Customer service reply method, device and equipment based on service agent

Abstract

The invention relates to the field of computers, and discloses a customer service reply method and device based on a business agent, equipment and a storage medium. The method comprises the steps of obtaining a dialogue request sent by a user; querying configuration information of the intelligent agent according to the identifier of the intelligent agent; obtaining a question input by a user, and retrieving a plurality of pieces of related data corresponding to the question through a configured retriever; through a configured rearrangement model, according to the semantic relevancy of the multiple pieces of related data and the problem, the multiple pieces of related data are optimized and sorted, and related data conforming to the semantic meaning of the problem is obtained; and generating a reply through the configured AI model according to the related data and the problem conforming to the semantic meaning of the problem, and returning the generated reply to the client. In the embodiment of the invention, different agents can be used for different business fields, and the accuracy of the reply content of the AI large model is improved.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a customer service reply method, device, equipment and storage medium based on business intelligence. Background Art

[0002] As the integrated budget management platforms at the provincial level across the country are gradually improved, the scope of business coverage is getting wider and wider, the number of users is gradually increasing, and the demand for business consulting services is increasing. Therefore, higher requirements are put forward for the operation and maintenance of the integrated budget management platform, such as whether customer consultations can be handled in a timely manner, whether problems can be solved more accurately, and whether the operation services can be more efficient.

[0003] The accuracy of the responses of existing financial customer service based on the AI ​​big model is low. For example, procurement business exists in different business systems, and when asking business question A, the answer is related to business B, and the efficiency of searching the knowledge base is low, all of which affect the accuracy of the responses of the AI ​​big model. Summary of the invention

[0004] The main purpose of this invention is to solve the technical problem of how to improve the accuracy of the AI ​​large model reply content.

[0005] A first aspect of the present invention provides a customer service response method based on a business intelligence agent, and the customer service response method based on a business intelligence agent comprises: Acquire a dialog request sent by a user, wherein the dialog request carries an agent selected by the user, and the agent is related to a business field; According to the identification of the agent, query the configuration information of the agent, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; Obtaining a question input by a user, and retrieving multiple pieces of relevant data corresponding to the question through a configured retriever; By using the configured rearrangement model, the plurality of relevant data are optimized and sorted according to the semantic relevance between the plurality of relevant data and the question, so as to obtain relevant data that conforms to the semantics of the question; Based on the relevant data that conforms to the semantics of the question and the question, a response is generated through the configured AI model, and the generated response is returned to the client.

[0006] Optionally, in a first implementation of the first aspect of the present invention, obtaining a question input by a user and retrieving a plurality of relevant data corresponding to the question through a configured retriever includes: According to the configuration information of the agent, loading the configured retriever; Obtain the question input by the user, and convert the text corresponding to the question into a vector representation through a preset vectorization model; According to the vectorized question text, multiple pieces of relevant data corresponding to the question are retrieved through the configured retriever.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the retrieving, according to the vectorized question text, the plurality of relevant data corresponding to the question through the configured retriever comprises: According to the configured retriever, the vectorized question text is input into a preset vertical business vector library for similarity retrieval; The vectorized question text is compared with the vector data in the vertical business vector library by using an approximate nearest neighbor algorithm, and multiple pieces of relevant data corresponding to the question are retrieved.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the configuration information further includes tool information, and the tool information is used to configure a tool interface to provide an external interaction function for the agent; The step of generating a reply based on the relevant data that conforms to the semantics of the question and the question through a configured AI model, and returning the generated reply to the client comprises: Acquire the user's historical conversation record, combine the relevant data that conforms to the question semantics, the question, the tool information, and the historical conversation record to generate combined information; Passing the combined information to a configured AI model, and generating a response through the AI ​​model; The generated reply is returned to the client, so that the client renders and displays the reply according to a preset message format.

[0009] Optionally, in a fourth implementation manner of the first aspect of the present invention, the transmitting the combined information to a configured AI model, and generating a reply through the AI ​​model includes: Obtaining a user ID, and based on the user ID, loading user-related context information through a database or cache query; Passing the context information and the combined information to a configured AI model; A response is generated through the AI ​​model.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, before obtaining the dialog request sent by the user, the method further includes: Acquire diversified data and construct a knowledge base based on the diversified data, wherein the diversified data is acquired by at least one of file uploading, website synchronization and API access; Updating the content of the knowledge base through API and website synchronization mechanism; The historical version records of the knowledge base are recorded to roll back the updated or misoperated contents of the knowledge base.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, after generating a reply according to the relevant data that conforms to the semantics of the question and the question through the configured AI model and returning the generated reply to the client, it also includes: Evaluating the agent using a preset number of evaluations and a standardized score; According to the evaluation results, the agent is optimized.

[0012] A second aspect of the present invention provides a customer service response device based on a business intelligence agent, and the customer service response device based on a business intelligence agent comprises: A dialogue request acquisition module is used to acquire a dialogue request sent by a user, wherein the dialogue request carries an agent selected by the user, and the agent is related to a business field; A configuration information query module, used to query the configuration information of the agent according to the identifier of the agent, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; A question input acquisition module is used to obtain a question input by a user and retrieve multiple pieces of relevant data corresponding to the question through a configured retriever; An optimization sorting module, used to optimize and sort the multiple pieces of relevant data according to the semantic relevance between the multiple pieces of relevant data and the question through a configured rearrangement model, so as to obtain relevant data that meets the semantics of the question; The reply generation module is used to generate a reply based on the relevant data that conforms to the semantics of the question and the question through the configured AI model, and return the generated reply to the client.

[0013] The third aspect of the present invention provides a customer service response device based on a business intelligence entity, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line; the at least one processor calls the instructions in the memory so that the customer service response device based on the business intelligence entity executes the above-mentioned customer service response method based on the business intelligence entity.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned customer service response method based on business intelligence.

[0015] In an embodiment of the present invention, a dialog request sent by a user is obtained; according to the identifier of the agent, the configuration information of the agent is queried, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; the question input by the user is obtained, and multiple relevant data corresponding to the question are retrieved through the configured retriever; through the configured rearrangement model, the multiple relevant data are optimized and sorted according to the semantic relevance of the multiple relevant data to the question to obtain relevant data that conforms to the semantics of the question; according to the relevant data that conforms to the semantics of the question and the question, a reply is generated through the configured AI model, and the generated reply is returned to the client. In the present invention, according to the configuration information of the agent, multiple relevant data corresponding to the question input by the user are retrieved through the configured retriever, and then, through the configured rearrangement model, the multiple relevant data are optimized and sorted to obtain relevant data that conforms to the semantics of the question and generate a reply. Different agents can be used for different business fields to improve the accuracy of the reply content of the AI ​​large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of an embodiment of a customer service reply method based on a business agent in an embodiment of the present invention; Figure 2 A schematic diagram of the logical interaction of the intelligent agent configuration in an embodiment of the present invention; Figure 3 A schematic diagram of a flow chart of a user interaction method in an embodiment of the present invention; Figure 4 A schematic diagram of an embodiment of a customer service reply device based on a business agent in an embodiment of the present invention; Figure 5 It is a schematic diagram of an embodiment of a customer service response device based on a business intelligence agent in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present invention provide a customer service reply method, device, equipment and storage medium based on business intelligence.

[0018] The embodiments disclosed in the present invention will be described in more detail below with reference to the accompanying drawings. Although certain embodiments disclosed in the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments disclosed in the present invention are only for exemplary purposes and are not intended to limit the scope of protection disclosed in the present invention.

[0019] In the description of the embodiments disclosed in the present invention, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a customer service reply method based on a business agent in an embodiment of the present invention includes: S100: Obtain a conversation request sent by a user.

[0021] In this embodiment, the agent is first configured, as follows: Figure 2 As shown, the following configuration is included: (1) Define AI model management, configure model parameters, and enter basic model parameters through AI model management, such as AI model name, model type, basic model, API domain name, APIKey, maximum number of tokens, model temperature, model chart, enabled status, whether to enable request logs, and whether to enable corresponding logs.

[0022] (2) Define vector model management, configure model parameters, and enter basic model parameters through AI model management, such as vector model name, model type, request type, API domain name, APIKey, dimension, description, model icon, enabled status, timeout, number of restarts, whether to enable request logs, and whether to enable corresponding logs.

[0023] (3) Define reorder model management, configure model parameters, and enter basic model parameters through AI model management, such as reorder model name, model type, API domain name, APIKey, minimum score, maximum number of entries, description, model icon, enabled status, timeout, number of restarts, whether to enable request logs, and whether to enable corresponding logs.

[0024] (4) Define retriever management, configure model parameters, and enter basic retriever parameters through retriever management, such as retriever name, retriever type, retriever icon, and enabled status.

[0025] (5) Define tool management. Tool management is used to configure business system API interfaces or third-party system APIs. Parameters include tool name, tool identification name, tool type (type includes http request, ToolApI, low-code API (process orchestration, input parameters [serial number, parameter name, parameter description, parameter type, whether required], return parameters [serial number, parameter name, parameter description, parameter type])), tool description, tool icon, and enabled status.

[0026] (6) Define the knowledge base, create different knowledge bases according to categories by configuring knowledge base management, provide different knowledge base knowledge entry methods and training document content maintenance, including parameters such as the knowledge base, knowledge source, knowledge tag, question, whether it is a common question, whether it is a general question, knowledge label, knowledge content, and answer.

[0027] (7) Definition and configuration of vertical business intelligence entities, as follows: Create an agent: agent parameters such as agent name, agent nickname, agent description, and agent icon.

[0028] Configure the agent: Configuration parameters include AI model, streaming output, retriever settings {retrieval vector information, parameters include retriever name, retriever type (sparse dense mixed retrieval, milvus retriever, BM25 retriever, ES mixed retrieval), associated knowledge base, parameter settings, where parameter settings include retrieval vector field configuration (such as field name, dense / sparse, maximum number of entries retained, vector model), total number of returned entries (1-100 positive integers), returned field name, similarity (0-1), sorter (weight sorter (each field weight is between 0-1), Rf sorter (factor 0-1)), vector database, metadata retrieval conditions, rearranger settings, user message template (custom prompt word or system default), RAG query message template (system default or custom prompt word), chat storage (storage type such as memory, MONGODB, REDIS, number of chat entries), tool selection (select third-party API tools).

[0029] (8) Agent evaluation: Agent evaluation is used to evaluate the accuracy of the question-and-answer knowledge base. It specifies the evaluator agent and the evaluated agent, and sets evaluation configuration parameters such as average pass rate %, single pass rate 0.0-5.0, and evaluation methods include automatic (setting the number of evaluations, automatic knowledge evaluation) and manual selection of question and answer content and reference answers, and importing evaluation question content and reference answers.

[0030] Specifically, users log in to the system through the client and interact with the customer service component. Users select a specific agent to initiate a conversation, which is usually related to a specific business field (such as finance, medical care, customer service, etc.).

[0031] S200, querying the configuration information of the intelligent agent according to the identification of the intelligent agent.

[0032] In this embodiment, after the user selects a specific agent on the front-end interface to initiate a conversation, the browser will transmit the agent's ID to the back-end server. The agent's ID is a unique identification code for each agent, and based on the agent's ID, the configuration information of the corresponding agent is queried through the API interface. The configuration information includes AI model configuration, retriever configuration, rearrangement model configuration, etc.

[0033] S300, obtaining a question input by a user, and retrieving a plurality of relevant data corresponding to the question through a configured retriever.

[0034] In this embodiment, the user enters a prompt (user's request or question) in the dialog box. The browser transmits data containing the agent ID and prompt content to the backend server through an HTTP request. This data transmission uses a standard Web request protocol (such as HTTP / HTTPS) to ensure the reliability and security of the data. According to the configuration information of the agent, the server loads the associated retriever configuration. An agent may have multiple retriever configurations (1-N). These retrievers are used in different business scenarios. The retrievers include Milvus hybrid retriever, single vector retriever, BM25 retriever, ES hybrid retriever, etc. The configuration description of each retriever is as follows: (1) Milvus hybrid searcher: combines dense vector search and sparse keyword matching to provide semantic and keyword hybrid matching capabilities.

[0035] Retrieval field configuration: field name, dense / sparse type, maximum number of entries to be retained, and vector model.

[0036] Number of results returned: Set the upper limit of the number of results returned from 1 to 100.

[0037] Sorter: Supports weight sorter and correlation factor sorting.

[0038] (2) Single vector retriever: Based on high-dimensional vector matching, it focuses on semantic relevance retrieval and is suitable for processing scenarios with a single vector field.

[0039] Retrieve field configuration: field name, feature type (dense / sparse) and vector model.

[0040] Return field configuration: Specify the set of fields that need to be returned.

[0041] (3) BM25 search engine: It uses inverted indexing technology to perform sparse keyword retrieval and is suitable for traditional text search.

[0042] Maximum number of results to retain: sets the upper limit for the number of results.

[0043] Filter conditions: Filter results by similarity threshold.

[0044] Quote paragraph TOP: limit the number of quoted paragraphs.

[0045] Quote character limit: Limit the number of characters in the returned text content.

[0046] (4) ES hybrid searcher: combines the dense and sparse search capabilities of Elasticsearch and is suitable for multi-source data retrieval.

[0047] Retrieve field configuration: field name, feature type (dense / sparse) and vector model.

[0048] Number of returned results: The total return result limit is 1-100.

[0049] Sorting strategy: supports weight sorting and factor sorting.

[0050] S400, optimizing and sorting the multiple pieces of relevant data according to the semantic relevance of the multiple pieces of relevant data to the question through the configured rearrangement model, to obtain relevant data that conforms to the semantics of the question.

[0051] In this embodiment, according to the configuration information of the agent, the server loads the configured rearrangement model, which sorts the retrieved relevant data according to the semantic relevance between the retriever query data and the question, ensuring that the most relevant information is ranked first. The rearrangement is usually based on more complex deep learning models, such as BERT, RoBERTa, etc. After the rearrangement model is adjusted, 10*N pieces of data that best match the semantics are finally returned. The rearrangement not only considers the similarity of the vectors, but also optimizes the sorting of the specific semantics of the business requirements.

[0052] Users can configure and manage various reordering models. The system supports inputting different reordering strategies to optimize the sorting according to business needs. Users can flexibly adjust the reordering rules and strategies through this module to ensure that the returned results meet the specific business scenarios. The parameter configuration of the reordering model is as follows: Reflow model name: Specifies a unique identifier for the reflow model, which is used to distinguish different reflow strategies.

[0053] Model type: including but not limited to rule-based sorting models, machine learning sorting models, learning sorting models, etc. Different model types will affect the way the sorting results are generated.

[0054] API domain name: Provides the API address for accessing the reordering model to ensure that the system can interact with the reordering model service through network requests.

[0055] APIKey: A key used for identity authentication to ensure access control.

[0056] Minimum score: Set the minimum score threshold for the results to be returned when re-ranking. Only results that reach this score will be included in the sorting results, filtering out irrelevant or low-quality results.

[0057] Maximum number of results: Specifies the maximum number of results returned after rearrangement, which is used to limit the number of results returned after rearrangement to ensure a balance between response speed and data volume.

[0058] Description: Briefly describe the functions, application scenarios, and advantages of the rearrangement model to help users understand the role of the model.

[0059] Model icon: Provide an icon representation for the rearrangement model to facilitate user identification and management.

[0060] Enabled status: Indicates whether the rearrangement model is enabled. Users can choose to enable or disable the model according to their needs to ensure system flexibility.

[0061] Timeout: Set the timeout for the rescheduling request to prevent the system from causing unnecessary waiting due to timeout.

[0062] Restart times: Configure the restart times when an exception occurs during the operation of the rearrangement model to ensure system stability.

[0063] Request log: Whether to record request logs for later analysis, debugging, and optimization.

[0064] Response log: Whether to record the response log to track the response results of the model and make improvements.

[0065] Through the rearrangement model management, the retrieved data can be intelligently sorted and optimized according to the configured parameters and strategies, ensuring that the most relevant results can be accurately and efficiently returned in different business scenarios. At the same time, through logging and timeout, restart control and other functions, the stability and high availability of the system are ensured.

[0066] S500 generates a response based on the relevant data and questions that meet the semantics of the question through the configured AI model, and returns the generated response to the client.

[0067] In this embodiment, relevant data and questions that conform to the semantics of the question are passed to the configured AI model to generate appropriate responses, and the responses generated by the agent are returned to the client.

[0068] In an optional implementation of the first aspect of the present invention, obtaining a question input by a user and retrieving a plurality of relevant data corresponding to the question through a configured retriever includes: According to the configuration information of the intelligent agent, the configured retriever is loaded; the question input by the user is obtained, and the text corresponding to the question is converted into a vector representation through a preset vectorization model; according to the vectorized question text, multiple relevant data corresponding to the question are retrieved through the configured retriever.

[0069] In this embodiment, first, through vector model management, the user is allowed to define and configure basic parameters of the vector model. The user needs to input the following key parameters: Vector model name: This parameter serves as a unique identifier for the vector model, ensuring that the system can accurately distinguish and call different vector models.

[0070] Model type: This parameter defines the type of model, such as sparse vector model or dense vector model (such as deep learning-based models, BERT, CLIP, etc.).

[0071] Request type: Users can choose synchronous or asynchronous requests to meet different business needs. Synchronous requests are suitable for scenarios that require immediate response, while asynchronous requests are suitable for more complex and computationally intensive tasks.

[0072] API domain name: Configure the API domain name of the model service, which is used to identify the access entry address of the service.

[0073] APIKey: This parameter is used to provide a key for secure access to the vector model service, ensuring that only authorized requests can access the model interface.

[0074] Dimension: The dimension of vector representation determines the representation ability of the vector in high-dimensional space. The larger the dimension, the stronger the feature expression ability of the model, but it also brings higher computational cost.

[0075] Description: Briefly describe the application scenarios and functions of the model to help users understand the role of the model.

[0076] Model icon: To facilitate model identification and management, you can set an icon for visual identification.

[0077] In addition to the basic model configuration, runtime parameter configuration options are also provided to ensure that the system can work stably and reliably in various operating environments, as shown below: Enabled state: Indicates whether the current model is enabled. When the model is enabled, the model will participate in actual vector calculation and information retrieval tasks.

[0078] Timeout: Configure the timeout for model service requests. The timeout mechanism can avoid long-term request blocking and help improve system response speed and stability.

[0079] Restart times: Set the number of automatic restarts to ensure stable system operation through the automatic restart function when the model has an abnormality.

[0080] Request log: Configure whether to enable request logging. This function is used to record all requests to the model service to facilitate subsequent debugging and performance analysis.

[0081] Response log: Configure whether to enable response log recording. Record detailed information of each model response to facilitate analysis and optimization of model performance.

[0082] To ensure high availability of the system, the Vector Model Management module supports an automatic restart mechanism through flexible parameter settings, and is equipped with a logging function for detailed tracking of model requests and responses. These mechanisms can help administrators monitor the health of the model in real time and take timely measures to deal with potential failures. By configuring parameters such as dimensions, timeouts, and request logs, you can optimize retrieval performance, reduce latency, and improve retrieval accuracy based on actual business needs.

[0083] The vector model management module is mainly aimed at the needs of various financial business agents, providing flexible vector model configuration and management functions. In this module, different financial business agents can configure different search engines and specify corresponding vector models according to their specific business scenarios, so as to achieve accurate information retrieval, intelligent recommendation, semantic matching and other tasks. Through vector model management, agents can select appropriate vector models according to different application scenarios, such as using dense vector models for deep semantic understanding, or using sparse vector models to process large-scale data sets. The configuration of vector model parameters such as dimension, request type, API interface, etc. ensures that the needs of different financial businesses can be accurately met, and the system can provide efficient and stable support.

[0084] Specifically, for each retriever, the system uses a predefined vectorization model (such as BERT, Sentence-BERT, etc.) to vectorize the prompt input by the user. At this point, the vectorization process converts the question text into a digital form (vector), so that the subsequent computer can perform vector comparisons (such as similarity calculations). This step is a key step in natural language processing (NLP) and determines the accuracy of subsequent processing.

[0085] In an optional implementation of the first aspect of the present invention, according to the vectorized question text, searching for multiple pieces of relevant data corresponding to the question by a configured retriever includes: According to the configured retriever, the vectorized question text is input into the preset vertical business vector library for similarity retrieval; through the approximate nearest neighbor algorithm, the vectorized question text is compared with the vector data in the vertical business vector library to retrieve multiple relevant data corresponding to the question.

[0086] In this embodiment, through the retriever management module, various types of retrievers and their related parameters are configured to achieve efficient retrieval and optimized sorting of data. It can adapt to different business scenarios, support multiple types of retrieval strategies, and provide efficient and accurate data support for intelligent entities.

[0087] Each retriever type supports personalized parameter settings, including: Retrieval field configuration: define the field name, dense / sparse type, maximum number of entries to be retained, and associated vector model.

[0088] Return result configuration: set the number of returned results, return field name and sorting strategy.

[0089] Sorting strategy: supports weight sorter, field weight configuration (between 0-1) and relevance factor sorting (Rf sorter) to optimize the priority and relevance of search results.

[0090] The retriever management module supports real-time parameter adjustment to adapt to dynamically changing business needs, and ensures the stability and efficiency of the retriever through logging and performance monitoring. As a key component in the general method of generating vertical business agents, the retriever management module provides accurate data support for the agent. Its specific functions include: (1) Knowledge retrieval and matching: When the intelligent agent interacts with the knowledge base, the retriever optimizes data queries and improves the response quality.

[0091] (2) Data query and sorting optimization: Through the weight and correlation factor sorter, the intelligent agent is provided with priority-optimized retrieval results.

[0092] (3) Dynamically adapt to business needs: Adjust retrieval strategies according to business scenarios to provide flexible query capabilities for intelligent agents.

[0093] Specifically, the vectorized text data will be fed into the pre-built vertical business vector library for similarity retrieval. Each retriever is configured with one or more vector libraries in business fields, and the text vector will be compared with the vectors in the library to find similar or related content. In the similarity calculation, the server uses the approximate nearest neighbor (ANN) algorithm (for example, using libraries such as FAISS and HNSW) to perform similarity retrieval. The system compares the vectors according to the configured search fields, such as text, tags, etc., and returns the most relevant 10*N pieces of data. This process is usually calculated in parallel to improve efficiency.

[0094] In an optional implementation of the first aspect of the present invention, generating a reply through a configured AI model according to relevant data and questions that conform to the semantics of the question, and returning the generated reply to the client includes: Obtain the user's historical conversation records, combine relevant data, questions, tool information, and historical conversation records that conform to the question semantics, and generate combined information; pass the combined information to the configured AI model, and generate a reply through the AI ​​model; return the generated reply to the client, so that the client renders and displays the reply according to the preset message format.

[0095] In this embodiment, the tool management module supports the definition and calling of multiple types of tools, and can realize complex business logic through process arrangement, helping the agent to efficiently answer various financial business questions. The tool management module is an important component of generating agents for different financial businesses, and aims to provide the agent with rich external interaction capabilities through flexible configuration of tool interfaces.

[0096] The tool management module supports the configuration of the following tool types for function expansion in different scenarios: HTTP request tool: used for the interaction between intelligent agents and financial business systems or external data interfaces, such as obtaining financial policies, budget data or system status information.

[0097] ToolAPI tool: call built-in general function modules, such as report generation tools, problem classification tools or data verification modules.

[0098] Low-code API tools: Support multi-step process orchestration to handle complex business scenarios. For example, by calling multiple sub-tools to integrate data and generate the final analysis results.

[0099] Low-code API tools: Support multi-step process orchestration to handle complex business scenarios. For example, by calling multiple sub-tools to integrate data and generate the final analysis results.

[0100] Each tool requires key properties to be set, such as tool name, tool identifier name, tool type, tool description, and enabled status.

[0101] The tool management module supports detailed parameter configuration to ensure the accuracy and flexibility of the call, including: Input parameters: Define the input data required to call the tool, including parameter name, type (string, integer, Boolean, etc.), description, and whether it is required.

[0102] Output parameters: Define the structure and content of the data returned by the tool, such as field names, types, and descriptions.

[0103] The tool management module also supports the orchestration of call logic through a low-code interface to meet the automation needs of complex operations in financial business. For example: first call the HTTP interface to obtain financial budget data; then call the ToolAPI tool to verify and organize the data; finally integrate the data to generate user-readable feedback information or analysis reports.

[0104] To ensure the stability and traceability of the agent's tool calls, the tool management module provides complete logging and performance monitoring functions, including: Request log: records the input parameters and time information when the tool is called.

[0105] Response log: records the result data and status returned by the tool.

[0106] Performance monitoring: Analyze tool call latency, success rate and other indicators to optimize system efficiency.

[0107] At the same time, the tool management module supports defining icons for each tool for quick identification in the configuration interface. The tool status (enabled / disabled) can be dynamically adjusted according to business needs.

[0108] Specifically, according to the configuration information of the agent, the following information is combined: 1) the user's historical conversation records; 2) information and descriptions of related tools; 3) system prompts (such as help information, guides, etc.); 4) the final vector data and the user's original prompt. With this information, the server generates a complete request message and prepares to pass it to the AI ​​model of the agent. In order to assemble messages efficiently, a templated approach is usually adopted to define the message structure in advance and then dynamically fill it according to real-time data, which can effectively reduce repeated calculations and improve processing efficiency. The assembled message is passed to the AI ​​model in the agent for processing to generate a suitable reply. The reply generated by the agent will be returned to the client, and the server will transmit the result back to the browser through the HTTP response. The client browser renders and displays it according to the specified message format (such as JSON, XML, etc.), and the user will see the final response result.

[0109] In an optional implementation of the first aspect of the present invention, passing the combined information to the configured AI model, and generating a reply through the AI ​​model includes: Obtain the user ID, and based on the user ID, load user-related context information through database or cache query; pass the context information and combined information to the configured AI model; generate a response through the AI ​​model.

[0110] In this embodiment, the server also needs to query the context information (such as chat history) related to the current user through the user ID, which is important for the subsequent dialogue generation. The loading of context information can be optimized through database query or cache system (such as Redis) to optimize the response time. The context information and the input prompt are passed to the AI ​​model to generate a suitable response.

[0111] In an optional implementation manner of the first aspect of the present invention, before obtaining the dialog request sent by the user, the method further includes: Acquire diversified data and build a knowledge base based on the diversified data. The diversified data is acquired through at least one of file upload, Web site synchronization and API access; update the content of the knowledge base through API and website synchronization mechanism; record the historical version records of the knowledge base to roll back the updated or erroneous content of the knowledge base.

[0112] In this embodiment, the knowledge base management module supports the classification of knowledge bases by business category to meet the needs of different financial business scenarios. The knowledge base management module is the core support module for the generation and response capabilities of intelligent agents, and is mainly used to manage and maintain various knowledge resources in the financial intelligent operation and maintenance customer service system. Through flexible knowledge base creation and document construction methods, the module can efficiently support intelligent agents to answer diverse questions in financial business and provide strong knowledge guarantee for intelligent operation and maintenance customer service. The module supports the configuration of the following properties of the knowledge base: Belonging knowledge base: clarify the classification of the knowledge base in the system.

[0113] Knowledge base name: The unique identifier of the knowledge base, used to distinguish different categories of knowledge resources.

[0114] Knowledge base description: Explain the purpose of the knowledge base and the financial business areas it covers.

[0115] The knowledge base management module supports building knowledge from multiple document sources, including: Upload general documents: Import finance-related policies, regulations or operation and maintenance manuals by uploading files (such as PDF, Word, Excel, etc.).

[0116] New Q&A: By uploading Q&A files, you can batch build the associated data of questions and answers to provide standardized response content for the intelligent agent.

[0117] Manual entry: supports administrators to manually enter personalized knowledge points, suitable for specific problems or customized content.

[0118] Web site synchronization: By configuring website links, the latest financial policies, regulations or business rules can be synchronized regularly to achieve dynamic updates.

[0119] API access: By connecting to the API interface of external systems, the knowledge base content can be obtained and updated in real time, such as connecting to the tax system, budget management system, etc., and automatically synchronizing the latest data.

[0120] The knowledge base management module supports adding or expanding the knowledge base to quickly adapt to new financial business needs. For example, when a special fund policy is added, relevant documents can be directly uploaded or rules can be manually entered to quickly integrate it into the agent's response capabilities.

[0121] Specifically, it supports building a knowledge base from multiple methods such as file upload, manual entry, and API access at the same time to improve knowledge coverage and accuracy. It also supports dynamic updates. For the configured Web site or API access method, it supports regular or real-time updates to ensure that the knowledge base content is always up to date. Through version management, the historical version changes of the knowledge base are recorded, and the rollback of the content of incorrect operations or incorrect updates is supported. Furthermore, by using the function of uploading general documents, financial policies and regulations are imported into the knowledge base to provide the intelligent agent with reliable policy interpretation capabilities. Through the Web site synchronization function, the legal content in the knowledge base is regularly updated to ensure that the intelligent agent can answer the latest policy questions. Upload operation and maintenance manuals or flow charts to build a guiding knowledge base for the financial business system. For example, the specific operation steps of the budget adjustment or reimbursement process are entered to provide users with clear operation instructions. Use API access to connect with the financial operation and maintenance system to obtain operation manuals or solution data in real time. And through the new question and answer function, common questions and answers in financial business are constructed. For example, for questions raised by users such as "What materials are required for budget approval?", the intelligent agent can directly retrieve standard answers from the knowledge base to reply. Combined with the manual entry function, it can quickly supplement and improve the knowledge points not covered by the intelligent agent.

[0122] In an optional implementation of the first aspect of the present invention, after generating a reply based on relevant data and questions that conform to the semantics of the question through a configured AI model and returning the generated reply to the client, the method further includes: The agent is evaluated using a preset number of evaluations and standardized scores, and the agent is optimized based on the evaluation results.

[0123] In this embodiment, the agent maintenance module allows users to create, define and configure different types of agents to meet the needs of various financial services. The agent maintenance module is designed to ensure that the agents generated in the financial intelligent operation and maintenance customer service system can operate stably and continuously optimize their service capabilities. This module provides functions such as agent definition, configuration, evaluation, update and monitoring, supports the full life cycle management of agents, and ensures that they can always efficiently and accurately answer various financial business questions. The definition of an agent includes the following aspects: Agent Name: A name that uniquely identifies the agent.

[0124] Agent nickname: an alias for easy user identification.

[0125] Agent Description: Briefly describe the purpose of the agent or the business problem it can solve.

[0126] Agent Icon: Set an icon for each agent to facilitate identification in the interface.

[0127] The configuration function includes the association setting between the agent and different models (AI model, vector model, etc.), retrievers and tools, ensuring that the agent can efficiently call various resources to answer user questions. The agent configuration supports flexible settings, including AI model, streaming output, retriever, rearranger, user message template, RAG query message template, chat memory, tool selection and other settings to meet the needs of different financial business scenarios, including: AI Model: Select the AI ​​model associated with the agent, such as the GPT series models for natural language understanding and generation.

[0128] Streaming output: Configure whether to enable streaming output, which supports returning the results of user interactions in real time.

[0129] Retriever: Configure the retriever used by the agent, support multiple retrieval methods (such as sparse / dense mixed retrieval, Milvus retriever, BM25 retriever, etc.), and set related parameters, including retrieval vector field information, return fields, similarity calculation, sorting method, etc.

[0130] Rearranger: Configure the rearranger used by the agent, which supports custom rearrangement logic to optimize the accuracy of query results.

[0131] User message template: defines the agent's reply format and prompt words, which can be customized or use the system default template.

[0132] RAG query message template: Configure the message template for RAG (retrieval enhancement generation) query according to the business scenario, and support customization or use of system default prompt words.

[0133] Chat storage: Configure the chat storage method of the agent, such as memory, MongoDB, Redis, etc., to determine the storage and management method of chat records.

[0134] Tool selection: Select third-party API tools for the intelligent agent, such as financial system interface, budget query interface, etc., to enhance its ability to handle financial business.

[0135] To ensure that the agent can continue to provide high-quality services, the agent evaluation function is used to evaluate the performance of the agent, evaluate the accuracy of the agent in answering financial business questions, and ensure that the information it provides is true and reliable. It supports both automatic evaluation and manual evaluation. Automatic evaluation is performed through preset evaluation quantities and standards to automatically detect the performance of the agent; manual evaluation allows manual selection of question content and scoring based on actual answers. The evaluation results can be quantified through standardized scores (such as standard%, score range), helping to optimize the performance of the agent.

[0136] With the continuous changes in financial operations and technological advances, agents need to be updated and optimized regularly. The agent maintenance module provides the following functions: Model Update: Supports automatic or manual updates of the AI ​​models used by the agent to ensure that it always has the latest functions and knowledge base.

[0137] Knowledge base update: Combined with the knowledge base management module, ensure that the agent can obtain the latest financial policies, regulations and operating manuals.

[0138] Restart and repair: When the agent fails or its performance degrades, the module supports automatic restart or repair to ensure the continuous availability of the agent.

[0139] Version control: Record the change history of each agent version to facilitate traceability and management, and ensure that the evolution process of the agent is controllable.

[0140] Agent monitoring and log management In order to promptly detect anomalies and problems in the operation of agents, the agent maintenance module provides comprehensive monitoring and log recording functions: Operation monitoring: Real-time monitoring of the running status of the intelligent agent, including response time, call success rate, failure rate, etc., to ensure efficient and stable operation of the intelligent agent.

[0141] Log management: Records the agent's request log and response log to help administrators quickly locate problems and optimize the agent's response effect.

[0142] Performance analysis: Through continuous analysis of the agent's performance, optimization suggestions are provided to further improve the agent's work efficiency and service quality.

[0143] In order to explain in detail the technical solution and effect of the customer service response method based on business agents in this application, a specific application example will be used to describe it. In actual applications, by defining AI model management, vector model management, rearrangement model management, retriever management, tool management, knowledge base, etc., vertical business agents can be quickly generated and created to achieve the use of different agents for different business systems and support the simultaneous application of multiple agents. The process of the specific user interaction method is as follows: Figure 3As shown, the following steps are included: S1. User submission prompt: User login and conversation initiation: After the user logs in to the system through the client, he can interact with the customer service component. The user selects a specific agent to initiate a conversation. This agent is usually related to a specific business field (such as finance, medical care, customer service, etc.).

[0144] Browser request: The user enters a prompt (user's request or question) in the dialog box. The browser transmits data containing the agent ID and prompt content to the backend server through an HTTP request. This data transmission uses a standard Web request protocol (such as HTTP / HTTPS) to ensure data reliability and security.

[0145] S2. The server queries the agent configuration information: Get configuration based on agent ID: After receiving the request, the server first queries the agent's configuration information from the database or configuration center based on the agent ID passed in. Configuration items include AI model configuration, retriever configuration, rearrangement model configuration, tool information, etc.

[0146] Context information: At the same time, the server also needs to query the context information related to the current user (such as chat history) through the user ID. This information is important for the subsequent dialogue generation. The loading of context information can be done through database query or cache system (such as Redis) to optimize the response time.

[0147] S3. Retriever configuration and Prompt vectorization: Retriever configuration loading: Based on the agent configuration, the server loads the associated retriever configuration. An agent may have multiple retriever configurations (1-N), which are used in different business scenarios and are responsible for converting prompt text into vector representation.

[0148] Vectorization process: For each retriever, the system uses a predefined vectorization model (such as BERT, Sentence-BERT, etc.) to vectorize the prompt input by the user. At this point, the vectorization process converts the text into a digital form (vector), so that subsequent computers can perform vector comparisons (such as similarity calculations). This step is a key step in natural language processing (NLP) and determines the accuracy of subsequent processing.

[0149] S4. Vector similarity retrieval: Vector library search: vectorized text data will be sent to the pre-built vertical business vector library for similarity search. Each searcher is configured with one or more vector libraries in business fields. The user's vector will be compared with the vectors in the library to find similar or related content.

[0150] Similarity calculation: In this step, the server uses the approximate nearest neighbor (ANN) algorithm (for example, using libraries such as FAISS and HNSW) to perform similarity search. The system compares the vectors according to the configured search fields, such as text, tags, etc., and returns the most relevant 10*N data. This process is usually calculated in parallel to improve efficiency.

[0151] S5. Semantic rearrangement: Reranking model loading: Through the agent configuration, the server loads the reranking model. The model sorts the retrieved vectors according to the semantic relevance of the query results, ensuring that the most relevant information is ranked first. Reranking is usually based on more complex deep learning models, such as BERT, RoBERTa, etc.

[0152] Final sorting: After the adjustment of the rearrangement model, 10*N vector data with the most semantics are finally returned. The rearrangement not only considers the similarity of the vectors, but also optimizes the sorting based on the specific semantics of the business requirements.

[0153] S6.Message assembly: Assembly process: Based on the configuration of the agent, the system will combine the following information: 1) the user's historical conversation records; 2) information and descriptions of related tools; 3) system prompts (such as help information, guides, etc.); 4) the final vector data and the user's original prompt. With this information, the server generates a complete request message and prepares to pass it to the agent's AI model.

[0154] Template and dynamic splicing: In order to efficiently assemble messages, the system usually adopts a template approach, defines the message structure in advance, and then dynamically fills it according to real-time data. This can effectively reduce repeated calculations and improve processing efficiency.

[0155] S7. The agent generates a response: Message passing to AI model: The assembled message is passed to the AI ​​model in the agent for processing. The agent usually uses a large language model (LLM) to generate responses to user prompts. This process includes text generation tasks, and the AI ​​model needs to understand user needs based on the input prompt and context information and generate appropriate responses.

[0156] Tool call: When generating a response, if the agent triggers certain specific tools (such as database queries, API calls, etc.), it will call the relevant tools or services, perform necessary operations, and return relevant data or results. This allows the system to not only generate dialogues, but also integrate other services to provide more complex functions.

[0157] S8. Result returned: Generated results are returned to the client: The responses generated by the agent (and the results of tool execution) are returned to the client. The server will transmit the results back to the browser via HTTP response, and the client browser will render and display them according to the specified message format (such as JSON, XML, etc.), and the user will see the final response results.

[0158] See also Figure 4 The second aspect of the present invention provides a customer service response device based on a business intelligence agent, and the customer service response device based on a business intelligence agent comprises: A dialogue request acquisition module 10 is used to acquire a dialogue request sent by a user, wherein the dialogue request carries an agent selected by the user, and the agent is related to a business domain; A configuration information query module 20, for querying the configuration information of the agent according to the agent's identifier, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; The question input acquisition module 30 is used to acquire the question input by the user and retrieve multiple pieces of relevant data corresponding to the question through a configured retriever; The optimization sorting module 40 is used to optimize the sorting of the multiple related data according to the semantic relevance of the multiple related data to the question through the configured rearrangement model, so as to obtain the related data that meets the semantics of the question; The reply generation module 50 is used to generate a reply based on the relevant data and questions that conform to the semantics of the question through the configured AI model, and return the generated reply to the client.

[0159] In an optional implementation of the second aspect of the present invention, the question input acquisition module 30 is also used to load a configured retriever according to the configuration information of the intelligent agent; obtain the question input by the user, and convert the text corresponding to the question into a vector representation through a preset vectorization model; and retrieve multiple related data corresponding to the question through the configured retriever based on the vectorized question text.

[0160] In an optional implementation of the second aspect of the present invention, the question input acquisition module 30 is also used to input the vectorized question text into a preset vertical business vector library for similarity retrieval according to the configured retriever; through the approximate nearest neighbor algorithm, the vectorized question text is compared with the vector data in the vertical business vector library to retrieve multiple related data corresponding to the question.

[0161] In an optional implementation of the second aspect of the present invention, the reply generation module 50 is also used to obtain the user's historical conversation records, combine relevant data, questions, tool information, and historical conversation records that conform to the question semantics, and generate combined information; pass the combined information to the configured AI model, and generate a reply through the AI ​​model; return the generated reply to the client, so that the client renders and displays the reply according to a preset message format.

[0162] In an optional implementation of the second aspect of the present invention, the reply generation module 50 is also used to obtain a user identifier, and based on the user identifier, load user-related context information through a database or cache query; pass the context information and combined information to a configured AI model; and generate a reply through the AI ​​model.

[0163] In an optional implementation of the second aspect of the present invention, the customer service reply device based on the business agent further includes: The knowledge base construction module is used to obtain diversified data and build a knowledge base based on the diversified data. The diversified data is obtained through at least one of file upload, Web site synchronization and API access; the content of the knowledge base is updated through the API and website synchronization mechanism; the historical version records of the knowledge base are recorded to roll back the updated or erroneous content of the knowledge base.

[0164] In an optional implementation of the second aspect of the present invention, the customer service reply device based on the business agent further includes: The intelligent agent evaluation module is used to evaluate the intelligent agent through a preset evaluation quantity and standardized score; and optimize the intelligent agent based on the evaluation results.

[0165] Figure 51 is a schematic diagram of the structure of a customer service response device based on a business intelligence provided by an embodiment of the present invention. The customer service response device 500 based on a business intelligence may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 (for example, one or more mass storage devices) storing application programs 533 or data 532. Among them, the memory 520 and the storage medium 530 may be short-term storage or permanent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the customer service response device 500 based on the business intelligence. Furthermore, the processor 510 may be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the customer service response device 500 based on the business intelligence.

[0166] The business intelligence-based customer service response device 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and / or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, Linux, Free BSD, etc. Those skilled in the art will appreciate that Figure 5 The structure of the business agent-based customer service response device shown does not constitute a limitation on the business agent-based customer service response device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0167] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the customer service response method based on business intelligence.

[0168] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer 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 foregoing.

[0169] In addition, although each operation is described in a specific order, this should be understood as requiring such operation to be performed in the specific order shown or in a sequential order, or requiring that all illustrated operations should be performed to obtain desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination mode.

[0170] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A customer service response method based on business intelligence, characterized in that: The customer service response method based on business intelligence agent includes: Acquire a dialog request sent by a user, wherein the dialog request carries an agent selected by the user, and the agent is related to a business field; According to the identification of the agent, query the configuration information of the agent, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; Obtaining a question input by a user, and retrieving multiple pieces of relevant data corresponding to the question through a configured retriever; By using the configured rearrangement model, the plurality of relevant data are optimized and sorted according to the semantic relevance between the plurality of relevant data and the question, so as to obtain relevant data that conforms to the semantics of the question; Based on the relevant data that conforms to the semantics of the question and the question, a response is generated through the configured AI model, and the generated response is returned to the client.

2. The customer service response method based on business intelligence according to claim 1 is characterized in that: The step of obtaining a question input by a user and retrieving a plurality of relevant data corresponding to the question through a configured retriever includes: According to the configuration information of the agent, loading the configured retriever; Obtain the question input by the user, and convert the text corresponding to the question into a vector representation through a preset vectorization model; According to the vectorized question text, multiple pieces of relevant data corresponding to the question are retrieved through the configured retriever.

3. The customer service response method based on business intelligence according to claim 2 is characterized in that: The retrieving a plurality of pieces of relevant data corresponding to the question through the configured retriever according to the vectorized question text includes: According to the configured retriever, the vectorized question text is input into a preset vertical business vector library for similarity retrieval; The vectorized question text is compared with the vector data in the vertical business vector library by using an approximate nearest neighbor algorithm, and multiple pieces of relevant data corresponding to the question are retrieved.

4. The customer service response method based on business intelligence according to claim 1 is characterized in that: The configuration information also includes tool information, which is used to configure a tool interface to provide an external interaction function for the agent; The step of generating a reply based on the relevant data that conforms to the semantics of the question and the question through a configured AI model, and returning the generated reply to the client comprises: Acquire the user's historical conversation record, combine the relevant data that conforms to the question semantics, the question, the tool information, and the historical conversation record to generate combined information; Passing the combined information to the configured AI model, and generating a response through the AI ​​model; The generated reply is returned to the client, so that the client renders and displays the reply according to a preset message format.

5. The customer service response method based on business intelligence according to claim 4 is characterized in that: The transmitting the combined information to the configured AI model and generating a response through the AI ​​model includes: Obtaining a user ID, and based on the user ID, loading user-related context information through a database or cache query; Passing the context information and the combined information to a configured AI model; A response is generated through the AI ​​model.

6. The customer service response method based on business intelligence according to claim 1 is characterized in that: Before obtaining the dialog request sent by the user, the method further includes: Acquire diversified data and construct a knowledge base based on the diversified data, wherein the diversified data is acquired by at least one of file uploading, website synchronization and API access; Updating the content of the knowledge base through API and website synchronization mechanism; The historical version records of the knowledge base are recorded to roll back the updated or misoperated contents of the knowledge base.

7. The customer service response method based on business agent according to claim 1, characterized in that: After generating a reply based on the relevant data that conforms to the semantics of the question and the question through the configured AI model and returning the generated reply to the client, it also includes: Evaluating the agent using a preset number of evaluations and a standardized score; According to the evaluation results, the agent is optimized.

8. A customer service response device based on business intelligence, characterized in that: The customer service reply device based on business intelligence comprises: A dialogue request acquisition module is used to acquire a dialogue request sent by a user, wherein the dialogue request carries an agent selected by the user, and the agent is related to a business field; A configuration information query module, used to query the configuration information of the agent according to the identifier of the agent, wherein the configuration information includes AI model configuration, retriever configuration, and rearrangement model configuration; A question input acquisition module is used to obtain a question input by a user and retrieve multiple pieces of relevant data corresponding to the question through a configured retriever; An optimization sorting module, used to optimize and sort the multiple pieces of relevant data according to the semantic relevance between the multiple pieces of relevant data and the question through a configured rearrangement model, so as to obtain relevant data that meets the semantics of the question; The reply generation module is used to generate a reply based on the relevant data that conforms to the semantics of the question and the question through the configured AI model, and return the generated reply to the client.

9. A customer service response device based on business intelligence, characterized in that: The customer service reply device based on business intelligence comprises: a memory and at least one processor, the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instructions in the memory to enable the business intelligence-based customer service reply device to execute the business intelligence-based customer service reply method according to any one of claims 1 to 7.

10. 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 response method based on business intelligence as described in any one of claims 1 to 7 is implemented.

Images