Customer service interaction optimization method and device, equipment, storage medium and program product

Abstract

The invention provides a customer service interaction optimization method and device, equipment, a storage medium and a program product, and relates to the technical field of computers. The customer service interaction optimization method comprises the following steps: in response to a received user question, performing intention recognition on the user question based on a mental model; determining a corresponding domain knowledge base for semantic enhancement based on a result of intention recognition; rewriting the user question based on a query operation on the semantic space alignment spectrogram to obtain to-be-retrieved information; retrieving a corresponding field knowledge base based on the to-be-retrieved information to obtain a business knowledge fragment for enhancing the user question; and inputting the business knowledge fragments into the large language model for response, and outputting reply information of the user question. Through the technical scheme disclosed by the invention, the semantic barrier between the knowledge base and the questions of the user can be effectively made up, the understanding accuracy of the big language model on the intention of the user is improved, and the use experience of the user in interaction of the intelligent customer service system is further improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a customer service interaction optimization method, a customer service interaction optimization device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In order to improve user experience and response accuracy, large language model technology can be applied to question-answering systems. When using a large language model to answer user questions, the knowledge base in the relevant field will be used. Since there is a gap between the standardized business information stored in the knowledge base and the language and background information used by users to ask questions, there is a storage difference between the semantic space of ordinary users and the semantic space of the knowledge base, which leads to a deviation in the large language model's understanding of user intentions, causing users to feel confused and dissatisfied during the interaction process, affecting the user's experience of the question-answering system.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of the present disclosure is to provide a customer service interaction optimization method, a customer service interaction optimization device, an electronic device, a computer-readable storage medium and a computer program product, which at least to some extent overcome the problem of the deviation in the understanding of user intention by the large language model in the customer service intelligent question-answering system in the related art.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0006] According to one aspect of the present disclosure, a customer service interaction optimization method is provided, including: in response to a received user question, identifying the intent of the user question based on a mental model; determining a corresponding domain knowledge base for semantic enhancement based on a result of the intent identification; rewriting the user question based on a query operation on a semantic space alignment spectrum to obtain information to be retrieved, the semantic space alignment spectrum being generated based on graph alignment of user question record logs and knowledge base documents in a corresponding field; retrieving the corresponding domain knowledge base based on the information to be retrieved to obtain a business knowledge fragment that enhances the user question; inputting the business knowledge fragment into a large language model for response, and outputting answer information for the user question.

[0007] In one embodiment of the present disclosure, before performing intent recognition on a received user question based on a mental model, the method further includes: performing triple information extraction on the user question record log to obtain a first group of information; performing triple information extraction on the knowledge base document to obtain a second group of information; performing knowledge disambiguation on the first group of information to obtain a user question record graph; performing knowledge disambiguation on the second group of information to obtain a knowledge base document graph; performing knowledge fusion based on spatial alignment on the user question record graph and the knowledge base document graph to obtain the semantic space alignment spectrum, wherein the triple information extraction includes entity extraction, attribute extraction, and relationship extraction.

[0008] In one embodiment of the present disclosure, the user question record graph and the knowledge base document graph are subjected to knowledge fusion based on space alignment to obtain the semantic space alignment spectrum, including: calculating the entity similarity between the user question record graph and the knowledge base document graph; aligning similar entities in the user question record graph and the knowledge base document graph based on the calculation results; merging and integrating the correlation and attribute information of the aligned entities to obtain the semantic space alignment spectrum.

[0009] In one embodiment of the present disclosure, the user question is rewritten based on a query operation on a semantic space alignment graph to obtain information to be retrieved, including: processing the user question based on a named entity recognition model to extract key entity information in the user question; performing a related search on the semantic space alignment graph for the key entity information based on a graph ranking calculation to obtain an identical semantic entity subgraph related to the key entity information; and rewriting the user question based on the identical semantic entity subgraph to obtain the information to be retrieved.

[0010] In one embodiment of the present disclosure, a graph-based ranking calculation is performed on the semantic space alignment graph to perform a related search for the key entity information to obtain an identical semantic entity subgraph related to the key entity information, including: taking the key entity information as the starting node for personalized page ranking PPR calculation to simulate a random walk in the semantic space alignment graph based on the PPR calculation from the starting node; calculating the importance scores of the entity nodes in the semantic space alignment graph relative to the starting node based on the walk results; determining the related entity nodes in the key entity information based on the importance scores, and generating the identical semantic entity subgraph based on the triple information of the related entity nodes.

[0011] In one embodiment of the present disclosure, the user question is rewritten based on the same semantic entity subgraph to obtain the information to be retrieved, including: inputting the same semantic entity subgraph into a query rewriting model to perform similar semantic retrieval and / or similar word expansion to obtain the information to be retrieved.

[0012] In one embodiment of the present disclosure, the intent of the user question is identified based on the mental model, including: determining the subject category in the mental model to which the user question belongs based on the intention semantic features of the user question; determining the question pattern in the mental model that matches the expression structure based on the user question; determining the entity relationship in the mental model that matches the user question; and identifying the intent of the user question based on the subject category, the question pattern and the entity relationship.

[0013] In one embodiment of the present disclosure, the business knowledge fragment includes a business background title and business background content, the large language model includes a first filtering model, a second filtering model and a main language model, the business knowledge fragment is input into the large language model for response, and the answer information of the user question is output, including: inputting the business background title and the user question into the first filtering model to detect the subject consistency between the business background title and the user question based on the first filtering model; if the subject is consistent, inputting the user question and the business background content into the main language model to generate information to be answered; inputting the information to be answered and the user question into the second filtering model to detect the correlation between the information to be answered and the user question based on the second filtering model; if it is determined to be related, determining the information to be answered as the answer information and feeding it back to the user end.

[0014] In one embodiment of the present disclosure, before determining the corresponding domain knowledge base for semantic enhancement based on the result of intent recognition, it also includes: detecting whether the result of the intent recognition hits a preset question library; if it hits the preset question library, querying the preset question library for matching fixed-word replies, wherein if it does not hit the preset question library, determining the corresponding domain knowledge base for semantic enhancement based on the result of the intent recognition.

[0015] According to another aspect of the present disclosure, there is provided a customer service interaction optimization device, including: an identification module, for performing intent identification on the user question based on a mental model in response to a received user question; a determination module, for determining a corresponding domain knowledge base for semantic enhancement based on a result of intent identification; a rewriting module, for rewriting the user question based on a query operation on a semantic space alignment spectrum to obtain information to be retrieved, wherein the semantic space alignment spectrum is generated based on a graph alignment of a user question record log and a knowledge base document in a corresponding field; a retrieval module, for searching the corresponding domain knowledge base based on the information to be retrieved to obtain a business knowledge fragment that enhances the user question; and a response module, for inputting the business knowledge fragment into a large language model for response, and outputting reply information to the user question.

[0016] According to another aspect of the present disclosure, there is provided an electronic device, including: a processor; and a memory for storing executable instructions of the processor; the processor is configured to execute the above-mentioned customer service interaction optimization method by executing the executable instructions.

[0017] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned customer service interaction optimization method is implemented.

[0018] According to another aspect of the present disclosure, a computer program product is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned customer service interaction optimization method is implemented.

[0019] The customer service interaction optimization solution provided by the embodiments of the present disclosure configures a mental model and a semantic space alignment spectrum, identifies user intent based on the mental model, determines the relevant knowledge base, and rewrites user questions based on the semantic space alignment spectrum to implement knowledge graph retrieval technology to achieve conversion between user query semantics and knowledge base document semantics. The information to be retrieved obtained is conducive to more accurate retrieval of business knowledge fragments. Inputting the business knowledge fragments into the large language model to generate replies can effectively bridge the semantic gap between the knowledge base and user questions, and improves the accuracy of the large language model's understanding of user intent, which in turn helps to improve the user experience in the interaction with the intelligent customer service system.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0022] Figure 1 A schematic diagram showing a customer service interaction optimization system in an embodiment of the present disclosure is shown;

[0023] Figure 2 A schematic flow chart showing a customer service interaction optimization method in an embodiment of the present disclosure is shown;

[0024] Figure 3 A schematic flow chart showing another customer service interaction optimization method in an embodiment of the present disclosure;

[0025] Figure 4 A schematic flow chart showing another method for optimizing customer service interaction in an embodiment of the present disclosure is shown;

[0026] Figure 5 A schematic flow chart showing another method for optimizing customer service interaction in an embodiment of the present disclosure is shown;

[0027] Figure 6 A schematic flow chart showing another method for optimizing customer service interaction in an embodiment of the present disclosure is shown;

[0028] Figure 7 A schematic flow chart showing another method for optimizing customer service interaction in an embodiment of the present disclosure is shown;

[0029] Figure 8 A schematic diagram of a customer service interaction optimization device in an embodiment of the present disclosure is shown;

[0030] Fig. 9 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0031] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0032] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0033] Applying large language models to the field of question-and-answer systems to realize artificial intelligence customer service question-and-answer (Q&A) and combining them with RAG (Retrieval-Augmented Generation) technology can integrate general open source large models into industry customer service Q&A at a lower cost. By combining information retrieval and generation capabilities, the system can extract and generate relevant answers from a rich knowledge base, significantly improving user experience and response accuracy.

[0034] However, due to the gap between the language and background information used by users when asking questions and the standardized information stored in the knowledge base, the semantics of the ordinary user semantic space and the knowledge base semantic space are not aligned, resulting in the failure of effective connection between the large language model and the customer service question-and-answer system, which not only affects the system's understanding of user intent, but may also cause users to feel confused and dissatisfied during the interaction process. Therefore, bridging the semantic gap between the two has become the key to improving the performance of intelligent customer service systems.

[0035] Query rewriting is a natural language processing technology, which mainly includes rule-based methods, template matching, synonym expansion and other rewriting methods. By adjusting and optimizing user queries to improve their matching degree with the information in the knowledge base, the accuracy and effectiveness of the question-answering system are enhanced. However, these methods have limitations such as difficult maintenance, poor adaptability, inability to effectively handle diverse user expressions, and insufficient understanding of complex semantic relationships. Therefore, it is necessary to propose a new query rewriting scheme to solve the inconsistency problem between user semantics and knowledge base semantic space, so as to effectively connect the large language model with the customer service question-answering system, enable the intelligent customer service system to understand user needs more accurately, and improve the accuracy and satisfaction of responses.

[0036] To facilitate understanding, several terms involved in this application are first explained below.

[0037] RAG (Retrieval-Augmented Generation): is a model architecture that combines information retrieval and text generation. It enhances the output of the generative model by retrieving relevant information (such as documents or data fragments), thereby improving the accuracy and relevance of answers. RAG is often used in question-answering systems and dialogue generation, leveraging information from external knowledge bases to support the generation of richer and more accurate content.

[0038] LLM (Large Language Model): refers to a language model trained using deep learning technology, which is usually very large in size, containing hundreds of millions to trillions of parameters. These models are trained with large amounts of text data and can understand and generate natural language. LLM performs well in a variety of natural language processing tasks, including text generation, translation, question answering, dialogue systems, etc.

[0039] Figure 1 1 is a schematic diagram of a customer service interaction system provided by an exemplary embodiment of the present application. The system includes: a plurality of user terminals 120 and a server terminal 140, wherein the user terminal 120 sends a user question and the server terminal 140 responds.

[0040] The user end 120 can be a mobile terminal such as a mobile phone, a game console, a tablet computer, an e-book reader, smart glasses, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a smart home device, an AR (Augmented Reality) device, a VR (Virtual Reality) device, or the user end 120 can also be a personal computer (PC), such as a laptop computer and a desktop computer.

[0041] Among them, the user terminal 120 may be installed with an application for providing customer service interaction optimization.

[0042] The user end 120 is connected to the server end 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0043] The server 140 is a server, or is composed of several servers, or is a virtualization platform, or is a cloud computing service center. The server 140 is used to provide background services for applications that provide customer service interaction optimization. Optionally, the server 140 undertakes the main computing work and the user end 120 undertakes the secondary computing work; or, the server 140 undertakes the secondary computing work and the user end 120 undertakes the main computing work; or, the user end 120 and the server 140 adopt a distributed computing architecture for collaborative computing.

[0044] In some optional embodiments, the server 140 is used to store customer service interaction optimization program information.

[0045] Optionally, the logistics user terminals of the application programs installed in different user terminals 120 are the same, or the logistics user terminals of the application programs installed on the two user terminals 120 are logistics user terminals of the same type of application programs on different control system platforms. Based on the different terminal platforms, the specific forms of the logistics user terminals of the application programs may also be different, for example, the logistics user terminals of the application programs may be mobile phone logistics user terminals, PC logistics user terminals, or World Wide Web (Web) logistics user terminals.

[0046] Those skilled in the art will appreciate that the number of the user terminals 120 may be more or less. For example, the terminal may be only one, or the terminal may be dozens or hundreds, or more. The embodiment of the present application does not limit the number and device type of the terminal.

[0047] Optionally, the system may also include a management device ( Figure 1 (not shown), the management device is connected to the server 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0048] Optionally, the above-mentioned wireless network or wired network uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0049] like Figure 2 As shown, according to an embodiment of the present disclosure, a customer service interaction optimization method includes:

[0050] Step S202: in response to the received user question, identify the intent of the user question based on the mental model.

[0051] In some embodiments, by constructing a corresponding mental model based on the business field involved in the intelligent customer service system, the mental model may include data such as common business concepts, business processes, common user needs and question patterns in the field.

[0052] In some embodiments, the user questions with extracted features are matched with the content in the mental model to judge the user's intention from multiple aspects such as themes, patterns, and relationships.

[0053] Step S204: determine the corresponding domain knowledge base for semantic enhancement based on the result of intent recognition.

[0054] In some embodiments, by configuring multiple knowledge bases divided according to different business fields or business functions in the intelligent customer service system, each knowledge base stores a specific type of standardized business information. According to the identified user question intention, a matching domain knowledge base for semantic enhancement is determined as the basis for subsequent semantic enhancement operations.

[0055] Step S206, rewrite the user question based on the query operation of the semantic space alignment spectrum to obtain the information to be retrieved. The semantic space alignment spectrum is generated based on the graph alignment of the user question record log and the knowledge base document in the corresponding field.

[0056] In some embodiments, user problem record logs and knowledge base document materials in the corresponding field can be collected, and natural language processing technology can be used to extract entities and relationships from the user problem record logs and knowledge base documents respectively, and a graph can be constructed with these entities as nodes and relationships as edges. Then, a semantic space alignment spectrum can be generated through operations such as entity alignment, relationship alignment, and knowledge fusion.

[0057] In some embodiments, the user's original question is queried in the semantic space alignment spectrum to find semantically similar and related entity and relationship information, and the user's question is reasonably rewritten using this information to obtain the information to be retrieved, so that the expression is closer to the standardized language style in the knowledge base.

[0058] Step S208: searching the corresponding domain knowledge base based on the information to be retrieved to obtain business knowledge fragments that enhance the user's question.

[0059] In some embodiments, the information to be retrieved is used as a search term to search in the corresponding domain knowledge base to find business knowledge content that is highly relevant to the information to be retrieved.

[0060] Step S210: Input the business knowledge fragment into the large language model for answering, and output the answer information of the user's question.

[0061] In this embodiment, by configuring the mental model and the semantic space alignment spectrum, the user intention is identified based on the mental model, the relevant knowledge base is determined, and the user question is rewritten based on the semantic space alignment spectrum to realize the conversion between the user query semantics and the knowledge base document semantics by the knowledge graph retrieval technology. The information to be retrieved is conducive to more accurate retrieval of business knowledge fragments. The business knowledge fragments are input into the large language model to generate answers, which can effectively bridge the semantic gap between the knowledge base and the user's questions, improve the accuracy of the large language model's understanding of the user's intention, and thus help improve the user's experience in interacting with the intelligent customer service system.

[0062] In one embodiment of the present disclosure, before performing intent recognition on the user question based on the mental model in response to the received user question, the method further includes:

[0063] The triplet information is extracted from the user question record log to obtain the first set of information.

[0064] The triple information is extracted from the knowledge base documents to obtain the second set of information.

[0065] Among them, triple information extraction includes entity extraction, attribute extraction and relationship extraction.

[0066] In some embodiments, the business-related knowledge base documents may be product specifications, operation manuals, knowledge materials in professional fields, etc. in text form.

[0067] Collect query record log data left by users during their interactions with customer service question-and-answer systems, etc. This data reflects the users' actual questioning situations, focus, language expression habits, and other information.

[0068] In some embodiments, entity extraction methods in natural language processing are used to process text content in knowledge base documents and user query record log data, including but not limited to rule-based entity extraction (identifying specific types of entities based on pre-set vocabulary and grammar rules, such as defining all words that mention product names as product entities), machine learning-based entity extraction (using training data of labeled entities to train classification models to identify entities in new texts, such as using support vector machines, named entity recognition models in deep learning, etc.) and deep learning-based pre-trained models (such as BERT and improved models, which perform entity extraction tasks through fine-tuning).

[0069] For the entities in the extracted knowledge base documents and user query record log data, further analyze and extract their relevant attributes, including but not limited to descriptive statements based on parsed text, specific format tags (if the document has a certain structured format), etc.

[0070] Analyze the relationship between different entities in the knowledge base documents and the relationship between different entities in the user query record log, using rule-based pattern matching or machine learning-based relationship classification models.

[0071] Perform knowledge disambiguation on the first set of information to obtain a user question record graph.

[0072] Perform knowledge disambiguation on the second set of information to obtain a knowledge base document graph.

[0073] In some embodiments, since there may be entities with the same name but different meanings or synonyms but different names in the knowledge base documents and user query records, entity disambiguation operations are required. In addition, if there is ambiguity or ambiguity in the understanding of the extracted relationships and attributes, they also need to be processed.

[0074] The user question record graph and the knowledge base document graph are fused based on spatial alignment to obtain a semantic space alignment spectrum.

[0075] In some embodiments, the process of performing knowledge fusion based on spatial alignment includes but is not limited to: fusing and unifying the same or similar entities extracted from knowledge base documents and user query records, and integrating the relationships and attributes involving these entities based on similar relationship expressions and attribute descriptions.

[0076] In this embodiment, based on operations such as entity extraction, attribute extraction, relationship extraction, knowledge disambiguation, and knowledge fusion, the knowledge in the knowledge base documents and user query records is integrated and optimized, and a graph structure is constructed with entities as nodes, relationships as edges, and attributes as supplementary descriptive information of the nodes. This allows the knowledge involved in the user question and the knowledge in the knowledge base documents to be aligned in the semantic space in this graph, even if the two can correspond and connect with each other in terms of semantic understanding. When the user question is subsequently received, the semantic conversion of the user query semantics and the knowledge base document is realized through the knowledge graph retrieval technology, so as to achieve accurate matching of the user query, thereby improving the response efficiency and accuracy of the intelligent customer service system.

[0077] In one embodiment of the present disclosure, a user question record graph and a knowledge base document graph are fused based on space alignment to obtain a semantic space alignment graph, including:

[0078] Calculate the entity similarity between the user question record graph and the knowledge base document graph; align similar entities in the user question record graph and the knowledge base document graph based on the calculation results; merge and integrate the relevant relationships and attribute information of the aligned entities to obtain a semantic space alignment spectrum.

[0079] In some embodiments, the similarity of entities at the name level can be determined based on the edit distance of the entity names (such as the Levenshtein distance, which measures the minimum number of times an entity name is transformed into another entity name by inserting, deleting, or replacing characters, where fewer times indicate higher similarity), or by using cosine similarity to calculate the word vectors of the entity names. Entity similarity can also be calculated using methods based on semantic similarity, such as using a pre-trained language model to put the entity into a sentence context to analyze its semantic similarity.

[0080] Furthermore, a similarity threshold may be set based on the characteristics of the business domain and actual application requirements, and based on the similarity threshold, it may be determined which entities can be considered similar and aligned.

[0081] For the aligned entities, we sort out the relationships and attribute information involved in their respective graphs, and formulate merging and integration rules based on business logic, knowledge consistency requirements and other factors. For relationships, if there are relationships with similar semantics and the same functions in the two graphs, they are merged into a unified relationship expression; for attributes, if it is determined that the attribute names are different but the actual meanings are the same, the attribute names are unified, and the attribute values ​​are processed reasonably. The aligned entities are used as nodes, the integrated relationships are used as edges, and the unified attributes are used as supplementary descriptive information of the nodes to construct a semantic space alignment spectrum.

[0082] In this embodiment, by calculating entity similarity, aligning similar entities, and merging and integrating related relationships and attribute information to construct a semantic space alignment spectrum, it is helpful to eliminate the semantic gap between user questions and the knowledge base due to differences in expression, realize the unified integration of knowledge at the semantic level, and facilitate more accurate use of knowledge to serve the question-answering system.

[0083] like Figure 3 As shown, in one embodiment of the present disclosure, the user question is rewritten based on the query operation of the semantic space alignment spectrum to obtain the information to be retrieved, including:

[0084] Step S302: Process the user question based on the named entity recognition model to extract key entity information in the user question.

[0085] In some embodiments, the named entity recognition model can be generated based on pre-set vocabulary and grammatical rules, or it can be generated using a machine learning-based model. For example, using training data of labeled entities to train traditional machine learning models such as support vector machines and decision trees. It can also be generated after fine-tuning a pre-trained language model for named entity recognition tasks.

[0086] Step S304, based on the sorting calculation of the graph, the semantic space alignment spectrum graph is searched for the key entity information, and the same semantic entity subgraph related to the key entity information is obtained.

[0087] In some embodiments, graph-based ranking calculations include but are not limited to PageRank algorithm, PersonalizedPageRank (PPR) algorithm, HITS algorithm (Hyperlink-Induced Topic Search), SimRank algorithm, LeaderRank algorithm, and the like.

[0088] In some embodiments, key entity information is used as the starting node or key basis for retrieval, and a selected graph-based sorting calculation method is used to perform retrieval in the semantic space alignment spectrum. Based on the structure of the spectrum and the connection relationship between the nodes, the importance score or similarity of each node relative to the key entity information is calculated by simulating random walks and other mechanisms. Then, based on the set thresholds, sorting rules, etc., nodes with higher importance or similarities that meet the requirements and their connection relationships are screened out. The subgraph composed of these nodes and relationships is the same semantic entity subgraph related to the key entity information.

[0089] For example, by using the PPR algorithm to search the corresponding semantic space alignment spectrum with "a certain brand of mobile phone" as the key entity information, we may obtain the same semantic entity subgraph containing entities such as its camera components, image processing algorithms, related shooting modes, and the relationships between them. These are all knowledge content related to the camera function of "a certain brand of mobile phone".

[0090] Step S306: rewrite the user question based on the same semantic entity subgraph to obtain information to be retrieved.

[0091] In this embodiment, the key entity information of the user's question is extracted through the named entity recognition model, and the relevant identical semantic entity subgraphs are retrieved in the semantic space alignment spectrum using a graph-based sorting calculation. The user's question is then rewritten based on the subgraph to obtain the information to be retrieved, and the knowledge base is searched based on the information to be retrieved. This can effectively connect the user's question with the knowledge base knowledge system, and can more accurately locate the user's needs, so that the subsequent knowledge content that is helpful in answering the user's question can be more accurately obtained.

[0092] like Figure 4 As shown, in one embodiment of the present disclosure, the semantic space alignment spectrogram is subjected to relevant retrieval of key entity information based on the sorting calculation of the graph, and the same semantic entity subgraph related to the key entity information is obtained, including:

[0093] Step S402: Use the key entity information as the starting node for calculating the personalized page ranking (PPR), and simulate random walk in the semantic space alignment spectrum based on the PPR calculation from the starting node.

[0094] Step S404, calculating the importance score of the entity node in the semantic space alignment spectrum relative to the starting node based on the walking result.

[0095] In some embodiments, in the graph structure of the knowledge graph, the PPR algorithm can calculate the importance score of each node in the graph (i.e., multiple entities in the knowledge graph) relative to the starting node based on a given starting node, and will consider factors such as the connection relationship between nodes and the probability of wandering.

[0096] Step S406 , determining entity nodes related to the key entity information based on the importance scores, and generating a subgraph of entities with the same semantics based on triple information of the related entity nodes.

[0097] In some embodiments, for the screened out related entity nodes, the triple relationships involved in the semantic space alignment spectrum are searched, these related entity nodes and their corresponding triple information are extracted, and a subgraph is constructed in the form of entities as nodes and relationships as edges. This subgraph is an entity subgraph with the same semantics as the key entity information.

[0098] In this embodiment, through the process of searching in the knowledge graph based on the PPR algorithm, entities and triples related to user questions can be effectively mined, forming the same semantic entity subgraph that is semantically aligned with the knowledge base, thereby achieving the purpose of mining semantically related knowledge based on the key content that users are concerned about, thereby effectively improving the efficiency and accuracy of retrieving knowledge related to user questions in the knowledge graph.

[0099] In one embodiment of the present disclosure, user questions are rewritten based on the same semantic entity subgraph to obtain information to be retrieved, including: inputting the same semantic entity subgraph into a query rewriting model to perform similar semantic retrieval and / or similar word expansion to obtain information to be retrieved.

[0100] In some embodiments, taking the rewriting based on the question-answer library as an example, preset questions can be configured in the question-answer library. When it is detected that the user question is consistent with the preset question, it can flexibly respond to generate semantically similar user questions, significantly improve the matching accuracy of the question-answer library, and improve the user satisfaction of the customer service robot. In the telecommunications customer service interaction system, more than 100 external service skill calls can be supported, including traffic, call charges, package inquiries, etc.

[0101] In this embodiment, by optimizing semantic mapping and query rewriting, the question-answering system can more accurately capture user needs, improve the accuracy of responses and overall satisfaction, and thus help to more effectively meet the user's true intentions and enhance the user experience.

[0102] like Figure 5 As shown, in one embodiment of the present disclosure, the intention of user questions is identified based on the mental model, including:

[0103] Step S502: determining the subject category in the mental model based on the intended semantic features of the user's question.

[0104] In some embodiments, the mental model can be understood as an internal cognitive structure or knowledge framework based on knowledge of a specific business domain (such as telecommunications business), common user cognition and behavior patterns, and used to understand and analyze the user's question intentions, including subject categories, question patterns, and entity relationships. The subject category is used to distinguish different knowledge sections in the business domain, the question pattern summarizes the common questioning methods and expression structures of users in this domain, and the entity relationship reflects the way different entities in the business domain (which can be specific objects, people, concepts, etc.) are interconnected and interact with each other.

[0105] In some embodiments, the semantic features of intent can be mined through operations such as keyword extraction (using word frequency statistics, TF-IDF, etc. to determine the key words in the question), part-of-speech tagging (analyzing the parts of speech of words, such as nouns, verbs, adjectives, etc., to clarify the grammatical structure and semantic emphasis of the sentence), and named entity recognition (identifying specific entities in the question, such as people, places, product names, etc.).

[0106] Furthermore, the extracted semantic features of intent are compared and matched with various subject categories in the constructed mental model to preliminarily narrow the scope of user intent judgment.

[0107] Step S504: determine a question pattern in the mental model that matches the expression structure based on the user question.

[0108] In some embodiments, by determining the textual expression structure of the user's question, the mental model is searched for corresponding common problem patterns, and through comparative analysis, the user's question and similar problem patterns in the mental model are determined to further clarify the general type and intended direction of the user's question.

[0109] Step S506, determining the entity relationship in the mental model that matches the user question.

[0110] In some embodiments, the entity relationships in the user's question are compared with the entity relationships constructed in the mental model to help more accurately determine the user's intent.

[0111] Step S508, identifying the intention of the user's question based on the subject category, question pattern and entity relationship.

[0112] In this embodiment, by analyzing and matching user questions from multiple dimensions such as subject category, question pattern, and entity relationship, the mental model is used to comprehensively and deeply identify the intention of the user question, which helps to overcome the deviation that may occur in a single judgment method, so as to achieve accurate positioning of user intentions and ensure the reliability of the process of understanding and predicting user intentions.

[0113] In some embodiments, the business knowledge segment includes a business background title and business background content, and the large language model includes a first filtering model, a second filtering model and a main language model, that is, a large language model of a 1+2LLM mode.

[0114] In one embodiment of the present disclosure, it also includes: performing intent scenario recognition on user questions based on an intent scenario recognition model, and outputting probability distribution of different intent scenario categories; determining corresponding intent scenarios based on the probability distribution, and the scenario intentions include consultation, inquiry, processing, fault reporting, and complaint categories, etc.

[0115] In some embodiments, taking the telecommunications customer service interaction system as an example, on the basis of retaining the robot function model, query call interface module, and front-end function module of the telecommunications customer service interaction system, a scenario intention recognition model is added to divide the entire telecommunications customer service scenario into 5 major categories, including inquiry, processing, consultation, fault reporting, and complaint. Among them, the four categories of inquiry, processing, fault reporting, and complaint can reply to user questions through the fixed scripts configured in the scenario. For example, for package inquiry and consultation scenarios, the retrieval enhanced large language model of the 1+2LLM mode can be used to reply to user questions.

[0116] In this embodiment, the intent scenario recognition model is used to perform intent scenario recognition on user questions, and the probability distribution of different intent scenario categories is output. The corresponding intent scenario is determined based on the probability distribution. This approach helps to accurately distinguish the different business scenarios to which the user questions belong, and then take targeted follow-up processing measures based on the characteristics of the scenarios, thereby improving the accuracy and effectiveness of handling various types of user problems and optimizing the overall service quality and user experience.

[0117] like Figure 6 As shown, in one embodiment of the present disclosure, the business knowledge fragment is input into the large language model for answering, and the answer information of the user question is output, including:

[0118] Step S602: input the business background title and the user question into the first filtering model to detect the subject consistency between the business background title and the user question based on the first filtering model.

[0119] In some embodiments, the first filtering model is used to receive business background titles and user questions, and based on semantic analysis, topic modeling, keyword extraction, etc., to measure whether the business background titles and user questions are consistent at the subject level. By comparing the semantic topics of the two, it is determined whether the topic involved in the business background title can match the topic of the user question, thereby playing a role in preliminary screening and preventing the subsequent use of irrelevant business background content to generate replies.

[0120] Step S604: if the topics are consistent, the user question and business background content are input into the main language model to generate information to be answered.

[0121] In some embodiments, after the first filtering model detects that the business background title is consistent with the user question topic, the main language model receives the user question and the corresponding business background content as input information. The main language model conducts in-depth analysis of the input content and integrates relevant knowledge to generate reply information that meets the user's question requirements and fits the business background. It is the key execution link in the entire reply generation process and determines the quality of the final reply information.

[0122] Step S606: If the subject is inconsistent, it will be transferred to manual processing.

[0123] Step S608: input the information to be answered and the user's question into the second filtering model to detect the correlation between the information to be answered and the user's question based on the second filtering model.

[0124] In some embodiments, the second filtering model takes effect after the main language model generates the information to be replied, and is used to take the information to be replied and the user's question as input, and use semantic understanding, text similarity calculation, etc. to accurately determine the degree of correlation between the information to be replied and the user's question. For example, by calculating the semantic similarity between the two, if the similarity reaches a certain threshold, it is determined that they are related, thereby ensuring that the reply information finally fed back to the user is closely centered on the question raised by the user, preventing the reply generated by the main language model from deviating from the core of the user's question, and improving the quality of the reply.

[0125] Step S610: If it is determined to be relevant, the information to be replied is determined as reply information and fed back to the user end.

[0126] Step S612: If it is determined to be irrelevant, it is transferred to manual processing.

[0127] In this embodiment, the large language model of the 1+2LLM mode uses the complementary advantages of each model through pre-filtering, main language model generation and post-filtering processes to accurately screen out business background knowledge that matches the user's question for reply generation, and strictly control the relevance of the final reply to the question, so as to prevent the introduction of irrelevant business background information or the generation deviation of the main language model itself from causing content inconsistent with the user's question, thereby making the reply information provided by the entire question-and-answer system to the user more accurate and reliable, greatly improving the reply accuracy of the large model and greatly reducing the hallucination rate.

[0128] like Figure 7 As shown, according to another embodiment of the present disclosure, a customer service interaction optimization method includes:

[0129] Step S702: In response to the received user question, identify the intent of the user question based on the mental model.

[0130] Step S702, detect whether the result of intent recognition hits the preset question library, if "yes", go to step S706, if "no", go to step S708.

[0131] Step S706, hit the preset question library, and search for matching fixed-word responses in the preset question library.

[0132] In some embodiments, in response to received user questions, intent recognition is first performed. Intent recognition can usually be implemented based on a classification method, that is, user questions are divided into several categories. For example: in a telecommunications customer service question and answer system, user questions can be divided into package inquiries, call fee inquiries, bill inquiries, traffic inquiries, etc. The results of intent recognition are matched with a common question library. If there is a match, the question library is queried and a reply is given directly. If there is no match, step S708 is entered.

[0133] Step S708, determining a corresponding domain knowledge base for semantic enhancement based on the result of intent recognition.

[0134] Step S710, rewrite the user question based on the query operation of the semantic space alignment spectrum to obtain the information to be retrieved. The semantic space alignment spectrum is generated based on the graph alignment of the user question record log and the knowledge base document in the corresponding field.

[0135] In some embodiments, in order to improve the intelligence of responses based on large language models, in the present disclosure, corresponding graphs are constructed for knowledge base documents and user query record log data through knowledge graph construction technology, including entity extraction, attribute extraction, relationship extraction, knowledge disambiguation, knowledge fusion and other operations. Based on the above operations, a semantic space alignment graph of user questions and knowledge base documents is constructed to achieve alignment of the user semantic space and the knowledge base document semantic space based on the knowledge graph.

[0136] For the received user questions, the user question entities are extracted and searched in the semantic space alignment spectrum using the PPR algorithm to obtain the entities and triples related to the user questions, that is, the subgraphs of the same semantic entities.

[0137] Step S712: searching the corresponding domain knowledge base based on the information to be retrieved to obtain business knowledge fragments that enhance the user's question.

[0138] In some embodiments, the query is rewritten based on the subgraph to obtain a modified question, that is, information to be retrieved.

[0139] Step S714: input the business knowledge fragment into the large language model for answering, and output the answer information of the user's question.

[0140] In some embodiments, based on the rewritten question, a search is performed in the knowledge base to obtain knowledge fragments related to the user's question, and the knowledge fragments are given to the large model as online text integration instructions for answering.

[0141] S716, perform relevance check on the reply information to determine whether to transfer to manual, if "yes", go to step S718, if "no", go to step S720.

[0142] Step S718: transfer the user issue to manual processing.

[0143] Step S720, directly output the reply information.

[0144] In this embodiment, in order to fully understand the user's intentions and locate the user's needs through the intention recognition method, and to solve the alignment problem between the user query semantics and the knowledge base document semantics, a semantic space alignment method based on the knowledge graph is introduced. This method rewrites the user query and effectively converts its semantics into the semantics of the knowledge base document. Retrieval based on the rewritten query can accurately locate the relevant knowledge information, thereby achieving accurate responses to user queries and improving the overall service quality of the intelligent customer service system.

[0145] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0146] Refer to the following Figure 8 To describe the customer service interaction optimization device 800 according to an embodiment of the present disclosure. Figure 8 The customer service interaction optimization device 800 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0147] The customer service interaction optimization device 800 is presented in the form of a hardware module or a software module. The components of the customer service interaction optimization device 800 may include but are not limited to: an identification module 802, which is used to identify the intent of the user question based on the mental model in response to the received user question; a determination module 804, which is used to determine the corresponding domain knowledge base for semantic enhancement based on the result of the intent recognition; a rewriting module 806, which is used to rewrite the user question based on the query operation of the semantic space alignment spectrum to obtain the information to be retrieved, and the semantic space alignment spectrum is generated based on the alignment of the user question record log and the knowledge base document in the corresponding field; a retrieval module 808, which is used to retrieve the corresponding domain knowledge base based on the information to be retrieved, and obtain the business knowledge fragment enhanced for the user question; a response module 810, which is used to input the business knowledge fragment into the large language model for response, and output the answer information of the user question.

[0148] like Fig. 9 As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include but are not limited to: at least one processing unit 910, at least one storage unit 920, and a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910).

[0149] The storage unit stores a program code, which can be executed by the processing unit 910, so that the processing unit 910 performs the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 910 can perform the following steps: Figure 2 The scheme described in .

[0150] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202 , and may further include a read-only storage unit (ROM) 9203 .

[0151] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0152] Bus 930 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0153] The electronic device 900 may also communicate with one or more external devices 970 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device 900, and / or any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 950. Furthermore, the electronic device 900 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 960. Fig. 9 As shown, the network adapter 960 communicates with other modules of the electronic device 900 via the bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0154] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or an electronic device, etc.) to execute the method according to the implementation of the present disclosure.

[0155] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on an electronic device, the program code is used to enable the electronic device to perform the steps according to various exemplary implementations of the present disclosure described in the above "Exemplary Method" section of the present specification.

[0156] According to the program product for implementing the above method according to the embodiment of the present disclosure, it can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on an electronic device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0157] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0158] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0159] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0160] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the IoT terminal, as an independent software package, partially on the user computing device and partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0161] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of a module or unit described above can be further divided into multiple modules or units to be concretized. In addition, although the various steps of the method in the present disclosure are described in a specific order in the accompanying drawings, this does not require or imply that these steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired result. In addition or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution, etc.

[0162] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions so that a computing device (which can be a personal computer, a server, a mobile terminal, or an electronic device, etc.) executes the method according to the embodiment of the present disclosure. After considering the disclosure of the specification and practicing the disclosure here, it will be easy for those skilled in the art to think of other embodiments of the present disclosure. This application is intended to cover any variation, use or adaptive change of the present disclosure, which follows the general principles of the present disclosure and includes the common knowledge or customary technical means in the technical field that are not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are indicated by the attached claims.

Claims

1. A customer service interaction optimization method, characterized in that: include: In response to a received user question, identifying the intent of the user question based on the mental model; Determine the corresponding domain knowledge base for semantic enhancement based on the result of intent recognition; Rewrite the user question based on a query operation of a semantic space alignment spectrum to obtain information to be retrieved, wherein the semantic space alignment spectrum is generated based on a graph alignment of a user question record log and a knowledge base document in a corresponding field; Retrieving the corresponding domain knowledge base based on the information to be retrieved to obtain a business knowledge fragment that enhances the user's question; The business knowledge fragment is input into a large language model for answering, and answer information of the user's question is output.

2. The customer service interaction optimization method according to claim 1, characterized in that: Before performing intent recognition on the user question based on the mental model in response to the received user question, the method further includes: Extracting triplet information from the user question record log to obtain a first set of information; Extracting triple information from the knowledge base document to obtain a second set of information; Performing knowledge disambiguation on the first set of information to obtain a user question record graph; Performing knowledge disambiguation on the second group of information to obtain a knowledge base document graph; Performing knowledge fusion based on space alignment on the user question record graph and the knowledge base document graph to obtain the semantic space alignment graph, The triple information extraction includes entity extraction, attribute extraction and relationship extraction.

3. The customer service interaction optimization method according to claim 2, characterized in that: Performing knowledge fusion based on space alignment on the user question record graph and the knowledge base document graph to obtain the semantic space alignment graph includes: Calculating the entity similarity between the user question record graph and the knowledge base document graph; Aligning similar entities in the user question record graph and the knowledge base document graph based on the calculation results; The relevant relationships and attribute information of the aligned entities are merged and integrated to obtain the semantic space alignment spectrum.

4. The customer service interaction optimization method according to claim 1, characterized in that: The user question is rewritten based on the query operation of the semantic space alignment spectrum to obtain the information to be retrieved, including: Processing the user question based on a named entity recognition model to extract key entity information in the user question; Based on the sorting calculation of the graph, the semantic space alignment spectrum graph is subjected to relevant retrieval of the key entity information to obtain the same semantic entity subgraph related to the key entity information; The user question is rewritten based on the same semantic entity subgraph to obtain the information to be retrieved.

5. The customer service interaction optimization method according to claim 4, characterized in that: The semantic space alignment spectrum is subjected to a related search of the key entity information based on the sorting calculation of the graph, and the same semantic entity subgraph related to the key entity information is obtained, including: The key entity information is used as a starting node for calculating a personalized page ranking (PPR), so as to simulate a random walk in the semantic space alignment spectrum based on the PPR calculation from the starting node; Calculate the importance score of the entity node in the semantic space alignment spectrum relative to the starting node based on the walking result; Entity nodes related to the key entity information are determined based on the importance scores, and the same semantic entity subgraph is generated based on triple information of the related entity nodes.

6. The customer service interaction optimization method according to claim 4, characterized in that: Rewriting the user question based on the same semantic entity subgraph to obtain the information to be retrieved includes: The same semantic entity subgraph is input into a query rewriting model to perform similar semantic retrieval and / or similar word expansion to obtain the information to be retrieved.

7. The customer service interaction optimization method according to claim 1, characterized in that: The user's question is identified based on the mental model, including: Determine the subject category in the mental model to which the question belongs based on the semantic features of the user's question; Determining a question pattern in the mental model that matches a representation structure based on the user question; Determining entity relationships in the mental model that match the user question; The intention of the user question is identified based on the subject category, the question pattern and the entity relationship.

8. The customer service interaction optimization method according to claim 1, characterized in that: The business knowledge fragment includes a business background title and business background content, and the large language model includes a first filtering model, a second filtering model and a main language model. The business knowledge fragment is input into the large language model for answering, and the answer information of the user question is output, including: Inputting the business background title and the user question into the first filtering model to detect the subject consistency between the business background title and the user question based on the first filtering model; If the topics are consistent, the user question and the business background content are input into the main language model to generate information to be answered; Inputting the information to be answered and the user question into the second filtering model to detect the correlation between the information to be answered and the user question based on the second filtering model; If it is determined to be relevant, the information to be replied is determined as the reply information and fed back to the user end.

9. The customer service interaction optimization method according to claim 1, characterized in that: Before determining the corresponding domain knowledge base for semantic enhancement based on the result of intent recognition, it also includes: Detecting whether the result of the intention recognition matches a preset question library; If the preset question library is hit, the preset question library is searched for a matching fixed-word response. If the preset question library is not matched, the corresponding domain knowledge base for semantic enhancement is determined based on the result of the intent recognition.

10. A customer service interaction optimization device, characterized in that: include: an identification module, configured to, in response to a received user question, identify the intent of the user question based on a mental model; A determination module, used for determining a corresponding domain knowledge base for semantic enhancement based on the result of intent recognition; A rewriting module, used to rewrite the user question based on a query operation on a semantic space alignment spectrum to obtain information to be retrieved, wherein the semantic space alignment spectrum is generated based on a graph alignment of a user question record log and a knowledge base document in a corresponding field; A retrieval module, used to search the corresponding domain knowledge base based on the information to be retrieved, and obtain a business knowledge fragment that enhances the user's question; The answering module is used to input the business knowledge fragment into the large language model for answering and output the answer information of the user question.

11. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the customer service interaction optimization method described in any one of claims 1 to 9 by executing the executable instructions.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the customer service interaction optimization method described in any one of claims 1 to 9 is implemented.

13. A computer program product having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the customer service interaction optimization method described in any one of claims 1 to 9 is implemented.

Images