Intelligent customer service processing method, device and equipment based on large language model

Abstract

The invention provides an intelligent customer service processing method, device and equipment based on a large language model, and the method comprises the steps: calling an intention recognition large model to classify user questions, and obtaining a classification result; if the classification result is that the user question is related to the target business, rewriting the user question in combination with context information, and generating a corresponding inquiry prompt word; based on a preset hybrid retrieval strategy, calling a knowledge base node to obtain recall content related to the inquiry cue word from the target knowledge base; the target knowledge base comprises a business-related knowledge base and an FAQ database; integrating the recall content in combination with the context information to generate corresponding reply content; and constructing structured data according to the user question and the corresponding reply content, and writing the structured data into an FAQ database to complete updating of the target knowledge base. The intention recognition accuracy, the reply quality and the service efficiency of the intelligent customer service system are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an intelligent customer service processing method, device and equipment based on a large language model. BACKGROUND

[0002] In the current field of customer service, the customer service system as an important bridge for communication between enterprises and users directly affects user experience and enterprise image in terms of service quality and efficiency. However, the existing customer service system still has many problems to be solved: On the one hand, the existing customer service system relies on a pre-defined fixed reply mode and can only deal with standardized problems within a limited range. For diversified and personalized needs of users, it is often difficult to effectively meet the needs due to the lack of flexible response mechanism, resulting in a low problem solving rate of users; On the other hand, the customer service robot in the existing customer service system often has mechanical dialogue interaction process due to the design of the dialogue and the lack of context understanding ability, so that the dialogue is stiff and unnatural, and it is difficult to form a smooth experience close to human communication, which easily causes user resistance; In addition, the existing customer service system fails to establish an effective knowledge accumulation mechanism. For high-quality question and answer content that has been successfully solved, it cannot be integrated and re-integrated into the knowledge base in time, so that when similar problems occur repeatedly, the system still needs to be processed repeatedly, which not only increases the service cost, but also limits the dynamic optimization of the knowledge base and the continuous improvement of the service ability.

[0003] Therefore, the existing customer service system has the problem of low reply accuracy. SUMMARY

[0004] The embodiments of the present application provide an intelligent customer service processing method, device and equipment based on a large language model, aiming at solving the problem of low reply accuracy of the existing customer service system.

[0005] In a first aspect, the embodiments of the present application provide an intelligent customer service processing method based on a large language model, which comprises: calling an intent recognition large model to classify user problems and obtain a classification result; If the classification result is related to the target business of the user problem, the user problem is rewritten in combination with the context information to generate corresponding inquiry prompt words; Based on a preset hybrid retrieval strategy, a knowledge base node is called to obtain recall content related to the inquiry prompt words from a target knowledge base; the target knowledge base includes a business-related knowledge base and a FAQ database; The recall content is integrated in combination with the context information to generate corresponding reply content; According to the user question and the corresponding reply content, structured data is constructed, and the structured data is written into the FAQ database to complete the update of the target knowledge base.

[0006] In a second aspect, the embodiments of the present application further provide an intelligent customer service processing device based on a large language model, which comprises: A classification unit is configured to classify a user question by calling a large intent recognition model to obtain a classification result. A rewriting unit is configured to rewrite the user question in combination with context information to generate a corresponding inquiry prompt word if the classification result is related to the target business. A recall unit is configured to obtain recall content related to the inquiry prompt word from a target knowledge base by calling a knowledge base node based on a preset hybrid retrieval strategy; the target knowledge base comprises a business-related knowledge base and a FAQ database. An integration unit is configured to integrate the recall content in combination with the context information to generate corresponding reply content. An update unit is configured to construct structured data according to the user question and the corresponding reply content, and write the structured data into the FAQ database to complete the update of the target knowledge base.

[0007] In a third aspect, the embodiments of the present application further provide an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of the first aspect.

[0008] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, wherein the storage medium stores a computer program, and the computer program comprises program instructions, and the program instructions are executed by a processor to implement the method of the first aspect.

[0009] The application provides an intelligent customer service processing method, device and equipment based on a large language model, the method comprising: calling an intent recognition large model to classify a user question to obtain a classification result; if the classification result is that the user question is related to a target business, rewriting the user question in combination with context information to generate a corresponding inquiry prompt word; based on a preset hybrid retrieval strategy, calling a knowledge base node to obtain recall content related to the inquiry prompt word from a target knowledge base; the target knowledge base comprises a business-related knowledge base and an FAQ database; integrating the recall content in combination with the context information to generate corresponding reply content; and constructing structured data according to the user question and the corresponding reply content, and writing the structured data into the FAQ database to complete updating of the target knowledge base. The application classifies user questions accurately through an intent recognition large model, ensures that business-related questions are processed pertinently, generates inquiry prompt words in combination with context information and adopts a hybrid retrieval strategy to improve the accuracy and relevance of recall content of the target knowledge base, integrates recall content in combination with context information to generate reply content to enhance the logicality and naturalness of the reply content, and at the same time, updates the target knowledge base through structured data to realize self-iterative optimization of the target knowledge base. In this way, the accuracy of intent recognition, the quality of replies and the service efficiency of the intelligent customer service system are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 A flowchart of an intelligent customer service processing method based on a large language model provided by the embodiments of the present application is shown. Figure 2 A schematic block diagram of an intelligent customer service processing device based on a large language model provided by the embodiments of the present application is shown. Figure 3 A schematic block diagram of an electronic device provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0013] It should be understood that the terms "comprises" and "comprising," when used in this specification and the following claims, indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0014] It should also be understood that the terms used in the present specification and the attached claims are only for the purpose of describing particular embodiments and do not intend to limit the present application. As used in the present specification and the attached claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should be further understood that the term "and / or" used in the present specification and the attached claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. The embodiments of the present application provide a large language model-based intelligent customer service processing method, device and equipment. The large language model-based intelligent customer service processing method is described in detail below. Figure 1 , Figure 1 The flowchart of the large language model-based intelligent customer service processing method provided by the embodiments of the present application is shown in the figure. The large language model-based intelligent customer service processing method is applied to the controller of an intelligent customer service system. The present application will be described in detail below through specific embodiments.

[0016] Figure 1 The flowchart of the large language model-based intelligent customer service processing method provided by the embodiments of the present application is shown in the figure. As shown in Figure 1 The method includes the following steps S110-S150.

[0017] S110, calling an intent recognition large model to classify user problems to obtain a classification result.

[0018] In the embodiment, the user question is classified based on a standardized rule built in an intent recognition large model to obtain a classification result; the standardized rule includes an intent category, a judgment standard, an output format and a processing boundary condition, the intent category includes two categories of the user question being related to a target business and the user question being irrelevant to the target business; the judgment standard is based on the target business; the output format is specifically: when the intent category is category 1 (the user question is related to the target business), output sequence number 1; when the intent category is category 2 (the user question is irrelevant to the target business), output sequence number 2; the processing boundary condition is specifically: the user question that is difficult to understand or grasp is classified as category 2 and output sequence number 2; specifically, the intent type of the user question is determined based on a pre-defined intent recognition prompt word (such as an intent recognition role prompt word, a skill prompt word and the like) in the intent recognition large model and the above standardized rule, and then the classification result is obtained. Based on the above standardized rule, it is ensured that the selector node can directly trigger the subsequent different branch workflow (target business answer or non-target business answer) according to the output of the intent recognition large model.

[0019] In an embodiment, before step S110, it further includes: after pre-defining opening information in the initial question and answer agent, previewing and debugging the initial question and answer agent, and completing the creation of the question and answer agent after confirming that the function is correct; the opening information includes but is not limited to opening speech scripts, pre-set questions, background pictures; wherein the opening speech scripts are used to clearly indicate the service range of the intelligent customer service system to the user (such as focusing on the operation of the charging station and the related question and answer of the intelligent parking business), and the pre-set questions are set based on the common consultation scenarios of the charging station; the question and answer agent adopts a dialogue flow mode, strictly follows steps S110 to S150 in its execution process, and has a long-term memory function.

[0020] In an embodiment, after step S110, it further includes: if the classification result is that the user question is irrelevant to the target business, calling a non-product answer large model to generate answer content, and replying to the user question based on the answer content.

[0021] In the embodiment, if the classification result is that the user question is irrelevant to the target business, the answer content is generated based on the non-product answer prompt word built in the non-product answer large model, and the user question is replied based on the answer content; the non-product answer prompt word includes but is not limited to a non-product answer role prompt word and a non-product answer restriction (such as prohibiting answering personal privacy information and company sensitive data); specifically, a non-product answer role is defined based on an intelligent large model, and a non-product answer role prompt word and non-product answer restriction and the like are provided; the answer content replying to the user question is generated based on the non-product answer prompt word, and when the target business is related to the operation of the charging station, the answer content can be "I'm sorry, I can only answer the related questions of the intelligent parking product. If you have other questions related to the intelligent parking, I will try to help you answer" and the like.

[0022] In S120, if the classification result is that the user question is related to the target business, the user question is rewritten in combination with the context information to generate a corresponding inquiry prompt word.

[0023] In the embodiment, when the user question is related to the target business and the user question is more colloquial and fragmented, the user question is rewritten into a detailed inquiry prompt word in combination with the context information (such as the context historical dialogue).

[0024] For example, the user question 1 is "the charging pile cannot be charged", and the user question 2 is "I can't find the charging pile number". Based on the user question 1 (i.e. the context information), the user question 2 is analyzed to obtain the user core intent, i.e. "how to find the charging pile number in the case that the charging pile cannot be charged", and the inquiry prompt word is generated based on the obtained user core intent.

[0025] In an embodiment, S120 includes: judging whether the user question needs to be rewritten based on the rewriting prompt word built in the inquiry rewriting large model in combination with the context information; if the user question needs to be rewritten, the user question is rewritten based on the context information to generate an inquiry prompt word that can reflect the user core intent; and if the user question does not need to be rewritten, the user question is taken as the inquiry prompt word.

[0026] In the embodiment, the rewriting prompt word built in the rewriting large model is rewritten based on the query, and whether the user question needs to be rewritten is determined in combination with the context information; the rewriting prompt word includes but is not limited to a rewriting role prompt word (such as you are a user question understanding expert, your task is to understand the user question fully and express the user question accurately in combination with the context information) and a rewriting workflow (such as first determining the user core intent, and then determining whether the user question needs to be rewritten based on the user core intent); specifically, a rewriting role is defined based on an intelligent large model, and rewriting role prompt words, rewriting workflows and the like are provided; if the user question is not accurate enough, the user question is rewritten based on the context information to generate a query prompt word that can reflect the user core intent; if the user question has accurately expressed the core intent, the user question is taken as the query prompt word.

[0027] In the embodiment, the mixed retrieval strategy includes a full-text retrieval strategy and a semantic retrieval strategy; specifically, a plurality of first recall segments matched with the query prompt word are obtained from the target knowledge base according to the full-text retrieval strategy, a plurality of second recall segments semantically related to the query prompt word are obtained from the target knowledge base according to the semantic retrieval strategy, each recall segment is comprehensively sorted, and a content segment with a high sorting position is taken as the recall content related to the query prompt word.

[0028] In the embodiment, the mixed retrieval strategy includes a full-text retrieval strategy and a semantic retrieval strategy; specifically, a plurality of first recall segments matched with the query prompt word are obtained from the target knowledge base according to the full-text retrieval strategy, a plurality of second recall segments semantically related to the query prompt word are obtained from the target knowledge base according to the semantic retrieval strategy, each recall segment is comprehensively sorted, and a content segment with a high sorting position is taken as the recall content related to the query prompt word.

[0029] In an embodiment, step S130 includes: after extracting a target keyword from the query prompt word, a plurality of first content segments matched with the target keyword are retrieved from the target knowledge base; after obtaining the semantic features of the query prompt word, a plurality of second content segments semantically related to the query prompt word are retrieved from the target knowledge base based on the semantic features; the first content segments and the second content segments are comprehensively sorted, and a content segment with a high sorting position is taken as the recall content related to the query prompt word.

[0030] In the embodiment, the mixed retrieval strategy specifically includes the following steps: 1. Perform a full-text retrieval strategy: after extracting a target keyword (such as a specific noun, a proper noun, an abbreviation, etc.) from the query prompt word, a plurality of first content segments matched with the target keyword are retrieved from the target knowledge base; 2. Perform semantic retrieval strategy: after obtaining the semantic features of the query prompt words, retrieve a plurality of second content segments related to the semantic features of the query prompt words in the target knowledge base based on the semantic features; wherein the semantic features include context information of each character in the query prompt words or association information between adjacent words in the query prompt words; 3. Comprehensive processing: comprehensively sorting the first content segment and the second content segment, and taking the content segment with high ranking as the recall content related to the query prompt words.

[0031] The mixed retrieval strategy combines the technical advantages of full-text retrieval and semantic retrieval. Based on the mixed retrieval strategy, the knowledge base node retrieves the recall content with the key word precise matching characteristics of full-text retrieval and the deep semantic association characteristics of semantic retrieval from the target knowledge base.

[0032] In an embodiment, the comprehensive sorting of the first content segment and the second content segment, and taking the content segment with high ranking as the recall content related to the query prompt words, comprises: comprehensively scoring the first content segment and the second content segment based on the key word matching degree and the semantic association degree; after sorting from high to low according to the comprehensive score of each content segment, selecting the content segment with high ranking and comprehensive score not lower than a preset threshold as the recall content related to the query prompt words.

[0033] In this embodiment, the first content segment and the second content segment are comprehensively scored based on the key word matching degree (measuring the matching degree of the first content segment and the target key word) and the semantic association degree (measuring the semantic correlation degree of the second content segment and the query prompt words). After sorting from high to low according to the comprehensive score of each content segment, the content segment with high ranking and comprehensive score not lower than a preset threshold is selected as the recall content related to the query prompt words. The mixed retrieval strategy combines the technical advantages of full-text retrieval and semantic retrieval. Based on the mixed retrieval strategy, the knowledge base node retrieves the recall content with the key word precise matching characteristics of full-text retrieval and the deep semantic association characteristics of semantic retrieval from the target knowledge base.

[0034] S140, integrating the recall content based on the context information to generate corresponding reply content.

[0035] In this embodiment, the recall content is semantically fused and logically reorganized based on the context information (such as context history dialogue), so that the generated reply content not only meets the core needs of the current query prompt words, but also maintains semantic coherence with the historical interaction. Through the cooperation of context history dialogue association and historical FAQ knowledge base (i.e. FAQ database) reuse, the present application realizes dynamic optimization of reply content, and ensures the accuracy and applicability of the output result.

[0036] In an embodiment, step S140 comprises: integrating the recalled content based on product answer prompt words built in the product answer large model in combination with the context information to generate corresponding reply content.

[0037] In the present embodiment, the recalled content is integrated based on product answer prompt words built in the product answer large model in combination with the context information to generate corresponding reply content; the product answer prompt words include but are not limited to product answer role prompt words, product answer workflow, product answer restrictions, product answer question and answer examples; specifically, a product answer role is defined based on an intelligent large model, and product answer role prompt words, product answer workflow, product answer restrictions, product answer question and answer examples and other information are provided.

[0038] When the target business is related to charging station operation, the product answer role prompt words can be set as "the product answer role is a smart parking customer assistant, as an intelligent customer service of smart parking business, specially answering the problems encountered by users in the product use process, with all introduction content and use question and answer knowledge of smart parking business, the task is to provide accurate answers for user questions based on the above knowledge", and the product answer workflow can be set as "step one: understand the recalled content and user questions, and determine whether the recalled content matches the user questions; if it cannot be matched and / or the user questions cannot be understood (such as incomplete information, etc.), the user is asked to clarify the core intention of the user; step two: if it matches, the recalled content is semantically integrated and logically reorganized to output accurate and concise reply content".

[0039] S150, constructing structured data according to the user questions and the corresponding reply content, and writing the structured data into the FAQ database to complete the update of the target knowledge base.

[0040] In the present embodiment, structured data is constructed according to the user questions and the corresponding reply content, the structured data includes fields of user questions, reply content, submission date, question classification, solution status, etc., the structured data is written into the FAQ database with online record table function, and the target knowledge base (i.e. the FAQ database) is updated by covering the original repeated question and answer and adding high-quality question and answer, so as to realize dynamic optimization of the knowledge base.

[0041] In an embodiment, step S150 comprises: constructing structured data according to the user questions and the corresponding reply content; sending the structured data to a preset audit node to obtain an audit result returned based on the structured data; if the audit result is passed, writing the structured data into the FAQ database to complete the update of the target knowledge base.

[0042] In the embodiment, the table structure configuration of the FAQ database is specifically: a target data table is selected, a row where a table header is located and a data starting row are set, a "common question" field is set as a table structure index, data types of the "common question" and "solution" fields are both configured as string types, and the structured configuration is used to realize the standard storage of data.

[0043] According to the user question and the corresponding reply content, structured data is constructed, the structured data includes fields of a user question, reply content, a submission date, question classification, a solution state and the like, the structured data is sent to a preset audit node (audited by an operation and maintenance personnel), an audit result returned based on the structured data is acquired, if the audit result is passed, based on the table structure configuration of the FAQ database, the "user question" in the structured data is standardized and stored in the "common question" field, the "reply content" is directly mapped to the "solution" field, the writing of the structured data to the FAQ database and the updating of the target knowledge base are completed, and if the audit result is not passed, the structured data is not written into the FAQ database.

[0044] In summary, the application accurately classifies user questions through the intent recognition large model, ensures that business-related questions are processed in a targeted manner, generates inquiry prompt words combined with context information and uses a hybrid retrieval strategy to improve the accuracy and relevance of the recalled content of the target knowledge base, integrates the recalled content to generate reply content combined with context information to enhance the logicality and naturalness of the reply content, and updates the target knowledge base through structured data to realize self-iterative optimization of the target knowledge base, thereby significantly improving the intent recognition accuracy, reply quality and service efficiency of the intelligent customer service system.

[0045] Figure 2 A schematic block diagram of an intelligent customer service processing device based on a large language model is provided for the embodiments of the application. Figure 2 As shown in the figure, corresponding to the above intelligent customer service processing method based on a large language model, the application also provides an intelligent customer service processing device based on a large language model, which is configured in a controller of an intelligent customer service system. Specifically, please refer to Figure 2 The intelligent customer service processing device based on a large language model 700 includes: A classification unit 701 configured to call an intent recognition large model to classify a user question and obtain a classification result; A rewriting unit 702 configured to, if the classification result is that the user question is related to a target business, rewrite the user question combined with context information to generate a corresponding inquiry prompt word; The recall unit 703 is configured to invoke a knowledge base node to obtain recall content related to the inquiry prompt word from a target knowledge base based on a preset mixed retrieval strategy; the target knowledge base includes a business-related knowledge base and an FAQ database. The integration unit 704 is configured to integrate the recall content based on the context information to generate corresponding reply content. The update unit 705 is configured to construct structured data according to the user question and the corresponding reply content, and write the structured data into the FAQ database to complete updating of the target knowledge base.

[0046] In some embodiments, after the step of classifying the user question by invoking the intent recognition large model to obtain a classification result, the classification unit 701 is further configured to: If the classification result is that the user question is irrelevant to the target business, a non-product answer large model is invoked to generate answer content, and the user question is replied based on the answer content.

[0047] In some embodiments, when performing the step of rewriting the user question based on the context information to generate corresponding inquiry prompt words, the rewriting unit 702 is specifically configured to: Based on a rewriting prompt word built in an inquiry rewriting large model, it is determined whether the user question needs to be rewritten based on the context information; if the user question needs to be rewritten, the user question is rewritten based on the context information to generate inquiry prompt words that can reflect the core intent of the user; and if the user question does not need to be rewritten, the user question is taken as the inquiry prompt word.

[0048] In some embodiments, when performing the step of invoking a knowledge base node to obtain recall content related to the inquiry prompt word from a target knowledge base based on a preset mixed retrieval strategy, the recall unit 703 is specifically configured to: After extracting a target keyword from the inquiry prompt word, a plurality of first content segments matching the target keyword are retrieved from the target knowledge base; after obtaining the semantic features of the inquiry prompt word, a plurality of second content segments semantically related to the inquiry prompt word are retrieved from the target knowledge base based on the semantic features; and the first content segments and the second content segments are comprehensively sorted, and the content segments with high ranking are taken as the recall content related to the inquiry prompt word.

[0049] In some embodiments, when performing the step of comprehensively sorting the first content segments and the second content segments, and taking the content segments with high ranking as the recall content related to the inquiry prompt word, the recall unit 703 is specifically configured to: The first content segment and the second content segment are comprehensively scored based on keyword matching degree and semantic association degree; after ordering from high to low according to the comprehensive scores of each content segment, a content segment with a high ranking and a comprehensive score not lower than a preset threshold is selected as recall content related to the query prompt word.

[0050] In some embodiments, the integration unit 704, when performing the step of integrating the recall content in combination with the context information to generate corresponding reply content, is specifically configured to: Integrating the recall content in combination with the context information based on the product answer prompt word built in the product answer large model to generate corresponding reply content.

[0051] In some embodiments, the updating unit 705, when performing the step of updating the target knowledge base according to the user question and the corresponding reply content to construct structured data and writing the structured data into the FAQ database, is specifically configured to: Constructing structured data according to the user question and the corresponding reply content; sending the structured data to a preset audit node to obtain an audit result returned based on the structured data; if the audit result is audit passed, writing the structured data into the FAQ database to complete the updating of the target knowledge base.

[0052] It should be noted that the specific implementation process of the above-mentioned intelligent customer service processing device based on a large language model and each unit can be clearly understood by those skilled in the art, which can be referred to the corresponding description in the foregoing method embodiments. For the convenience and brevity of description, it will not be repeated here.

[0053] The above-mentioned intelligent customer service processing device based on a large language model can be realized in the form of a computer program, which can run on an electronic device as shown in Figure 3 .

[0054] Please refer to Figure 3 , Figure 3 is a schematic block diagram of an electronic device provided by an embodiment of the present application. The electronic device 800 can be a terminal or a server, wherein the terminal can be an electronic device with communication function. The server can be a stand-alone server or a server cluster composed of multiple servers.

[0055] Please refer to Figure 3 , the electronic device 800 includes a processor 802, a memory and a network interface 805 connected through a system bus 801, wherein the memory can include a non-volatile storage medium 803 and an internal memory 804.

[0056] The non-volatile storage medium 803 can store an operating system 8031 and a computer program 8032. The computer program 8032 includes program instructions that, when executed, cause the processor 802 to perform an intelligent customer service processing method based on a large language model.

[0057] The processor 802 is configured to provide computing and control capabilities to support the operation of the entire electronic device 800.

[0058] The memory 804 provides an environment for the execution of the computer program 8032 in the non-volatile storage medium 803, which, when executed by the processor 802, causes the processor 802 to perform an intelligent customer service processing method based on a large language model.

[0059] The network interface 805 is configured to communicate with other devices over a network. Those skilled in the art can understand that the network interface 805 can be configured to communicate with other devices over a network according to the structure shown in the figure, and does not constitute a limitation on the electronic device 800 to which the present application is applied. The specific electronic device 800 can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. Figure 3

[0060] The processor 802 is configured to run the computer program 8032 stored in the memory to implement the following steps: The processor 802 is configured to run the computer program 8032 stored in the memory to implement the following steps:

[0061] In some embodiments, the processor 802, after implementing the step of classifying the user question by calling the intent recognition large model to obtain a classification result, further implements the following steps: If the classification result is that the user question is not related to the target business, the processor 802 is configured to call a non-product answer large model to generate answer content and reply to the user question based on the answer content.

[0062] In some embodiments, the processor 802, when implementing the step of rewriting the user question in combination with the context information to generate a corresponding inquiry prompt word, specifically implements the following steps: ​The built-in rewriting prompt word in the rewriting large model is queried to determine whether the user question needs to be rewritten based on the context information; if the user question needs to be rewritten, the user question is rewritten based on the context information to generate a query prompt word that can reflect the core intent of the user; if the user question does not need to be rewritten, the user question is taken as a query prompt word.

[0063] In some embodiments, when the processor 802 implements the preset-based hybrid retrieval strategy and calls the knowledge base node to obtain the recall content related to the query prompt word from the target knowledge base, the following steps are implemented: After extracting the target keyword from the query prompt word, a plurality of first content segments matching the target keyword are retrieved in the target knowledge base; after obtaining the semantic features of the query prompt word, a plurality of second content segments semantically related to the query prompt word are retrieved in the target knowledge base based on the semantic features; and the first content segments and the second content segments are comprehensively sorted, and the content segments with high ranking are taken as the recall content related to the query prompt word.

[0064] In some embodiments, when the processor 802 implements the step of comprehensively sorting the first content segments and the second content segments and taking the content segments with high ranking as the recall content related to the query prompt word, the following steps are implemented: The first content segments and the second content segments are comprehensively scored based on the keyword matching degree and the semantic correlation degree; after sorting from high to low according to the comprehensive scores of each content segment, the content segments with high ranking and comprehensive scores not lower than a preset threshold are selected as the recall content related to the query prompt word.

[0065] In some embodiments, when the processor 802 implements the step of integrating the recall content based on the context information to generate corresponding reply content, the following steps are implemented: The recall content is integrated based on the product answer prompt word built in the product answer large model and the context information to generate corresponding reply content.

[0066] In some embodiments, when the processor 802 implements the step of constructing structured data according to the user question and the corresponding reply content, writing the structured data into the FAQ database to complete the update of the target knowledge base, the following steps are implemented: The structured data is constructed according to the user question and the corresponding reply content; the structured data is sent to a preset audit node to obtain an audit result returned based on the structured data; if the audit result is passed, the structured data is written into the FAQ database to complete the update of the target knowledge base.

[0067] It should be understood that in the embodiments of the present application, the processor 802 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0068] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiments.

[0069] Therefore, the present application also provides a storage medium. The storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. The program instructions are executed by a processor to make the processor perform the following steps: invoking an intent recognition large model to classify the user question to obtain a classification result; if the classification result is that the user question is related to a target business, rewriting the user question in combination with context information to generate a corresponding inquiry prompt word; based on a preset hybrid retrieval strategy, invoking a knowledge base node to obtain recall content related to the inquiry prompt word from a target knowledge base; the target knowledge base includes a business-related knowledge base and an FAQ database; integrating the recall content in combination with the context information to generate corresponding reply content; and constructing structured data according to the user question and the corresponding reply content, and writing the structured data into the FAQ database to complete updating of the target knowledge base.

[0070] In an embodiment, the processor, after executing the program instructions to implement the step of invoking an intent recognition large model to classify the user question to obtain a classification result, further implements the following steps: if the classification result is that the user question is not related to the target business, invoking a non-product answer large model to generate answer content, and replying to the user question based on the answer content.

[0071] In an embodiment, the processor, when executing the program instructions to implement the step of rewriting the user question in combination with the context information to generate the corresponding inquiry prompt word, specifically implements the following steps: Based on the rewriting prompt word built in the inquiry rewriting large model, it is judged whether the user question needs to be rewritten in combination with the context information; if the user question needs to be rewritten, the user question is rewritten based on the context information to generate an inquiry prompt word that can reflect the user's core intent; if the user question does not need to be rewritten, the user question is taken as an inquiry prompt word.

[0072] In an embodiment, the processor, when executing the program instructions to implement the step of calling the knowledge base node to obtain the recall content related to the inquiry prompt word from the target knowledge base based on the preset hybrid retrieval strategy, specifically implements the following steps: After extracting the target keyword from the inquiry prompt word, a plurality of first content segments matching the target keyword are retrieved in the target knowledge base; after obtaining the semantic features of the inquiry prompt word, a plurality of second content segments semantically related to the inquiry prompt word are retrieved in the target knowledge base based on the semantic features; the first content segments and the second content segments are comprehensively sorted, and the content segments with high ranking are taken as the recall content related to the inquiry prompt word.

[0073] In an embodiment, the processor, when executing the program instructions to implement the step of comprehensively sorting the first content segments and the second content segments, and taking the content segments with high ranking as the recall content related to the inquiry prompt word, specifically implements the following steps: The first content segments and the second content segments are comprehensively scored based on the keyword matching degree and the semantic correlation degree; after sorting from high to low according to the comprehensive scores of each content segment, the content segments with high ranking and comprehensive scores not lower than a preset threshold are selected as the recall content related to the inquiry prompt word.

[0074] In an embodiment, the processor, when executing the program instructions to implement the step of integrating the recall content in combination with the context information to generate the corresponding reply content, specifically implements the following steps: The recall content is integrated in combination with the context information based on the product answer prompt word built in the product answer large model to generate the corresponding reply content.

[0075] In an embodiment, the processor, when executing the program instructions to implement the step of constructing structured data according to the user question and the corresponding reply content, and writing the structured data into the FAQ database to complete the update of the target knowledge base, specifically implements the following steps: According to the user question and the corresponding reply content, structured data is constructed; the structured data is sent to a preset audit node, an audit result returned based on the structured data is acquired; if the audit result is an audit pass, the structured data is written into the FAQ database to complete updating of the target knowledge base.

[0076] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, and various computer readable storage media that can store program codes.

[0077] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0078] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented by other ways. For example, the apparatus embodiments described above are only schematic. The division of the units is only a logical function division, and there can be another division in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0079] The steps in the method embodiments of the present application can be adjusted, combined and reduced in sequence according to actual needs. The units in the apparatus embodiments of the present application can be combined, divided and reduced according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0080] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing an electronic device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0081] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent customer service processing based on a large language model, characterized in that, The method includes: The intent recognition model is invoked to classify user questions and obtain classification results. If the classification result indicates that the user question is related to the target business, the user question is rewritten based on the context information to generate corresponding query prompts; Based on a preset hybrid retrieval strategy, knowledge base nodes are invoked to retrieve recall content related to the query prompts from the target knowledge base; the target knowledge base includes a business-related knowledge base and a FAQ database. The recalled content is integrated based on the context information to generate corresponding response content; Structured data is constructed based on user questions and corresponding responses, and the structured data is written into the FAQ database to complete the update of the target knowledge base.

2. The intelligent customer service processing method based on a large language model according to claim 1, characterized in that, The large-scale model for identifying call intent classifies user questions, and after obtaining the classification results, it also includes: If the classification result indicates that the user's question is unrelated to the target business, the non-product answer model is invoked to generate answer content, and the user's question is answered based on the answer content.

3. The intelligent customer service processing method based on a large language model according to claim 1, characterized in that, The process of rewriting the user's question by combining contextual information to generate corresponding prompt words includes: Based on the rewriting prompts built into the query rewriting model, and combined with the context information, it is determined whether the user question needs to be rewritten. If it is necessary to rewrite the user question, the user question is rewritten based on the context information to generate query prompts that reflect the user's core intent; If there is no need to rewrite the user question, use the user question as the prompt word.

4. The intelligent customer service processing method based on a large language model according to claim 1, characterized in that, The preset hybrid retrieval strategy calls knowledge base nodes to retrieve recall content related to the query prompt words from the target knowledge base, including: After extracting the target keyword from the query prompt, multiple first content fragments matching the target keyword are retrieved from the target knowledge base; After obtaining the semantic features of the query prompt, multiple second content fragments semantically related to the query prompt are retrieved from the target knowledge base based on the semantic features; The first content fragment and the second content fragment are sorted together, and the content fragments with the highest ranking are used as recall content related to the query prompt words.

5. The intelligent customer service processing method based on a large language model according to claim 4, characterized in that, The step of comprehensively sorting the first content fragment and the second content fragment, and using the content fragments ranked higher as the recall content related to the query prompt, includes: The first and second content segments are comprehensively scored based on keyword matching degree and semantic relevance. After sorting the content segments from highest to lowest based on their overall scores, the content segments with the highest overall scores and no lower than a preset threshold are selected as the recall content related to the query prompts.

6. The intelligent customer service processing method based on a large language model according to claim 1, characterized in that, The step of integrating the recalled content with the context information to generate corresponding response content includes: Based on the product answer prompts built into the product answer model, the recalled content is integrated with the context information to generate corresponding response content.

7. The intelligent customer service processing method based on a large language model according to claim 1, characterized in that, The step of constructing structured data based on user questions and corresponding responses, and writing the structured data into the FAQ database to update the target knowledge base, includes: Structured data is built based on user questions and corresponding responses; The structured data is sent to a preset review node to obtain the review result returned based on the structured data; If the review result is "approved", the structured data will be written into the FAQ database to complete the update of the target knowledge base.

8. An intelligent customer service processing device based on a large language model, characterized in that, The device includes: The classification unit is used to call the intent recognition model to classify the user's question and obtain the classification result; The rewriting unit is used to rewrite the user question in combination with context information and generate corresponding query prompt words if the classification result indicates that the user question is related to the target business. The recall unit is used to retrieve recall content related to the query prompt words from the target knowledge base by calling knowledge base nodes based on a preset hybrid retrieval strategy; the target knowledge base includes a business-related knowledge base and a FAQ database. An integration unit is used to integrate the recalled content with the context information and generate corresponding response content; The update unit is used to construct structured data based on user questions and corresponding responses, and write the structured data into the FAQ database to complete the update of the target knowledge base.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent customer service processing method based on a large language model as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which includes program instructions that, when executed by a processor, cause the processor to perform the intelligent customer service processing method based on a large language model as described in any one of claims 1-7.

Images