Knowledge mining method and device, intelligent customer service system, equipment, medium and product

Abstract

The invention provides a knowledge mining method and device, an intelligent customer service system, equipment, a medium and a product, and relates to the technical field of computers, in particular to the technical fields of artificial intelligence, data processing, large models, intelligent customer service and the like. The method comprises the following steps: extracting candidate question and answer pairs and a first service label from an artificial customer service session log, and inputting questions in the candidate question and answer pairs into a question and answer model to determine a basic question and answer pair and a second service label; based on the first service label, the second service label and the basic question and answer pair, determining whether the candidate question and answer pair is a new question and answer pair; and in response to the candidate question and answer pair being a new question and answer pair, updating a knowledge base of the question and answer model based on the candidate question and answer pair. According to the invention, the knowledge base of the question and answer model of the existing intelligent customer service system can be automatically supplemented and optimized.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, specifically to the fields of artificial intelligence, data processing, large models, and intelligent customer service, and particularly to a knowledge mining method, a knowledge mining device, an intelligent customer service system, an electronic device, a computer storage medium, and a computer program product. Background Technology

[0002] With the development of internet applications, intelligent customer service systems have been widely used in various online service scenarios. Existing intelligent customer service systems primarily rely on question-and-answer models based on pre-set question-and-answer pairs in a knowledge base to quickly provide automatic answers. However, in actual business operations, due to the diverse forms of user questions and constantly changing business rules, the knowledge base often suffers from insufficient coverage or incomplete answers, causing the question-and-answer model to fail to provide effective solutions. This not only affects the efficiency of the intelligent customer service system but also reduces the user experience. Summary of the Invention

[0003] This disclosure presents a knowledge mining method and apparatus, an intelligent customer service system, equipment, medium, and product, which can automatically supplement and optimize the knowledge base of existing intelligent customer service system question-answering models.

[0004] In a first aspect, embodiments of this disclosure propose a knowledge mining method, comprising: extracting candidate question-answer pairs and a first business tag from human customer service conversation logs, and inputting the questions in the candidate question-answer pairs into a question-answering model to determine a basic question-answer pair and a second business tag; determining whether the candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair; and updating the knowledge base of the question-answering model based on the candidate question-answer pair if the candidate question-answer pair is a new question-answer pair.

[0005] Secondly, embodiments of this disclosure propose a knowledge mining apparatus, comprising: a question-answer pair acquisition module, configured to extract candidate question-answer pairs and a first business tag from human customer service conversation logs, and input the questions in the candidate question-answer pairs into a question-answer model to determine a basic question-answer pair and a second business tag; a new question-answer pair determination module, configured to determine whether the candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair; and a knowledge mining module, configured to update the knowledge base of the question-answer model based on the candidate question-answer pair in response to the candidate question-answer pair being a new question-answer pair.

[0006] Thirdly, embodiments of this disclosure propose an intelligent customer service system, including: a question-answering model, including a knowledge base; a human customer service conversation log; and a knowledge mining apparatus for performing the methods described in any implementation of the first aspect.

[0007] Fourthly, embodiments of this disclosure provide an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first aspect.

[0008] Fifthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions that enable a computer to perform the method as described in any of the implementations in the first aspect.

[0009] In a sixth aspect, embodiments of this disclosure provide a computer program product including a computer program that, when executed by a processor, can implement the method as described in any of the implementations of the first aspect.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is an exemplary system architecture diagram to which this disclosure can be applied; Figure 2 This is a flowchart of a knowledge mining method provided by some embodiments of this disclosure; Figure 3 This is a flowchart of some embodiments of the present disclosure for determining whether a candidate question-answer pair is a new question-answer pair; Figure 4 This is a flowchart of the knowledge base for updating the question-answering model provided by some embodiments of this disclosure; Figure 5 This is a flowchart of the knowledge base for updating the question-answering model provided in some other embodiments of this disclosure; Figure 6 This is a flowchart of a knowledge mining method provided by some other embodiments of this disclosure; Figure 7 This is a schematic diagram illustrating an application scenario of the knowledge mining method provided in the embodiments of this disclosure; Figure 8 This is a structural block diagram of the knowledge mining apparatus provided in the embodiments of this disclosure; Figure 9This is a schematic diagram of the structure of an electronic device suitable for knowledge mining methods provided by an embodiment of this disclosure. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding; these should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0013] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0014] Figure 1 An exemplary system architecture 100 is shown that can be implemented using the application processing methods and apparatus, electronic devices, storage media and program products disclosed herein.

[0015] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0016] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications for communication between the two devices can be installed on terminal devices 101, 102, and 103, such as cloud storage applications, e-commerce applications, and financial applications.

[0017] Terminal devices 101, 102, and 103 and server 105 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices, and can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here.

[0018] Server 105 can provide various services through various built-in applications. Taking a cloud storage application that can provide cloud storage services as an example, server 105 can perform the following operations through the cloud storage application: extract candidate question-answer pairs and a first business tag from the manual customer service conversation log, and input the questions in the candidate question-answer pairs into the question-answering model to determine the basic question-answer pairs and the second business tag; based on the first business tag, the second business tag, and the basic question-answer pairs, determine whether the candidate question-answer pairs are new question-answer pairs; in response to the candidate question-answer pairs being new question-answer pairs, update the knowledge base of the question-answering model based on the candidate question-answer pairs.

[0019] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0020] Please refer to Figure 2 , Figure 2 A flowchart of a knowledge mining method provided for some embodiments of this disclosure, wherein process 200 may include the following steps: Step 201: Extract candidate question-answer pairs and the first business tag from the human customer service conversation log, and input the questions in the candidate question-answer pairs into the question-answer model to determine the basic question-answer pairs and the second business tag.

[0021] This step is intended for the execution body of the application processing method (e.g., Figure 1 The server 105 shown extracts candidate question-and-answer pairs and the first business label from the human customer service session log, and inputs the questions in the candidate question-and-answer pairs into the question-and-answer model to determine the basic question-and-answer pairs and the second business label.

[0022] In this embodiment, existing intelligent customer service systems typically include chatbots and human customer service representatives. Chatbots usually prioritize handling user inquiries, primarily using question-and-answer models to automatically answer user questions. When a user's question is covered by the question-and-answer model's knowledge base, the chatbot can provide a quick answer. However, if the question is not covered by the knowledge base or the answers in the knowledge base are incomplete, the chatbot cannot provide an effective answer, and the user needs to be transferred to a human customer service representative. Human customer service representatives typically provide more complete and personalized answers. Alternatively, users can choose to consult a human representative directly instead of using a chatbot. When handling user inquiries, human customer service representatives generate a session log containing the user's questions and the human representative's answers.

[0023] In this embodiment, the executing entity can parse the human customer service conversation logs over a period of time, extract the core questions and corresponding answers from each conversation record, and construct candidate question-answer pairs based on the core questions and corresponding answers. The executing entity can process the human customer service conversation logs using an artificial intelligence model to extract candidate question-answer pairs; for example, the artificial intelligence model could be a large language model. Alternatively, the executing entity can also process the human customer service conversation logs using template matching to extract candidate question-answer pairs. The implementation method of extracting candidate question-answer pairs from human customer service conversation logs in this disclosure is not limited.

[0024] In this embodiment, customer service conversation logs typically generate business tags for each conversation record. Business tags are unique category tags set for each conversation record based on preset rules. Business tags are usually keywords or phrases and can be used to distinguish different types of conversation content. Business tags can be single-level tags, such as product inquiries, order queries, return and exchange processing, payment issues, and logistics tracking. Business tags can also be multi-level tags; for example, the first-level tag might be the account scenario, the second-level tags might include basic account attributes and account behavior patterns, the third-level tags for basic account attributes might include account type, registration time, and authentication status, and the third-level tags for account behavior patterns might include login frequency, function usage records, and operating habits. The embodiments disclosed herein do not limit this.

[0025] In this embodiment, the executing entity can also extract business tags corresponding to candidate question-answer pairs from the human customer service conversation log, as the first business tag. The executing entity can extract the first business tag simultaneously with the candidate question-answer pairs from the human customer service conversation log. The method by which the executing entity extracts the first business tag from the human customer service conversation log can be the same as the method for extracting candidate question-answer pairs, for example, both can be achieved using artificial intelligence models. Alternatively, the method by which the executing entity extracts the first business tag from the human customer service conversation log can differ from the method for extracting candidate question-answer pairs. This disclosure does not limit the specific method used.

[0026] In this embodiment, the question-answering model is the main structure of the intelligent customer service system's chatbot for automatically answering user inquiries. The question-answering model includes a knowledge base containing pre-set basic question pairs consisting of standard questions and corresponding answers. When a user inputs a question, the chatbot can use the question-answering model to match the user's input with questions in the knowledge base, identifying similar questions. The chatbot can then use the corresponding answers from the knowledge base as its response to the user's inquiry. The knowledge base may include a Frequently Asked Questions (FAQ) knowledge base. Common question matching methods may include keyword retrieval, BM25 text matching, and semantic similarity calculations such as Word2Vec. The FAQ knowledge base also includes a skill process tree. This skill process tree, through manually configured business processes and rules, calls the backend application programming interface (API) to complete transactional operations, such as querying order status and verifying account permissions.

[0027] In this embodiment, after extracting candidate question-and-answer pairs from the human customer service conversation logs and determining the first business tag for each pair, the executing entity can input the questions from the candidate pairs as the user's inquiry questions into the question-and-answer model. For example, the executing entity can input the questions from the candidate pairs into the question-and-answer model by calling the corresponding API. The question-and-answer model can construct basic question-and-answer pairs by matching the input questions with questions in the knowledge base and outputting similar questions and corresponding answers from the knowledge base. For example, when the executing entity inputs question XXX from the candidate pair into the question-and-answer model, the model will output a question-and-answer pair where the possible question is xxx and the answer is yyy. The executing entity can also obtain the business tag corresponding to the basic question-and-answer pair from the knowledge base as the second business tag.

[0028] Step 202: Based on the first business tag, the second business tag, and the basic question-answer pair, determine whether the candidate question-answer pair is a new question-answer pair.

[0029] Based on step 201, this step aims to have the aforementioned executing entity determine whether a candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair.

[0030] In this embodiment, after determining the candidate question-answer pair, the first business tag, the basic question-answer pair, and the second business tag, the executing entity can match the first business tag with the second business tag and match the candidate question-answer pair with the basic question-answer pair. Based on the matching results, it can determine whether the candidate question-answer pair is similar to the basic question-answer pair in the knowledge base or a new question-answer pair different from the basic question-answer pair in the knowledge base. Specifically, the executing entity can directly match the first business tag with the second business tag and the candidate question-answer pair with the basic question-answer pair to determine whether the candidate question-answer pair is a new question-answer pair. For example, the matching method can include keyword retrieval, BM25 text matching, Word2Vec semantic similarity calculation, etc. Alternatively, the executing entity can input the first business tag, the second business tag, the candidate question-answer pair, and the basic question-answer pair into an artificial intelligence model, and use the artificial intelligence model for matching to determine whether the candidate question-answer pair is a new question-answer pair. For example, the artificial intelligence model can be a large language model. The implementation method of determining whether a candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair is not limited in this disclosure.

[0031] In this embodiment, new question-and-answer pairs can include: entirely new question-and-answer pairs, new question-and-answer pairs, and new answer question-and-answer pairs. Specifically, entirely new question-and-answer pairs can include: question-and-answer pairs where the first business tag and the second business tag do not match, and question-and-answer pairs where the first business tag and the second business tag match but the question and answer do not match. New question-and-answer pairs can include question-and-answer pairs where the first business tag and the second business tag match, the question is different but the answer is the same. New answer question-and-answer pairs can include: question-and-answer pairs where the first business tag and the second business tag match, the question is similar but the answer is different, for example, question-and-answer pairs where the questions are similar and the answers in the candidate question-and-answer pairs contain new information. In this embodiment, the executing entity can further determine the type of the candidate question-and-answer pair in the new question-and-answer pairs based on the first business tag, the second business tag, and the basic question-and-answer pairs.

[0032] Step 203: In response to the candidate question-answer pair being a new question-answer pair, update the knowledge base of the question-answering model based on the candidate question-answer pair.

[0033] Building upon step 202, this step aims to have the aforementioned executing entity update the knowledge base of the question-answering model based on the candidate question-answer pair in response to the candidate question-answer pair being a new question-answer pair.

[0034] In this embodiment, after determining whether a candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair, the executing entity can update the knowledge base of the question-answer model based on whether the candidate question-answer pair is a new question-answer pair and according to the type to which the candidate question-answer pair belongs in the new question-answer pair. Specifically, when the candidate question-answer pair is a completely new question-answer pair, the executing entity can write both the question and the answer in the candidate question-answer pair into the knowledge base of the question-answer model; when the candidate question-answer pair is a new question question-answer pair, the executing entity can write the question in the candidate question-answer pair into the knowledge base of the question-answer model to supplement similar questions in the knowledge base; when the candidate question-answer pair is a new answer question-answer pair, the executing entity can write the answer in the candidate question-answer pair into the knowledge base of the question-answer model to supplement the answers in the knowledge base.

[0035] The knowledge mining method provided in this disclosure obtains candidate question-and-answer pairs by performing knowledge mining on human customer service conversation logs. It then combines business tags to perform similarity and difference analysis between the candidate question-and-answer pairs and the basic question-and-answer pairs in the existing question-and-answer model knowledge base. This determines whether the candidate question-and-answer pairs contain new content. The knowledge base of the question-and-answer model is updated using candidate question-and-answer pairs containing new content. This enables automatic supplementation and optimization of the knowledge base of the existing intelligent customer service system's question-and-answer model, reduces the manual cost of knowledge base maintenance, ensures the accuracy and usability of newly added content, guarantees the quality and stability of knowledge base updates, continuously improves the automatic answering capability of the intelligent customer service system, reduces the proportion of transfers to human customer service, reduces the pressure on human customer service personnel, improves the efficiency of the intelligent customer service system, and enhances the user experience.

[0036] Please refer to Figure 3 , Figure 3 The flowchart for determining whether a candidate question-answer pair is a new question-answer pair provided for some embodiments of this disclosure, wherein determining whether a candidate question-answer pair is a new question-answer pair based on a first service tag, a second service tag, and a basic question-answer pair may include the following steps: Step 301: In response to the first business tag matching the second business tag, determine the matching result between the candidate question-answer pair and the basic question-answer pair.

[0037] This step is intended for the execution body of the component mounting method (e.g., Figure 1 The terminal devices 101, 102, and 103 shown respond to the matching of the first service tag and the second service tag, and determine the matching result of the candidate question-answer pair and the basic question-answer pair.

[0038] In this embodiment, the executing entity can first match the first business tag with the second business tag. Based on the matching of the first and second business tags, the executing entity then matches the candidate question-answer pairs with the basic question-answer pairs to determine the matching result. Matching the candidate question-answer pairs with the basic question-answer pairs may include: matching the questions in the candidate question-answer pairs with the questions in the basic question-answer pairs, and matching the answers in the candidate question-answer pairs with the answers in the basic question-answer pairs. Matching the first business tag with the second business tag indicates that the candidate question-answer pair and the basic question-answer pair belong to the same business category, such as both being product inquiries or both being return and exchange processing.

[0039] Step 302: In response to a matching result of question mismatch and / or answer mismatch, determine the candidate question-answer pair as a new question-answer pair.

[0040] Based on step 301, this step aims to have the aforementioned executing entity determine the candidate question-answer pair as a new question-answer pair in response to the matching result being a question mismatch and / or answer mismatch.

[0041] In this embodiment, the matching result between the candidate question-answer pair and the basic question-answer pair can include: neither the question nor the answer matches, the question matches but the answer does not match, the question does not match but the answer matches, and both the question and the answer match. After determining the matching result between the candidate question-answer pair and the basic question-answer pair based on the matching of the first business tag and the second business tag, the executing entity can determine the candidate question-answer pair as a new question-answer pair based on one of the following matching results: neither the question nor the answer matches, the question matches but the answer does not match, or the question does not match but the answer matches.

[0042] Step 303: In response to the mismatch between the first business tag and the second business tag, determine the candidate question-answer pair as a new question-answer pair.

[0043] Based on step 302, this step aims to have the aforementioned executing entity determine the candidate question-answer pair as a new question-answer pair in response to the mismatch between the first business tag and the second business tag.

[0044] In this embodiment, the executing entity can first match the first business tag with the second business tag. If the first business tag and the second business tag do not match, the candidate question-and-answer pair is determined to be a new question-and-answer pair. A mismatch between the first business tag and the second business tag indicates that the candidate question-and-answer pair and the basic question-and-answer pair belong to different business categories. For example, the candidate question-and-answer pair belongs to product inquiries, while the basic question-and-answer pair belongs to return and exchange processing. In this case, regardless of whether the candidate question-and-answer pair matches the basic question-and-answer pair, the candidate question-and-answer pair is a completely new question-and-answer pair for return and exchange processing.

[0045] This embodiment combines business tags to perform similarity and difference analysis on candidate question-answer pairs and basic question-answer pairs, which can simplify the identification and analysis of candidate question-answer pairs and basic question-answer pairs, improve the efficiency of identification and analysis of new content, and further improve the accuracy of new content.

[0046] Please refer to Figure 4 , Figure 4 A flowchart for updating the knowledge base of a question-answering model provided for some embodiments of this disclosure, wherein updating the knowledge base of the question-answering model based on the candidate question-answer pair in response to the candidate question-answer pair being a new question-answer pair, may include the following steps: Step 401: In response to the candidate question-answer pair being a new question-answer pair, divide the candidate question-answer pair into different business sets based on the first business label.

[0047] This step is intended for the execution body of the component mounting method (e.g., Figure 1 The terminal devices 101, 102, and 103 shown respond to the candidate question-answer pair being a new question-answer pair by dividing the candidate question-answer pair into different service sets based on the first service label.

[0048] In this embodiment, the executing entity can classify candidate question-and-answer pairs into different business sets according to their business type, based on the first business tag of the candidate question-and-answer pairs, since the candidate question-and-answer pairs are new question-and-answer pairs. In an optional example, the first business tag is a multi-level tag, and the executing entity can bin the candidate question-and-answer pairs determined as new question-and-answer pairs according to the first-level tag to form candidate question-and-answer pair sets of different business categories. For example, based on the first-level tag, the executing entity separates member-type candidate question-and-answer pairs from account-type candidate question-and-answer pairs.

[0049] Step 402: Based on the questions in the candidate question-answer pairs, cluster each business set separately.

[0050] Building upon step 401, this step aims to have the aforementioned executing entities cluster each business set based on the questions in the candidate question-answer pairs.

[0051] In this embodiment, after dividing the candidate question-answer pairs into different business sets according to the first business label, the executing entity can cluster the questions in the candidate question-answer pairs for each business set to determine similar sub-questions under each business category. The clustering of each business set can use the DBSCAN clustering algorithm, the k-means clustering algorithm, or hierarchical clustering; the implementers of this disclosure do not limit the specific method used.

[0052] Step 403: Determine the candidate question-answer pairs corresponding to the cluster centers as the target question-answer pairs, and update the knowledge base of the question-answering model based on the target question-answer pairs.

[0053] Based on step 402, this step aims to have the aforementioned executing entity determine the candidate question-answer pairs corresponding to the cluster centers as target question-answer pairs, and update the knowledge base of the question-answering model based on the target question-answer pairs.

[0054] In this embodiment, after clustering each business set according to the questions in the candidate question-answer pairs, the executing entity can take the questions in the candidate question-answer pairs corresponding to each cluster center as representative questions of a class of questions, and determine the corresponding candidate question-answer pairs as target question-answer pairs. The knowledge base of the question-answering model is then updated using these target question-answer pairs. The target question-answer pairs can be entirely new question-answer pairs, new question-answer pairs, or new answer question-answer pairs. Completely new question-answer pairs include new questions and new answers; new question-answer pairs include new questions; and new answer question-answer pairs include new answers.

[0055] In some alternative implementations, the new question-answer pair may include a new question and / or a new answer; updating the knowledge base of the question-answering model based on the target question-answer pair may include: updating the knowledge base of the question-answering model based on the question in the target question-answer pair in response to the target question-answer pair including a new question; and / or updating the knowledge base of the question-answering model based on the answer in the target question-answer pair in response to the target question-answer pair including a new answer.

[0056] This embodiment clusters and organizes new question-answer pairs to determine representative target question-answer pairs, which can effectively reduce repetitive expressions and dispersion of expressions in the question-answer model knowledge base and improve the standardization of the knowledge base.

[0057] Please refer to Figure 5 , Figure 5 A flowchart for updating the knowledge base of a question-answering model provided for other embodiments of this disclosure, wherein updating the knowledge base of the question-answering model based on the candidate question-answering pair in response to a new question-answering pair may include the following steps: Step 501: In response to the candidate question-answer pair being a new question-answer pair, divide the candidate question-answer pair into different business sets based on the first business tag.

[0058] Step 502: Based on the questions in the candidate question-answer pairs, cluster each business set separately.

[0059] Step 503: Determine the candidate question-answer pairs corresponding to the cluster centers as the target question-answer pairs.

[0060] Steps 501, 502, and 503 above are the same as those in the example below. Figure 4 Steps 401, 402, and 403 shown are the same; please refer to the original text for the identical parts. Figure 4 The corresponding parts of the illustrated implementation are not described in detail here.

[0061] Step 504: In response to the target question-answer pair including new answers, rewrite the answers in the target question-answer pair based on the answer style in the base question-answer pair.

[0062] Building upon step 504, this step aims to have the aforementioned executing entity rewrite the answer in the target question-answer pair based on the answer style in the base question-answer pair in response to the target question-answer pair being a new answer.

[0063] In this embodiment, after determining the target question-answer pairs based on clustering, the executing entity can further identify target question-answer pairs that include new answers from the target question-answer pairs, and rewrite the answers in these target question-answer pairs according to the answer style in the basic question-answer pairs to optimize the target question-answer pairs. Specifically, the executing entity can rewrite the answers in the target question-answer pairs according to preset rules based on the answer style in the basic question-answer pairs. Alternatively, the executing entity can input the target question-answer pairs and the basic question-answer pairs into an artificial intelligence model, which will then rewrite the answers in the target question-answer pairs according to the answer style in the basic question-answer pairs. For example, the artificial intelligence model can be a large language model. The embodiments of this disclosure limit the implementation method of rewriting the answers in the target question-answer pairs according to the answer style in the basic question-answer pairs.

[0064] For example, a target question-answer pair is one where the questions are similar but the answers contain new information. The questions and answers from the target pair, along with the questions and answers from the base question-answer pair, can be input into a large language model. The large language model can then further determine whether the answers in the target pair contain valid new information or optimized content based on the answers in the base pair, and rewrite the answers in the base pair according to their specific style. The large language model can then output a new answer that conforms to the style of answers in the existing knowledge base and incorporates new, reusable, valid information.

[0065] Step 505: Review the target question-and-answer pair.

[0066] Based on step 504, this step can be further reviewed by the aforementioned executing entity or by staff members. The implementation method disclosed herein does not limit this.

[0067] In this embodiment, preset review rules can be used to review the target question-and-answer pairs from multiple dimensions. For example, the review can be conducted on the vocabulary used, the readability of the text, and the logic of the questions and answers, in order to further ensure the accuracy and usability of the content of the target question-and-answer pairs. The embodiments of this disclosure do not limit the content of the review.

[0068] Step 506: Update the knowledge base of the question-answering model based on the approved target question-answer pairs.

[0069] Building upon step 505, this step aims to have the aforementioned implementing entity update the knowledge base of the question-answering model based on the approved target question-answer pairs.

[0070] In this embodiment, after reviewing the target question-answer pair, the executing entity can use the approved target question-answer pair to update the knowledge base of the question-answer model. Updating the knowledge base of the question-answer model can include: updating the knowledge base of the question-answer model based on the question in the target question-answer pair in response to the target question-answer pair including a new question; and / or updating the knowledge base of the question-answer model based on the answer in the target question-answer pair in response to the target question-answer pair including a new answer.

[0071] This embodiment rewrites the new answers in the target question-and-answer pairs, which can reaffirm the validity of the added or optimized information in the answers, and remove sensitive or private information involved in the answers, thereby improving the standardization of the target question-and-answer pairs. This allows the optimized question-and-answer pairs to be directly applied to expand the knowledge base. By reviewing the target question-and-answer pairs before updating the knowledge base, this embodiment further ensures the accuracy and usability of the target question-and-answer pairs, guaranteeing the quality and stability of the knowledge base updates.

[0072] Please refer to Figure 6 , Figure 6 A flowchart of a knowledge mining method provided for other embodiments of this disclosure, wherein the method may include the following steps: Step 601: Extract questions and corresponding answers from the human customer service conversation logs, determine candidate question-answer pairs, and determine the first business tag for the candidate question-answer pairs.

[0073] Step 602: Input the questions from the candidate question-answer pairs into the question-answering model, determine the basic question-answer pairs based on the output questions and answers, and determine the second business tag of the basic question-answer pairs.

[0074] Steps 601 and 602 above are the same as those in the example below. Figure 2 The steps shown are the same as in step 201; please refer to the original text for the same parts. Figure 2 The corresponding parts of the illustrated implementation are not described in detail here.

[0075] Step 603: Based on the questions in the candidate question-answer pair, determine the most similar question in the knowledge base and determine the similarity value between the questions in the candidate question-answer pair and the most similar question.

[0076] Based on step 602, this step aims to have the aforementioned executing entity determine the most similar question among the questions in the knowledge base based on the questions in the candidate question-answer pair, and determine the similarity value between the questions in the candidate question-answer pair and the most similar question.

[0077] In this embodiment, after determining candidate question-answer pairs, the executing entity can identify the question most similar to the question in the question-answering model's knowledge base based on the questions in the candidate question-answer pairs, and determine the similarity value between the question in the candidate question-answer pairs and the most similar question. The implementation of this disclosure does not limit the method for determining the most similar question and the similarity value in the knowledge base. In some optional implementations, the executing entity can semantically vectorize the questions in the knowledge base and construct a vector database based on the encoded questions; then semantically vectorize the questions in the candidate question-answer pairs to determine query vectors; subsequently, it can perform a retrieval in the vector database based on the query vectors to determine the most similar question and the similarity value.

[0078] In an optional example, standard questions and similar questions in the existing question-answering model's knowledge base can be semantically vectorized and encoded. Encoding methods could include BM25, TF-IDF, etc. The encoded results are stored in a vector database, forming the basis for retrieval. Then, the questions in the candidate question-answer pairs are embedded to obtain query vectors. Next, the query vectors are retrieved from the vector database to obtain similarity values ​​between the query vector and each vector in the database. The question corresponding to the vector with the highest similarity value is determined as the most similar question in the candidate question-answer pair.

[0079] Step 604: Input the first business label, the second business label, candidate question-answer pairs, basic question-answer pairs, most similar questions, and similarity values ​​into the artificial intelligence model for classification, and the type of candidate question-answer pairs, which includes new question-answer pairs.

[0080] Based on step 603, this step aims to have the aforementioned executing entity input the first business label, the second business label, candidate question-answer pairs, basic question-answer pairs, most similar questions, and similarity values ​​into the artificial intelligence model for classification, and the type of candidate question-answer pairs, which includes new question-answer pairs.

[0081] In this embodiment, after determining the candidate question-answer pairs, the first business tag, the basic question-answer pairs, the second business tag, the most similar question, and the similarity value, the executing entity can input the first business tag, the second business tag, the candidate question-answer pairs, the basic question-answer pairs, the most similar question, and the similarity value into the artificial intelligence model. The artificial intelligence model classifies the candidate question-answer pairs to determine their type. The type of candidate question-answer pairs can include new question-answer pairs and non-new question-answer pairs. New question-answer pairs can also include completely new question-answer pairs, new question-answer pairs, and new answer question-answer pairs. Optionally, to improve the accuracy of the artificial intelligence model's classification, a preset similarity threshold can be input into the artificial intelligence model. The preset similarity threshold can be a similarity value with good performance determined based on experience or experiments.

[0082] Step 605: In response to the candidate question-answer pair being a new question-answer pair, update the knowledge base of the question-answering model based on the candidate question-answer pair.

[0083] The above steps 605 and as follows Figure 2 The steps shown in step 203 are the same; please refer to the original text for the identical parts. Figure 2 The corresponding parts of the illustrated implementation are not described in detail here.

[0084] This embodiment improves the accuracy of AI model classification by adding the most similar question and similarity value as input. Furthermore, by performing finer-grained classification on newly added content in candidate question-answer pairs, it accurately distinguishes between entirely new pairs and supplements to existing ones. This allows for targeted and accurate updates to the knowledge base, effectively avoiding repetitive expressions and further ensuring the accuracy and usability of newly added content, thus guaranteeing the quality and stability of knowledge base updates.

[0085] To deepen understanding, this disclosure also provides a specific implementation scheme in the context of a particular application scenario. In this scenario, the knowledge base is shifted to a FAQ library, such as... Figure 7 As shown, the following steps may be included: 701: Construct a vector database of existing knowledge; 702: Customer service chat log data cleaning and processing; 703: Use the prompt project and LLM to extract core questions and answers; 704: Extract the core question and answer vectors and perform similarity comparison to recall the top 1 similar question among existing knowledge questions; 705: Combining existing recall, vector database recall, and conversation classification tags, use LLM to determine whether a question and answer can be classified as one of the following: [new FAQ, similar questions to existing FAQs, new answer information for existing questions, or no new information]. 706: Clustering questions containing new information and selecting cluster representatives; 707: Combine LLM to rewrite answers based on the specific answer style of machine-generated customer service; 708: Manual review and confirmation 709: New knowledge is written into the customer service system's knowledge base.

[0086] In this implementation example, the chatbot in the intelligent customer service system primarily relies on a pre-set FAQ library and the skill process tree within it to provide automated answers. When a user's question is covered by the FAQ library, the chatbot can respond quickly. However, in actual business scenarios, due to the diverse forms of user questions and constantly changing business rules, the FAQ library often suffers from insufficient coverage or incomplete answers, leading to the chatbot's inability to provide effective solutions, requiring users to be transferred to human customer service. Human customer service representatives typically provide more complete and personalized answers when handling user questions.

[0087] This embodiment, through automated mining and structured organization of human customer service conversation records, can supplement and optimize the existing FAQ database, improve the automatic answering capability of the intelligent customer service system, reduce the pressure on human customer service personnel, and improve the user's service experience.

[0088] The method for automatic mining and knowledge base expansion of question-and-answer pairs based on human customer service conversation records provided in this embodiment can achieve continuous updating and optimization of the FAQ library of the intelligent customer service system by introducing large language models, vector retrieval, cluster analysis and human review mechanism.

[0089] This embodiment can be applied to online customer service systems, such as those in the finance, e-commerce, cloud computing, and telecommunications industries. It can also be applied to the automatic construction and maintenance of enterprise knowledge bases, as well as to online education Q&A systems and intelligent government service consultation systems.

[0090] For example, in a customer service scenario for a certain product, a user asked, "How do I cancel automatic renewal for my membership?" The chatbot was unable to answer and was transferred to a human customer service representative. The human representative then provided the answer: "Open Settings - Account Management - Turn off automatic renewal."

[0091] This embodiment can automatically extract the question-and-answer pair, determine it as a "supplement to existing FAQ answers," cluster it into "value-added / refund" questions, and add it to the FAQ database after manual review. The next time a user raises a similar question, the chatbot can directly provide an answer.

[0092] Further reference Figure 8 As a response to the above Figure 2 The implementation of the method shown in this disclosure provides a knowledge mining apparatus. For example... Figure 8 As shown, the knowledge mining device 800 of this embodiment may include: a question-answer pair acquisition module 801, a new question-answer pair determination module 802, and a knowledge base update module 803. Among them, The question-answer pair acquisition module 801 is configured to extract candidate question-answer pairs and a first business tag from the human customer service conversation log, and input the questions in the candidate question-answer pairs into the question-answering model to determine the basic question-answer pairs and the second business tag. The new question-answer pair determination module 802 is configured to determine whether a candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pairs. The knowledge mining module 803 is configured to update the knowledge base of the question-answering model based on the candidate question-answer pairs in response to the candidate question-answer pairs being new question-answer pairs.

[0093] In this embodiment, the specific processing of the question-answer pair acquisition module 801, the new question-answer pair determination module 802, and the knowledge base update module 803 in the knowledge mining device 800, and the resulting technical effects, can be referred to respectively. Figure 2 The relevant descriptions of steps 201-203 in the corresponding implementation method will not be repeated here.

[0094] In some optional implementations of this embodiment, the question-answering pair acquisition module 801 includes: The candidate question-answer pair acquisition unit is configured to extract questions and corresponding answers from the human customer service conversation log, determine the candidate question-answer pairs, and determine the first business tag of the candidate question-answer pairs; The basic question-answer pair acquisition unit is configured to input the questions from the candidate question-answer pairs into the question-answering model, determine the basic question-answer pairs based on the output questions and answers, and determine the second business tag of the basic question-answer pairs.

[0095] In some optional implementations of this embodiment, the new question-and-answer pair determination module 802 is further configured as follows: In response to the first business tag matching the second business tag, the matching result between the candidate question-answer pair and the basic question-answer pair is determined; In response to the matching result being a question mismatch and / or an answer mismatch, the candidate question-answer pair is determined as a new question-answer pair.

[0096] In some optional implementations of this embodiment, the new question-and-answer pair determination module 802 is further configured as follows: In response to the mismatch between the first service tag and the second service tag, the candidate question-answer pair is determined to be a new question-answer pair.

[0097] In some optional implementations of this embodiment, the knowledge base update module 803 includes: The set partitioning unit is configured to, in response to the candidate question-answer pair being a new question-answer pair, divide the candidate question-answer pair into different business sets based on the first business tag; The clustering unit is configured to cluster each of the business sets based on the questions in the candidate question-answer pairs; The knowledge base update unit is configured to determine the candidate question-answer pairs corresponding to the cluster centers as target question-answer pairs, and update the knowledge base of the question-answering model based on the target question-answer pairs.

[0098] In some optional implementations of this embodiment, the new question-and-answer pair includes new questions and / or new answers; The knowledge base update unit is further configured as follows: In response to the target question-answering pair including a new question, the knowledge base of the question-answering model is updated based on the questions in the target question-answering pair; and / or In response to the target question-answer pair including a new answer, the knowledge base of the question-answering model is updated based on the answer in the target question-answer pair.

[0099] In some optional implementations of this embodiment, the following is also included: The answer rewriting module is configured to rewrite the answer in the target question-answer pair based on the answer style in the base question-answer pair in response to the target question-answer pair including a new answer.

[0100] In some optional implementations of this embodiment, the knowledge base update unit is further configured as follows: The target question-and-answer pairs will be reviewed; The knowledge base of the question-answering model is updated based on the approved target question-answer pairs.

[0101] In some optional implementations of this embodiment, the following is also included: The similarity question determination module is configured to determine the most similar question among the questions in the knowledge base based on the questions in the candidate question-answer pair, and to determine the similarity value between the questions in the candidate question-answer pair and the most similar question; The new question-answer pair determination module 803 is configured to input the first business tag, the second business tag, the candidate question-answer pair, the basic question-answer pair, the most similar question, and the similarity value into an artificial intelligence model for classification, wherein the candidate question-answer pair is of a type, and the type includes the new question-answer pair.

[0102] In some optional implementations of this embodiment, the similar problem determination module includes: The vector database construction unit is configured to semantically vectorize the questions in the knowledge base and construct a vector database based on the encoded questions. The query vector determination unit is configured to perform semantic vectorization encoding on the questions in the candidate question-answer pair to determine the query vector; The similar question retrieval unit is configured to perform a retrieval in the vector database based on the query vector to determine the most similar question and the similarity value.

[0103] According to embodiments of this disclosure, this disclosure also provides an intelligent customer service system, which includes: a question-and-answer model including a knowledge base; a human customer service conversation log; and a knowledge mining device capable of implementing the knowledge mining method described in any of the above embodiments when executed.

[0104] According to embodiments of this disclosure, this disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to implement the knowledge mining method described in any of the above embodiments.

[0105] According to embodiments of this disclosure, this disclosure also provides a readable storage medium storing computer instructions that enable a computer to implement the knowledge mining method described in any of the above embodiments when executed.

[0106] According to embodiments of this disclosure, this disclosure also provides a computer program product that, when executed by a processor, can implement the knowledge mining method described in any of the above embodiments.

[0107] Figure 9 A schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0108] like Figure 9 As shown, device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 902 or a computer program loaded from storage unit 908 into random access memory (RAM) 903. RAM 903 may also store various programs and data required for the operation of device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0109] Multiple components in device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of monitors, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0110] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as knowledge mining methods. For example, in some embodiments, the knowledge mining method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the knowledge mining method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform knowledge mining methods by any other suitable means (e.g., by means of firmware).

[0111] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0112] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0113] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0114] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0115] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0116] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0117] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0118] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A knowledge mining method, comprising: Extract candidate question-and-answer pairs and first business tags from the human customer service conversation logs, and input the questions in the candidate question-and-answer pairs into the question-and-answer model to determine the basic question-and-answer pairs and second business tags; Based on the first business tag, the second business tag, and the basic question-answer pair, determine whether the candidate question-answer pair is a new question-answer pair; In response to the candidate question-answer pair being a new question-answer pair, the knowledge base of the question-answering model is updated based on the candidate question-answer pair.

2. The method according to claim 1, wherein, The step of extracting candidate question-and-answer pairs and a first business tag from the human customer service conversation log, and inputting the questions in the candidate question-and-answer pairs into the question-and-answer model to determine the basic question-and-answer pairs and the second business tag, includes: Extract questions and corresponding answers from the human customer service conversation logs, determine the candidate question-answer pairs, and determine the first business tag of the candidate question-answer pairs; The questions in the candidate question-answer pairs are input into the question-answering model. Based on the output questions and answers, the basic question-answer pairs are determined, and the second business tag of the basic question-answer pairs is determined.

3. The method according to claim 1, wherein, The step of determining whether the candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair includes: In response to the first business tag matching the second business tag, the matching result between the candidate question-answer pair and the basic question-answer pair is determined; In response to the matching result being a question mismatch and / or an answer mismatch, the candidate question-answer pair is determined as a new question-answer pair.

4. The method according to claim 3, further comprising: In response to the mismatch between the first service tag and the second service tag, the candidate question-answer pair is determined to be a new question-answer pair.

5. The method according to any one of claims 1-4, wherein, The step of updating the knowledge base of the question-answering model based on the candidate question-answer pair in response to the candidate question-answer pair being a new question-answer pair includes: In response to the candidate question-answer pair being a new question-answer pair, the candidate question-answer pair is divided into different business sets based on the first business tag; Based on the questions in the candidate question-answer pairs, each of the business sets is clustered. The candidate question-answer pairs corresponding to the cluster centers are determined as target question-answer pairs, and the knowledge base of the question-answering model is updated based on the target question-answer pairs.

6. The method according to claim 5, wherein, The new question-and-answer pair includes new questions and / or new answers; The process of updating the knowledge base of the question-answering model based on the target question-answer pair includes: In response to the target question-answering pair including a new question, the knowledge base of the question-answering model is updated based on the question in the target question-answering pair; and / or In response to the target question-answer pair including a new answer, the knowledge base of the question-answering model is updated based on the answer in the target question-answer pair.

7. The method according to claim 6, further comprising: In response to the target question-answer pair including a new answer, the answer in the target question-answer pair is rewritten based on the answer style in the base question-answer pair.

8. The method according to claim 6, wherein, The process of updating the knowledge base of the question-answering model based on the target question-answer pair includes: The target question-and-answer pairs will be reviewed; The knowledge base of the question-answering model is updated based on the approved target question-answer pairs.

9. The method according to any one of claims 1-4, further comprising: Based on the questions in the candidate question-answer pairs, the most similar question is determined among the questions in the knowledge base, and the similarity value between the questions in the candidate question-answer pairs and the most similar question is determined; The step of determining whether the candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair includes: The first business tag, the second business tag, the candidate question-answer pair, the basic question-answer pair, the most similar question, and the similarity value are input into an artificial intelligence model for classification. The type of the candidate question-answer pair includes the new question-answer pair.

10. The method according to claim 9, wherein, The step of determining the most similar question among the questions in the knowledge base based on the questions in the candidate question-answer pairs, and determining the similarity value between the questions in the candidate question-answer pairs and the most similar question, includes: The questions in the knowledge base are semantically vectorized and encoded, and a vector database is constructed based on the encoded questions. The questions in the candidate question-answer pairs are semantically vectorized and encoded to determine the query vector; Based on the query vector, a search is performed in the vector database to determine the most similar question and the similarity value.

11. A knowledge mining device, comprising: The question-answer pair acquisition module is configured to extract candidate question-answer pairs and a first business tag from the human customer service conversation log, and input the questions in the candidate question-answer pairs into the question-answer model to determine the basic question-answer pairs and the second business tag. The new question-answer pair determination module is configured to determine whether the candidate question-answer pair is a new question-answer pair based on the first business tag, the second business tag, and the basic question-answer pair. The knowledge base update module is configured to update the knowledge base of the question-answering model based on the candidate question-answer pair in response to the candidate question-answer pair being a new question-answer pair.

12. An intelligent customer service system, comprising: Question answering models, including knowledge bases; Customer service conversation logs; as well as A knowledge mining apparatus that performs the knowledge mining method according to any one of claims 1-10.

13. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the knowledge mining method according to any one of claims 1-10.

14. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the knowledge mining method according to any one of claims 1-10.

15. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the knowledge mining method according to any one of claims 1-10.

Images