recent behavior data set of each user terminal; Based on the recent consultation question set, determining the target question type that each user terminal is concerned about; Based on the recent behavior data set, determining the target product type that each user terminal is interested in; Based on the target commodity type and the target question type, determining the commodity question that each user terminal is concerned about, and associating each user terminal identifier with the commodity question that each user terminal is concerned about and storing it in a cache database; After detecting that the user terminal has entered the customer service system, the commodity issues that the user terminal is concerned about are sent to the user terminal, so that the user terminal generates a predicted question list for display based on the commodity issues that the user terminal is concerned about.

2. The method according to claim 1, characterized in that: The method further comprises: Input the consulting questions collected by the customer service system into the trained question bias classification model for type labeling to obtain the question type of the consulting question; Determining the target question type that the user is concerned about based on the recent consultation question set includes: Statistics are collected for each consulting question in the recent consulting question set according to question type, and N question types with the highest number of questions are selected as target question types of interest to the user end, where N is an integer greater than or equal to 1.

3. The method according to claim 2, characterized in that The determining of the commodity issues that each user terminal is concerned about based on the target commodity type and the target issue type includes: Count the proportion of each target question type in the recent consultation question set; Obtain a set of commodity consultation questions corresponding to the target commodity type; According to the N target question types and the proportion of the number of questions corresponding to each target question type, a preset first number of product consulting questions are screened out from the product consulting question set as the product questions that each user terminal is concerned about.

4. The method according to claim 1, characterized in that: The method further comprises: Acquire customer service question and answer data and product consultation data, merge the customer service question and answer data and the product consultation data based on the questions to generate all question and answer data, and store all question and answer data in the cache database and the vector database respectively, wherein the customer service question and answer data includes multiple custom questions and corresponding custom answers, and the product consultation data includes consultation questions of each user terminal based on each product type and corresponding reply answers; Clustering is performed on the consulting questions in the product consulting data to generate product FAQs, and the product FAQs and corresponding reply answers are formed into product FAQ data and stored in the cache database.

5. The method according to claim 4, characterized in that The method further comprises: When detecting that the user terminal manually inputs a question in the customer service system, determining a completion mode based on the manual input method and / or the manual input content; If the completion mode is primary completion, extract keywords from the manually input content, match the keywords with the question list data in the cache database, generate a primary completion question list based on a number of successfully matched questions, and send it to the user terminal; If the completion mode is secondary completion, the manually input content is matched with the question list data in the vector database for similarity, and a secondary completion question list is generated based on a number of successfully matched questions and sent to the user terminal; If the completion mode is three-level completion, a number of top-ranked questions are selected from the commodity FAQ data in the cache database to generate a three-level completion question list and send it to the user terminal.

6. The method according to claim 5, characterized in that Determining the completion mode based on the manual input method and / or the manual input content includes: When it is detected that the pause duration of the manual input is greater than the first preset pause duration, determining the completion mode to be the first-level completion; When it is detected that the pause duration of the manual input is longer than the second preset pause duration and the first-level completion question list is empty, determining that the completion mode is the second-level completion, and the second preset pause duration is longer than the first preset pause duration; When it is detected that the manually input content does not contain the product type and the secondary completion question list is empty, the completion mode is determined to be the tertiary completion.

7. The method according to claim 5, characterized in that The performing similarity matching between the manually input content and the question list data in the vector database, and generating a secondary completion question list based on a number of successfully matched questions and sending the list to the user terminal comprises: The manually input content is matched with the question list data in the vector database for similarity, and a number of questions whose similarity values ​​exceed a preset similarity threshold and whose similarity values ​​are ranked high are selected to generate a secondary completion question list.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Count the number of conversations and messages before and after the question-guided strategy is implemented; The efficiency index before and after the question-asking guidance strategy is calculated based on the number of conversations and messages before and after the question-asking guidance strategy is implemented; The effectiveness of the question-guiding strategy was evaluated based on the changes in efficiency indicators before and after the implementation of the question-guiding strategy.

9. A customer service server, characterized in that: The method comprises at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that: The computer storage medium stores instructions or programs, and when the instructions or programs are executed by at least one processor, the at least one processor is caused to perform the method according to any one of claims 1 to 8.

Images