An intelligent customer service assistance system based on browser plugins and a large language model

Abstract

This invention relates to the field of artificial intelligence technology and proposes an intelligent customer service assistance system based on browser plugins and a large language model. The system includes an intelligent plugin front-end, a large model service layer, and a multi-source data integration layer. The intelligent plugin front-end is deployed in the customer service browser and monitors the customer service system page to capture the current working context. The multi-source data integration layer retrieves supplementary business data related to the current working context from multiple business systems corresponding to the customer service system. The large model service layer generates content based on the current working context and supplementary business data using a large language model to obtain auxiliary processing information. The intelligent plugin front-end also transforms the auxiliary processing information into visual and operational elements that are presented and interacted with on the customer service system page. This system significantly reduces the workload of customer service staff manually switching, querying, and copying between different systems, greatly improving the efficiency and accuracy of problem handling.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent customer service assistance system based on browser plugins and a large language model. Background Technology

[0002] As enterprises accelerate their digital transformation, customer service operations face common challenges such as fragmented systems, data silos, and insufficient intelligence. To handle a single customer request, customer service personnel often need to repeatedly switch between multiple independent systems, such as ticketing systems, customer relationship management systems, and knowledge bases, to query and copy information. This results in cumbersome procedures, delayed responses, and a high risk of service accuracy and customer experience being affected by missing or outdated information.

[0003] To address these issues, two main integration solutions currently exist: First, data integration via developed system interfaces. This solution requires complex joint debugging of multiple systems, resulting in long implementation cycles, high transformation costs, and difficult maintenance. Second, reconstructing independent customer service workbenches by integrating the functional modules of various systems. While this unifies the interface, it necessitates the duplication of existing functions, leading to resource waste and difficulty in adapting to rapid business iterations. Both solutions require deep modifications to existing systems and have significant limitations in scalability, flexibility, and implementation efficiency.

[0004] Therefore, how to achieve intelligent integration of multi-source data, assistance to business processes, and efficient empowerment of artificial intelligence without altering the existing system architecture as much as possible has become an urgent technical requirement for improving customer service efficiency and service quality. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent customer service assistance system based on browser plugins and a large language model, so as to solve the above-mentioned technical problem.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: An intelligent customer service assistance system based on browser plugins and a large language model, comprising an intelligent plugin front-end, a large model service layer, and a multi-source data integration layer; the intelligent plugin front-end is deployed in the customer service browser and is used to monitor the customer service system page to capture the current working context; the multi-source data integration layer is used to obtain supplementary business data associated with the current working context from multiple business systems corresponding to the customer service system; the large model service layer is used to generate content through a large language model based on the current working context and the supplementary business data to obtain auxiliary processing information; the intelligent plugin front-end is also used to transform the auxiliary processing information into visual and operational elements that are presented and interacted with on the customer service system page.

[0007] The beneficial effects of this invention are as follows: It non-intrusive monitoring of customer service work pages is achieved through a smart plugin front-end deployed in the browser. Combined with a multi-source data integration layer, it acquires cross-system business data in real time. Finally, a large model service layer is used for deep semantic understanding and content generation, thus realizing intelligent assistance for customer service personnel. This system significantly reduces the burden on customer service staff by manually switching, querying, and copying data between different systems, greatly improving the efficiency and accuracy of problem handling, while also reducing training costs and reliance on staff experience.

[0008] Based on the above technical solution, the present invention can be further improved as follows.

[0009] Furthermore, the smart plugin front end includes: A context-aware unit is used to monitor changes in the customer service system page in order to capture the current working context; The intelligent triggering unit is used to decide, based on the auxiliary processing information, which business information and functions need to be pushed to the customer service system page. An intelligent display unit is used to display the business information and business functions on the customer service system page; The natural language interaction unit is used to receive natural language instructions input by customer service personnel and transmit the natural language instructions to the large model service layer.

[0010] Furthermore, the context-aware unit is used to monitor changes in the customer service system page to capture the current working context, specifically for: The changes in DOM nodes of the customer service system page are monitored through the DOM monitoring interface provided by the client browser. When a change is detected in the DOM node of the customer service system page, structured work order data and customer information are extracted from the changed DOM node to obtain the current working context.

[0011] Furthermore, the auxiliary processing information includes at least one of the following: response content script, processing step suggestions, and work order summary information.

[0012] Furthermore, the multi-source data integration layer includes: The data access module is used to obtain multiple related data associated with the current working context from multiple business systems corresponding to the customer service system; The data aggregation module is used to integrate and process multiple related data to obtain supplementary business data; The knowledge base module is used to convert the supplementary business data into semantic vectors and store them in a vector database.

[0013] Furthermore, the large model service layer includes: A multi-task reasoning engine is used to perform reasoning analysis based on the current working context using a large language model and generate reasoning analysis results. The knowledge-enhanced retrieval module is used to generate retrieval queries based on the reasoning and analysis results, and to perform similarity searches in the vector database based on the retrieval queries to obtain relevant business rules and knowledge fragments. The content generation module is used to generate content based on the reasoning and analysis results and related business rules and knowledge fragments to obtain auxiliary processing information.

[0014] Furthermore, the large model service layer also includes: The continuous learning and optimization module is used to collect feedback operation data from customer service personnel and adjust the prompt word template of the multi-task inference engine, the retrieval strategy of the knowledge-enhanced retrieval module, or the generation rules of the content generation module based on the feedback operation data.

[0015] Furthermore, the system also includes: The intelligent workflow engine is used to generate a workflow that includes business operation steps, execution roles, and verification conditions based on the auxiliary processing information and in combination with pre-configured process decision rules and role permissions.

[0016] Furthermore, the system also includes: The security control module is used to encrypt the communication link between the front end of the smart plugin and the large model service layer, and to verify the operation permissions of customer service personnel.

[0017] Furthermore, when the data access module is used to obtain multiple related data associated with the current working context from multiple business systems corresponding to the customer service system, it is specifically used for: Based on the current working context, construct the query conditions; Based on the query conditions, multiple data interfaces of the business systems are called concurrently through a unified application programming interface gateway to obtain multiple related data. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an intelligent customer service assistance system based on browser plugins and a large language model according to the present invention. Figure 2 This is an architecture diagram of an intelligent customer service assistance system based on browser plugins and a large language model, according to the present invention. Detailed Implementation

[0019] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0020] like Figure 1 As shown, this embodiment provides an intelligent customer service assistance system based on browser plugins and a large language model, including an intelligent plugin front-end, a large model service layer, and a multi-source data integration layer. The intelligent plugin front-end is deployed in the customer service browser and is used to monitor the customer service system page to capture the current working context. The multi-source data integration layer is used to obtain supplementary business data related to the current working context from multiple business systems corresponding to the customer service system. The large model service layer is used to generate content based on the current working context and supplementary business data through a large language model to obtain auxiliary processing information. The intelligent plugin front-end is also used to transform the auxiliary processing information into visual and operational elements that are presented and interacted with on the customer service system page.

[0021] This system mainly consists of an intelligent plugin front-end, a large model service layer, and a multi-source data integration layer. The intelligent plugin front-end, based on browser extension technology, is responsible for monitoring user operation behavior and injecting functional entry points, thereby integrating the interactive capabilities of the large language model and displaying processing results. The large model service layer provides intelligent analysis, content generation, and decision support based on the large language model. The multi-source data integration layer integrates data from various business systems through a unified data gateway.

[0022] By deploying a smart plugin frontend in the browser to non-intrusively monitor customer service work pages, and combining this with a multi-source data integration layer to acquire cross-system business data in real time, and finally utilizing a large model service layer for deep semantic understanding and content generation, intelligent assistance for customer service personnel is achieved. This system significantly reduces the workload for customer service personnel by manually switching, querying, and copying data between different systems, greatly improving the efficiency and accuracy of problem handling, while also reducing training costs and reliance on personnel experience.

[0023] Optionally, in this embodiment, the smart plugin front end includes: Context-aware units are used to monitor changes to customer service system pages in order to capture the current working context; The intelligent triggering unit is used to decide, based on auxiliary processing information, which business information and functions need to be pushed to the customer service system page. Intelligent display units are used to display business information and business functions on the customer service system page; The Natural Language Interaction Unit is used to receive natural language commands input by customer service personnel and transmit the natural language commands to the large model service layer.

[0024] Optionally, in an embodiment, the context-aware unit is used to monitor changes in the customer service system page to capture the current working context. Specifically, it is used to: listen for changes in the DOM nodes of the customer service system page through the DOM monitoring interface provided by the client browser; when a change in the DOM nodes of the customer service system page is detected, extract structured work order data and customer information from the changed DOM nodes to obtain the current working context.

[0025] Specifically, the context-aware unit captures the work context in real time through DOM monitoring and user behavior analysis. It employs a combination of the MutationObserver API (the browser's native DOM monitoring interface) and element listening: during plugin initialization, the MutationObserver is bound to the root node of the existing customer service system page (or the core container of the work order system), and monitoring configurations are set (child node changes: childList; attribute changes: attributes; content changes: characterData; and subtree listening is enabled). When a user clicks to trigger work order loading, field editing, or page switching, a monitoring callback is triggered, extracting the text / attribute values ​​of the target element (e.g., extracting the customer's phone number from input#phone, and the complaint issue from textarea#problem-desc), thus obtaining the current work context.

[0026] The intelligent triggering unit analyzes the contextual semantics based on a large language model, intelligently displays relevant business information, and recommends relevant business functions.

[0027] Relevant business information is auxiliary data provided by the large model based on the analysis results to help customer service handle current business. It is key data that customer service can directly obtain without manually querying across systems, and mainly includes the following categories (customer service scenarios): Customer-related information includes: historical complaint records (e.g., the customer complained about the same issue 3 days ago), package information (e.g., the customer is currently subscribed to the 99 yuan unlimited data package), and customer level (e.g., high-value VIP customer). Knowledge base / rule information: standard answers matching the current question (such as the conditions and procedures for refunds of value-added service fees), enterprise business rules (such as refunds of less than 15 yuan can be directly approved without the need for confirmation from superiors); Historical case information: Past cases of handling similar issues (e.g., "a refund was issued for the same value-added service fee issue, and the solution was to directly return the money to the original payment method"). Work order association information: upstream transfer records of the work order (such as whether it was transferred by other departments) and associated business order number (such as the order number for value-added services).

[0028] The natural language interaction unit allows users to interact with plugins using natural language and directly express their needs; the intelligent display unit dynamically generates the most suitable display format based on the analysis results of the large language model.

[0029] Optionally, in an embodiment, the large model service layer includes: A multi-task reasoning engine is used to perform reasoning analysis based on the current working context using a large language model and generate reasoning analysis results. The knowledge-enhanced retrieval module is used to generate retrieval queries based on the reasoning and analysis results, and to perform similarity searches in the vector database based on the retrieval queries to obtain relevant business rules and knowledge fragments. The content generation module is used to generate content based on the reasoning and analysis results and related business rules and knowledge fragments to obtain auxiliary processing information.

[0030] The multi-task inference engine is based on a large language model to perform multi-task processing such as work order classification, sentiment analysis, and intent recognition. The input data comes from the complete working context captured by the context-aware unit, such as basic work order information (number, type, customer mobile number, customer level, work order status, creation time), customer tags (VIP, high complaint rate), customer service operation behavior (such as "open work order" and "click reply box"), business parameters (disputed fee amount, fault address), and detailed description of the work order problem.

[0031] The output is a structured, confidence-based multi-task reasoning result (rather than plain text), which can directly drive subsequent intelligent triggering, content generation, and other processes, such as work order classification, sentiment analysis, intent recognition, and strategy triggering.

[0032] Specifically, the structured information captured by the context-aware unit (such as work order data, customer information, customer service operation behavior, complaint issues, and processing information) is input into the large language model. After the large model performs reasoning analysis, it outputs structured semantic analysis conclusions that can directly drive triggering logic, rather than simple text content. For example: 1. Problem Scenario / Business Type Identification: Determine the current business scenario of the customer service representative, such as "handling phone bill disputes and complaints", "consultation on cancellation of value-added services", "reporting broadband faults", etc.

[0033] 2. Extracting core customer needs / intents: Analyze the customer's core needs (from the work order problem description / processing record), such as "the customer requests a refund of the overcharged 10 yuan value-added service fee" or "the customer needs on-site broadband repair".

[0034] 3. Key information annotation: Extract core business parameters from the context, such as customer mobile phone number, work order number, disputed fee amount, fault address, and work order urgency level.

[0035] 4. Trigger Priority / Strategy: Determine the information / functions to be pushed based on the information returned by the large model, as well as the push format (such as pop-up, sidebar display, one-click access), such as "displaying the refund process with standard answers from the knowledge base".

[0036] The content generation module automatically generates response content, processing suggestions, and summaries based on business scenarios and user needs. Business scenarios refer to issues such as billing disputes, broadband fault reports, inquiries about canceling value-added services, package subscriptions, and escalation of customer complaints. Further supplementary scenarios include customer level (VIP / regular), urgency level (high / medium / low), and processing stage (initial reception / follow-up processing / case closed).

[0037] Business scenarios are automatically obtained through three steps: "contextual data collection + large-scale model semantic analysis + rule matching". The specific process is as follows: 1. Collect basic data: The context-aware unit captures basic data such as work order problem descriptions, customer information, and operational behaviors.

[0038] 2. Large-scale model semantic analysis: The multi-task reasoning engine performs reasoning on the basic data and outputs results such as work order classification and intent recognition (e.g., "Work order classification: broadband fault reporting; intent: emergency on-site repair"). This is the core basis for business scenarios.

[0039] 3. Rule matching and scenario integration: Combine the enterprise's pre-set business scenario system, match the reasoning results with structured data (customer level, urgency, processing stage) and integrate them into complete business scenario tags.

[0040] The knowledge-enhanced retrieval module combines knowledge-enhanced generation (RAG) technology with enterprise knowledge base and large language model capabilities.

[0041] Combining RAG technology can achieve the following technical advantages: 1) Make up for the shortcomings of large models that lack enterprise proprietary knowledge: Large models only contain general public knowledge, lack internal business rules of enterprises, and without combining with a knowledge base, they are prone to outputting content that does not meet the requirements of enterprises; 2) Solving knowledge illusions in large models: Large models may output false / erroneous information (such as mistakenly saying "customer service refunds take 3 days to arrive" when the actual company rule is 24 hours). Combining with a knowledge base can constrain large models with real business knowledge and avoid illusions. 3) Solving the knowledge update problem: Enterprise knowledge bases need to iterate with business. If large models are fine-tuned and knowledge is updated, a lot of data and computing power are required, which is costly and time-consuming. RAG can directly update the knowledge in the vector database and synchronize business changes in real time without the need to fine-tune large models.

[0042] Knowledge-enhanced retrieval serves as the system's knowledge hub, connecting the context-aware and multi-task reasoning modules at the beginning and supporting the content generation and intelligent triggering modules at the end. 1. Relationship with preceding technologies (receiving input, supporting retrieval accuracy) Connecting to the context-aware unit: The work context captured by context awareness (work order issues, customer information, operational behavior) is the core basis for RAG to generate search terms. Without this data, search terms will be out of context (e.g., blindly searching for "refund rules" instead of "value-added service 15 yuan refund rules"), ensuring the relevance of the search; Connecting to a multi-task inference engine: The work order classification, intent recognition, and scene tags (such as "fee dispute complaint - VIP customer") output by multi-task inference can optimize the search term dimensions, further narrow the search scope, and improve the accuracy of knowledge fragment matching (such as prioritizing the search for VIP customer-specific refund knowledge, rather than general rules).

[0043] 2. Connection with subsequent technologies (outputting knowledge, enabling intelligent output and triggering) Supporting the content generation module: The enterprise knowledge fragments retrieved by RAG serve as the core business basis for content generation. Based on this knowledge, the content generation module must generate responses, processing suggestions, and work order summaries that comply with enterprise rules (e.g., generating compliant responses based on refund rules, rather than generic wording), ensuring accurate and compliant output. Supporting the intelligent triggering unit: Relevant business knowledge retrieved by RAG (such as business processes and standard rules) is the core source of business information pushed by the intelligent triggering unit. For example, after retrieving "value-added service refund process", the intelligent triggering unit will display the process as relevant business information in the sidebar to customer service, and recommend the "one-click submission of refund application" function to support intelligent assistance in implementation.

[0044] Optionally, in an embodiment, the auxiliary processing information includes at least one of the following: response content script, processing step suggestions, and work order summary information.

[0045] Optionally, in an embodiment, the large model service layer further includes: The continuous learning and optimization module is used to collect feedback operation data from customer service personnel and adjust the prompt word template of the multi-task inference engine, the retrieval strategy of the knowledge-enhanced retrieval module, or the generation rules of the content generation module based on the feedback operation data.

[0046] The continuous learning and optimization module continuously optimizes the output of the large language model based on user feedback and business data.

[0047] Specifically, large-scale model prompt optimization: For frequently reported issues, adjust the model's input prompt template. For example, if multiple reports indicate that "AI-generated replies missed mentioning VIP rules," add the following to the prompt: "Business rules corresponding to the customer level must be cited first; VIP customers must be explicitly marked 'priority processing'." RAG Knowledge Base Weight Adjustment: If feedback indicates that "the retrieved knowledge does not match the scenario" (e.g., the refund scenario matches the fault reporting rule), the retrieval weight of the vector database will be adjusted, and the matching weight of fields such as "customer level" and "business type" will be increased. At the same time, frequently missing knowledge will be supplemented (e.g., "VIP refund time limit" which is frequently supplemented by customer service will be added to the knowledge base). Rule engine optimization: To address "inaccurate dispatch recommendations," the rule weights of the dispatch decision model are adjusted based on feedback data. For example, if historical feedback indicates "90% accuracy rate for dispatching broadband outages to the repair department and 30% accuracy rate for dispatching to the customer service department," then the matching weight for "broadband outage → repair department" will be increased.

[0048] Optionally, in an embodiment, the multi-source data integration layer includes: The data access module is used to obtain multiple related data associated with the current working context from multiple business systems corresponding to the customer service system; The data aggregation module is used to integrate and process multiple related data to obtain supplementary business data; The knowledge base module is used to convert the supplementary business data into semantic vectors and store them in a vector database.

[0049] Optionally, in an embodiment, when the data access module is used to obtain multiple associated data related to the current working context from multiple business systems corresponding to the customer service system, it is specifically used to: construct query conditions based on the current working context; and based on the query conditions, concurrently call the data interfaces of multiple business systems through a unified application programming interface gateway to obtain multiple associated data.

[0050] Optionally, in an embodiment, the system further includes: The intelligent workflow engine is used to generate workflows that include business operation steps, execution roles, and verification conditions based on auxiliary processing information and pre-configured process decision rules and role permissions.

[0051] The intelligent workflow engine understands business needs through large language models, dynamically generates and optimizes workflows, and achieves true intelligent decision support. 1. Basic business context data: Work order core information (number, type, status, creation time), customer information (mobile phone number, level, package, historical interaction records), business parameters (dispute amount, fault address, urgency level tag); 2. Preceding AI analysis results: Scene recognition (e.g., "Broadband fault repair - urgent"), core intent ("Request on-site repair"), and sentiment analysis ("Angry - Strong") output by the multi-task inference engine; Enterprise knowledge base rules retrieved by RAG (e.g., "Emergency fault dispatch within 2 hours"); 3. Business knowledge or business process templates: fixed process specifications (such as "refund amount ≥ 100 yuan requires superior approval", "fault reporting requires verification of customer address first"), role and permission rules (such as "junior customer service cannot approve refunds ≥ 50 yuan"), department / personnel matching rules (such as "broadband fault → repair department").

[0052] Optionally, in an embodiment, the system further includes: The security control module is used to encrypt the communication link between the smart plugin front-end and the large model service layer, and to verify the operation permissions of customer service personnel.

[0053] Compared to existing solutions, this system has the following technical advantages: 1. Solving the problem of "high cost and long cycle of system integration and transformation": This solution uses a browser plugin as a non-intrusive carrier, requiring no modification to existing system code or interfaces. The browser plugin can directly capture page context data through DOM monitoring, and the unified data gateway of the multi-source data integration layer automatically adapts to different system interface protocols, eliminating the need for complex interface development and cross-departmental coordination between multiple systems, significantly reducing implementation costs, significantly shortening the launch cycle, and avoiding the high cost of traditional integration and transformation.

[0054] 2. Solving the problem of "repetitive development and resource waste in independent workbenches": The solution focuses on data integration and intelligent empowerment, avoiding the duplication of development of existing system functions. The multi-source data integration layer directly calls native data such as customer service knowledge base and work order system, and plugins are directly embedded into the existing customer service system. Through the embedding of plugin capabilities, the core capabilities of each system are integrated into a unified interface. Customer service representatives can call the original functions without switching systems, which not only makes full use of existing resources, but also completely avoids the resource waste caused by repeated development.

[0055] In summary, this invention achieves non-intrusive system integration and truly intelligent customer service assistance by deeply integrating the intelligent capabilities of a large language model into a browser plugin. The system utilizes the browser plugin to perform DOM monitoring and context awareness on the customer service work page, intelligently capturing the work scenario; through the large model service layer, it performs fusion analysis and semantic understanding of multi-source data, and incorporates enterprise-specific knowledge using RAG technology to generate auxiliary information including response content, processing suggestions, and summaries; based on context understanding and intelligent triggering mechanisms, it dynamically recommends business functions and information, and accurately displays them in various forms such as natural language interaction and intelligent workflows. This system significantly improves the accuracy and efficiency of customer service problem handling, reduces a significant amount of copywriting workload, lowers usage and training costs, and continuously optimizes through a continuous learning mechanism, driving the evolution of customer service business processes towards standardization and intelligence.

[0056] like Figure 2 As shown, the architecture of this system is divided into a front-end embedding layer (entry point) → a data service layer (data preparation) → a service layer (core intelligent processing) → a result display layer (output landing). Each module corresponds to a specific functional unit in the solution, as shown in Table 1: Table 1 Functional Unit Comparison Table Taking the customer service representative opening a refund request for value-added services (customer complaint about being overcharged by 15 yuan) as an example, the process is as follows: 1. Front-end Embedded Layer: Triggering Process + Initial Data Collection (The "Entry Point" of the Solution) Step 1: Start DOM monitoring and event capturing When a customer service representative opens the complaint ticket page, the browser plugin module uses the MutationObserver API to listen to the DOM and automatically capture the basic business data of the ticket (the "context-aware unit" in the corresponding solution): ticket ID=1001, customer mobile number 139xxxx7890, problem description "overcharged 15 yuan for value-added services, requesting a refund", ticket status "new"; at the same time, it captures the customer service representative's action: "open ticket details".

[0057] Step 2: Natural Language Interaction Interface The plugin embeds a "smart assistant" dialog box (corresponding to the "natural language interaction unit" in the solution) on the work order processing page, and automatically triggers the analysis and processing of the problem in the "service layer" based on the information obtained in step 1. Users can also manually enter "how should this work order be processed" for further chat interaction.

[0058] Step 3: Injecting Smart Functions By injecting intelligent capabilities from the service layer (such as "AI-generated replies" and "workflow recommendations") into the natural language interaction interface page, the relevant service capabilities can be automatically triggered when a work order is opened or triggered by clicking in the form of "buttons / floating panels". The subsequent automatic processing results can be interacted with, displayed, or operated on the current embedded page.

[0059] 2. Data Service Layer: Preparing multi-source data and knowledge support (the "data foundation" of the solution) The initial data collected by the front end triggers the data service layer to work, providing the service layer with "enterprise private data + knowledge": Step 1: External system data retrieval Retrieve relevant data from existing systems (customer service knowledge base, work order system, customer middle platform): customer level (VIP), customer's historical complaint records (none), and value-added service refund rules ("15 yuan and below can be directly approved").

[0060] Step 2: Data Aggregation Service Integration The retrieved multi-source data is cleaned, structured, and integrated into standardized data such as "Customer Information: VIP139xxxx7890; Business Rules: 15 RMB Refund Direct Approval".

[0061] Step 3: Store knowledge in a vector knowledge base The integrated "value-added service refund rules" are converted into semantic vectors and stored in the vector knowledge base (the enterprise knowledge vector base of the corresponding solution "Knowledge Enhancement Retrieval - RAG"), waiting for the service layer to call them.

[0062] 3. Service Layer: Core Intelligent Processing (the "brain" of the solution) Front-end data and knowledge from the data service layer are input into the service layer to complete the core logic of "analysis → retrieval → generation": Step 1: Multi-task inference engine analyzes the context Receive "work order data + operation behavior" from the front end, combine it with customer information from the data service layer, and complete multi-task reasoning: Work order category: "Fee dispute complaints - Value-added service refund"; Intent recognition: "The customer requests a refund of the 15 yuan value-added service fee"; Sentiment analysis: "Dissatisfaction (moderate)".

[0063] Step 2: Knowledge-Enhanced Retrieval - RAG Invokes Enterprise Knowledge The multi-task reasoning engine uses "value-added service refund + VIP + 15 yuan" as the search term and semantically matches relevant knowledge fragments from the vector knowledge base: "VIP customers can directly approve the 15 yuan refund for value-added services, which will be returned to the original payment method within 24 hours and should be marked as 'priority processing'."

[0064] Step 3: Content generation service output results Combining "multi-task reasoning results + RAG retrieval knowledge", the following is generated: Reply: "Dear VIP customer, we have verified that you were overcharged 15 yuan for value-added services. According to the rules, it will be approved directly and refunded to the original payment method within 24 hours. We will prioritize processing your case." Suggested steps: "1. Click 'Submit Refund Request'; 2. Simultaneously send an SMS to inform the customer of the progress."

[0065] Step 4: Continuously learn and optimize, and wait for feedback. Record the current "reasoning result + generated content" and wait for subsequent customer service feedback / business data for future optimization (such as triggering adjustments if customer service modifies the reply).

[0066] 4. Results Presentation Layer: Delivering Intelligent Results (The "Delivery End" of the Solution) The results generated by the service layer are then displayed as actionable functions directly on the work order page through the results presentation layer. Step 1: Intelligent Workflow Generation Based on the "processing flow" of the content generation service, a personalized workflow is generated: "Submit a refund request → Send a text message to the customer → Mark the work order as 'processing'".

[0067] Step 2: Personalized interface adaptation and display Based on the customer service role (junior customer service), hide the "approval authority" related functions and only display the "submit refund application" button.

[0068] Step 3: Autofill and Generation The AI-generated reply content can be directly filled into the reply input box of the work order, eliminating the need for customer service representatives to manually enter it.

[0069] Step 4: Display of Multimodal Results The work order page displays the following in a "floating panel": reply content (text) + workflow (step bar) + business rules (knowledge source). Customer service representatives can directly click "Apply Reply" and "Execute Workflow".

[0070] 5. Continuous learning and optimization: closed-loop iteration (the solution's "self-evolution") Once the customer service representative completes the work order processing, the "continuous learning and optimization" module in the service layer is activated: If the customer service representative directly clicks "Apply to Reply" (positive feedback), then "Reply acceptance rate for this scenario +1" will be recorded. If customer service changes the reply to "refund within 12 hours" (negative feedback), then mark it as "reply missing VIP time limit rules" and optimize the prompt words later (add "VIP customer refund time limit 12 hours"); Feedback data is regularly used for "fine-tuning the multi-task inference engine" and "adjusting the weights of the vector knowledge base" to continuously improve the intelligent effect.

[0071] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0072] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0073] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0074] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent customer service assistance system based on browser plugins and a large language model, characterized in that, This includes an intelligent plugin front-end, a large model service layer, and a multi-source data integration layer; The intelligent plugin front-end is deployed in the client browser and is used to monitor the client system page to capture the current working context. The multi-source data integration layer is used to obtain supplementary business data associated with the current working context from multiple business systems corresponding to the customer service system. The large model service layer is used to generate content through the large language model based on the current working context and the supplementary business data to obtain auxiliary processing information. The intelligent plugin front end is also used to transform the auxiliary processing information into visual and operational elements that are presented and interacted with on the customer service system page.

2. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 1, characterized in that, The smart plugin front end includes: A context-aware unit is used to monitor changes in the customer service system page in order to capture the current working context; The intelligent triggering unit is used to decide, based on the auxiliary processing information, which business information and functions need to be pushed to the customer service system page. An intelligent display unit is used to display the business information and business functions on the customer service system page; The natural language interaction unit is used to receive natural language instructions input by customer service personnel and transmit the natural language instructions to the large model service layer.

3. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 2, characterized in that, The context-aware unit is used to monitor changes in the customer service system page to capture the current working context, specifically for: The changes in DOM nodes of the customer service system page are monitored through the DOM monitoring interface provided by the client browser. When a change is detected in the DOM node of the customer service system page, structured work order data and customer information are extracted from the changed DOM node to obtain the current working context.

4. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 1, characterized in that, The auxiliary processing information includes at least one of the following: response content script, processing step suggestions, and work order summary information.

5. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 1, characterized in that, The multi-source data integration layer includes: The data access module is used to obtain multiple related data associated with the current working context from multiple business systems corresponding to the customer service system; The data aggregation module is used to integrate and process multiple related data to obtain supplementary business data; The knowledge base module is used to convert the supplementary business data into semantic vectors and store them in a vector database.

6. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 5, characterized in that, The large model service layer includes: A multi-task reasoning engine is used to perform reasoning analysis based on the current working context using a large language model and generate reasoning analysis results. The knowledge-enhanced retrieval module is used to generate retrieval queries based on the reasoning and analysis results, and to perform similarity searches in the vector database based on the retrieval queries to obtain relevant business rules and knowledge fragments. The content generation module is used to generate content based on the reasoning and analysis results and related business rules and knowledge fragments to obtain auxiliary processing information.

7. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 6, characterized in that, The large model service layer also includes: The continuous learning and optimization module is used to collect feedback operation data from customer service personnel and adjust the prompt word template of the multi-task inference engine, the retrieval strategy of the knowledge-enhanced retrieval module, or the generation rules of the content generation module based on the feedback operation data.

8. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 1, characterized in that, Also includes: The intelligent workflow engine is used to generate a workflow that includes business operation steps, execution roles, and verification conditions based on the auxiliary processing information and in combination with pre-configured process decision rules and role permissions.

9. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 1, characterized in that, Also includes: The security control module is used to encrypt the communication link between the front end of the smart plugin and the large model service layer, and to verify the operation permissions of customer service personnel.

10. The intelligent customer service assistance system based on browser plugins and a large language model according to claim 5, characterized in that, The data access module is used to obtain multiple related data associated with the current working context from multiple business systems corresponding to the customer service system, specifically for: Based on the current working context, construct the query conditions; Based on the query conditions, multiple data interfaces of the business systems are called concurrently through a unified application programming interface gateway to obtain multiple related data.

Images