E-commerce operation auxiliary method and system based on active AI intelligent agent
By employing a proactive AI-powered e-commerce management approach, we have achieved efficient and intelligent management of e-commerce operations. This approach addresses the passivity and fragmentation issues of existing tools, provides one-stop intelligent services, and enhances merchants' operational efficiency and risk warning capabilities.
Patent Information
- Application Number
- CN202511171839.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-21
- Publication Date
- 2026-01-09
AI Technical Summary
Existing e-commerce operation support tools mostly respond passively to user queries, making it difficult to proactively identify potential operational problems. Their functional modules are scattered, requiring switching between multiple systems. The integration of knowledge and business data is not high, making it difficult to provide comprehensive decision support. Their intelligence and convenience are insufficient, failing to meet merchants' multi-dimensional needs for cost reduction and efficiency improvement, risk warning, and other aspects.
This approach employs an e-commerce operation support method based on proactive AI agents. Through multi-source data access and status monitoring, proactive perception and intent recognition, dynamic task scheduling, and generation and output of intelligent early warning information, it integrates intelligent question answering, high-frequency business operations, content creation, and business analysis. It utilizes machine learning models to identify anomalies and risks and provides a unified interactive interface.
It enables efficient, intelligent, and closed-loop management of e-commerce operations, helps merchants promptly identify and warn of potential problems, simplifies operating procedures, lowers the barriers to creation and operation, provides a one-stop intelligent entry point, dynamically interprets changes in platform rules, reduces the risk of violations, and enhances commercial value.
Smart Images

Figure CN121304208A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of electronic commerce, in particular to an e-commerce operation auxiliary method and system based on an active AI agent. BACKGROUND
[0002] With the rapid development of e-commerce, the business operation scene of merchants is becoming increasingly complex, involving order processing, logistics tracking, after-sales service, marketing promotion and other multi-link cooperation. There is an increasing demand for efficient and intelligent operation assistance tools. Currently, artificial intelligence technology has been applied in the e-commerce field, such as answering common questions through customer service robots and providing sales data statistics with the help of data analysis tools. These technologies have improved the operational efficiency of merchants to some extent, but the overall service is still in the stage of decentralization and passivity, which cannot meet the comprehensive needs of merchants for active perception, intelligent decision-making and integrated services in complex business scenarios.
[0003] In practical applications, there is room for improvement in existing technical solutions. For example, most auxiliary tools mainly respond to user queries passively and are difficult to actively identify potential problems in the operation process. Different functional modules often run independently, and merchants need to switch between multiple systems, which affects processing efficiency. The integration of knowledge and business data is not high, making it difficult to provide comprehensive decision support. At the same time, in terms of content creation and dynamic rule adaptation, the intelligence level and convenience of existing tools still have room for improvement, and they cannot fully meet the multi-dimensional business needs of merchants for cost reduction and efficiency improvement, risk warning, etc. For this reason, we propose an e-commerce operation auxiliary method and system based on an active AI agent. SUMMARY
[0004] To solve the above technical problems, an e-commerce operation auxiliary method and system based on an active AI agent are provided. This technical solution solves the problems of passive response to user queries, difficulty in actively identifying potential business problems, scattered functional modules, the need to switch between multiple systems, affecting efficiency, insufficient integration of knowledge and business data, limited decision support, and the need for improvement in the intelligence and convenience of content creation and rule adaptation.
[0005] To achieve the above purposes, the technical solution adopted by the present application is as follows: An e-commerce operation auxiliary method based on an active AI agent includes the following steps: S1, multi-source data access and state monitoring: real-time or periodic access to e-commerce operation related data, synchronize update internal knowledge base, and monitor the state of preset key business indicators; S2, active perception and intent recognition: analyze user input through natural language processing to identify user intent, and scan the accessed business data through an active patrol engine to identify state abnormalities or risk opportunity points based on preset rules; S3. Task Scheduling and Intelligent Processing: Based on the identified user intent or proactive inspection results, dynamically schedule and execute at least one of the following functionalities: intelligent question answering based on a knowledge base; execution of preset high-frequency business operations; extraction of product selling points and generation of marketing content; processing of marketing images; generation of business analysis reports; and execution of rule monitoring and risk warnings. S4. Result Generation and Proactive Output: Integrate processing results and output them through a unified intelligent interactive interface; for abnormalities or risks identified by proactive inspection, generate early warning information containing problem descriptions and handling suggestions, and provide different levels of reminders according to preset rules.
[0006] Preferably, in step S2, the specific method for identifying abnormal states or risk / opportunity points based on preset rules is as follows: The proactive inspection engine periodically scans key business data tables in timed mode, and initiates scanning in response to specific business events or knowledge update events in event-triggered mode. During the scanning process, the current business data is compared with preset threshold rules to identify at least one of the following abnormal states or risk opportunities: Logistics timeliness anomaly: The duration of the calculated transportation status exceeds a preset threshold; After-sales processing timeout risk: Identify after-sales orders that are in the pending status and have less than a threshold before the preset processing deadline; A surge in complaints: The number of specific types of complaints exceeds the historical baseline or a preset threshold within a specific time period; Incomplete data: Key information fields are missing in the detected business data; Impact of the rules: Detect platform rule updates and identify potentially affected businesses or products; The identification results are associated with the corresponding business data details and trigger subsequent processing.
[0007] Preferably, in step S2, the proactive inspection engine uses a machine learning model trained based on historical business data to identify state anomalies or risk opportunity points: Collect historical business data and corresponding result labels, and train at least one of the following models: A time-series anomaly detection model is used to predict the probability of logistics and transportation delays or abnormal nodes. A classification model is used to assess the risk level of after-sales order timeout. Statistical process control or clustering models are used to detect sudden and abnormal increases in complaint data; A rule impact assessment model is used to predict the risk score of product violations; The trained model is integrated into the proactive inspection engine to replace or enhance the preset rule judgment logic, calculate the current business data and output the predicted abnormal state label or risk score as the basis for identification.
[0008] Preferably, in step S3, the execution of the preset high-frequency service operation specifically includes: When data query operations are involved, the corresponding business system database is accessed directly, relevant data is retrieved according to the query conditions, and then converted into a structured format for output. When business processing operations are involved, receive user operation instructions, complete status updates or data modifications by calling the application programming interface of the business system, and record operation logs. The unified interaction interface associates the structured data obtained from the query with the corresponding business processing operation functions and outputs them, allowing users to directly trigger processing operation instructions based on the query results.
[0009] Preferably, in step S3, the process of extracting product selling points and generating marketing content, as well as processing marketing images, specifically includes: Selling point extraction: Receive product information, identify key features by analyzing product attributes or image content, and extract core selling point keywords; Marketing content generation: Based on the extracted selling point keywords and user needs, a generative text model is used to generate marketing copy; Marketing image processing: Provides at least one of the following functions: It receives user-uploaded images and editing instructions, and calls image processing algorithms to perform cropping, stretching, or object erasure operations. Receive text description and parameters, and call the image and text generation model to generate marketing poster images; Receive images and parameters of clothing products, and call an image generation model to generate virtual model display images.
[0010] Preferably, in step S3, the extraction of product selling points and generation of marketing content are continuously optimized using user feedback data: Record user actions regarding the generated text or images, including acceptance, modification, or rejection. Collect the content samples that users ultimately adopt, and store them in association with the initial input instructions, product information, extracted selling point keywords, and the prompt word templates used; The selling point keyword extraction model was fine-tuned using the collected sample dataset; Analyze the differences between user-adopted content and the initially generated content to extract user-preferred language styles, structures, or image styles; Based on the results of the difference analysis and the adopted samples, the content and structure of the prompt word template library were adjusted and optimized. Use the updated model and template to process subsequent requests.
[0011] The process of generating early warning information for anomalies or risks identified through proactive inspections and providing different levels of alerts based on preset rules specifically includes: Preferably, in step S4, the step of generating early warning information for anomalies or risks identified by proactive inspection and providing different levels of alerts according to preset rules specifically includes: Access the knowledge base based on the identified status anomalies or risk types, combine specific business data details to conduct impact analysis and risk assessment, and classify different risk levels based on preset rules. Generate structured early warning information, including anomaly / risk descriptions, affected data, association rule knowledge references, risk levels, and handling suggestions; Based on the risk level field contained in the warning information, the output method is selected according to the preset push rules, including: for high risk levels, triggering pop-ups, push notifications and emails; for lower risk levels, displaying via notifications or in the information flow of the interactive interface; The warning message includes a link that redirects to the relevant business interface or knowledge details.
[0012] Preferably, it further includes: The generated proactive warning information or business suggestions will be automatically converted into to-do tasks, specifically including: Analyze the content of early warning information or business suggestions to extract task type, related object identifier, operation steps, priority and suggested completion time; Match a preset task template according to the task type. The template defines the task title rules, description field, assignment rules and deadline rules. Use the extracted elements to populate the template and generate a structured to-do task item that includes task title, description, responsible person, deadline and status fields; Synchronize the generated to-do items to the merchant's workflow management system; Provide a link to the task in the interactive interface and synchronize status updates.
[0013] An e-commerce operation assistance system based on proactive AI agents is used to implement the e-commerce operation assistance method based on proactive AI agents, including: The data access and fusion module is configured to access e-commerce operation-related data and internal knowledge base; The perception and recognition module includes a natural language processing engine for parsing user input and recognizing user intent, and an active inspection engine for scanning business data and recognizing abnormal statuses or risk opportunities. The core processing module is configured to perform at least one of the following functions: intelligent question answering; high-frequency business operations; selling point extraction and marketing content generation; marketing image processing; business analysis report generation; rule monitoring and risk warning; The proactive service module is used to generate early warning information or trigger tasks based on proactive inspection results, and to integrate the processing results. A unified interaction interface is used to receive user input and output processing results and warning information.
[0014] Preferably, the unified interaction interface supports multimodal interaction, specifically including: An integrated speech recognition engine converts received voice commands into text input for the perception and recognition module; An integrated speech synthesis engine converts text-based results into speech signals for output. Provides a graphical interface for receiving, editing, and displaying images; When processing high-risk warning information, trigger an audible alarm.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The e-commerce operation assistance method and system proposed in this invention can proactively sense and monitor e-commerce business data, promptly identify and warn of potential problems, and help merchants take measures to avoid risks in advance. It integrates multiple functions such as intelligent Q&A, high-frequency business processing, content creation, and business analysis, providing merchants with a one-stop intelligent entry point, simplifying the operation process, lowering the threshold for creation and operation, dynamically interpreting changes in platform rules, providing compliance guidance for merchants, reducing the risk of violations, and realizing the high efficiency, intelligence, and closed-loop management of e-commerce operations by transforming from passive response to proactive service, thus creating greater commercial value for merchants. Attached Figure Description
[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system framework diagram of the present invention. Detailed Implementation
[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0018] Reference Figure 1As shown, this is an e-commerce operation assistance method based on proactive AI agents. The core of this method lies in leveraging the proactive perception and processing capabilities of AI agents to assist e-commerce operators in efficient decision-making and execution. The entire process begins with the access and status monitoring of multi-source data. These data sources are diverse, including but not limited to e-commerce platform backends (such as order, product, and inventory data), logistics tracking system APIs, customer service ticket systems, user review data, platform rule announcements, and the merchant's internal knowledge base (such as operation manuals and historical cases). These data sources are chosen to comprehensively cover key aspects of operations. Data access can be real-time streaming (such as order status changes and logistics node updates) or timed batch extraction (such as daily sales statistics and inventory checks) to ensure that the internal knowledge base (such as product information databases, rule databases, and frequently asked questions databases) is updated synchronously, providing the latest basis for subsequent processing. Simultaneously, preset key business indicators (such as order fulfillment rate, after-sales response time, complaint rate, and inventory turnover rate) are continuously monitored, and changes in their status are important signals triggering subsequent actions.
[0019] Proactive perception and intent recognition constitute the key to the "proactive" nature of the intelligent agent. On the one hand, a natural language processing engine parses the text or voice commands entered by the user through the interactive interface to understand their explicit intent to query questions, request operations, or seek help. On the other hand, an independent proactive inspection engine operates continuously, with two operating modes: in timed mode, it scans key business data tables (such as logistics status tables, after-sales work order tables, and complaint record tables) according to a set period (such as every 5 minutes or every hour); in event-triggered mode, it immediately initiates scanning in response to specific business events (such as new complaints, platform rule change notifications) or knowledge base update events. During the scanning process, a dual mechanism is used to identify abnormal statuses or risk opportunities. The first is a pre-set rule comparison: matching the current business data with pre-set threshold rules derived from business experience and historical data analysis. For example, the system calculates the time elapsed for orders in the "in transit" state. If this time exceeds a preset reasonable timeliness threshold (which is usually set based on the historical average timeliness of the logistics route and product type plus a certain buffer period), it is marked as an abnormal logistics timeliness. It identifies after-sales orders in the "pending processing" state and calculates the remaining time until the preset processing deadline (set according to platform rules or the merchant's internal service commitment). If this time is less than the set warning threshold (e.g., 2 hours remaining), it determines that there is a risk of after-sales processing exceeding the time limit. It counts the number of complaints of a specific type (e.g., "damaged goods") within a specific time period (e.g., the past hour). If this number exceeds the historical baseline value for that time period (e.g., the average value of the same period in the past 7 days) or a directly set absolute value threshold, it is marked as a sudden increase in complaints. It directly detects whether key information fields in business data (e.g., missing main image on product details page, missing delivery address on order) are missing, determining that the data is incomplete. When a platform rule base update is detected, it identifies parts of currently sold products or business operations that may violate the new rules or be significantly affected by the new rules through keyword matching or rule impact scope analysis, marking them as rule-affected. The identification results are linked to specific business data details (such as problem order numbers, risky after-sales order IDs, lists of sudden increases in complaints, product IDs with missing fields, and lists of affected product IDs). Secondly, the engine also integrates machine learning models for more complex identification: it trains specific models using historical business data (such as logistics node time-series data and whether the event was delayed, historical after-sales order data and whether it was overdue, historical complaint data and whether it was a sudden increase, product attributes, and historical violation records).For example, training time-series anomaly detection models (such as LSTM or Prophet) analyzes logistics trajectories to predict the probability of current shipment delays or identify abnormal stalls; training classification models (such as XGBoost or Random Forest) assesses the timeout risk level (high, medium, low) based on features such as order type, processing time, and current status; applying statistical process control (SPC) charts or clustering algorithms (such as K-means, DBSCAN) detects whether complaint volumes significantly deviate from normal fluctuation ranges; and training rule impact assessment models (such as classifiers based on text features and product attributes) predicts the violation risk score for specific products under new rules. These trained models are integrated into the proactive inspection engine, which can replace simple threshold rules or supplement and enhance its judgment logic to calculate current business data and output predicted anomaly status indicators or risk scores as a more accurate identification basis.
[0020] Based on identified user intent or proactive inspection results, the task scheduling and intelligent processing modules dynamically schedule and execute various functions. Intelligent question answering based on a knowledge base: Utilizing a real-time updated knowledge base, it directly retrieves matching answers or generates concise and accurate responses to answer user questions. Execution of preset high-frequency business operations: For data query operations (such as "querying yesterday's unshipped orders"), the agent directly accesses the corresponding business system database (such as the order database), retrieves relevant data based on the parsed query conditions, and converts the database's native format results into a more easily understood structured format (such as a JSON table) for output. For business processing operations (such as "marking order 123 as shipped"), after receiving a clear operation instruction from the user, the agent completes the status update or data modification by calling the standardized application programming interface provided by the business system, and records detailed operation logs for auditing. In a unified interaction interface, the agent can associate the structured data obtained from the query with the corresponding business processing operation function for output. For example, while displaying a "list of pending after-sales orders," it provides a "one-click processing" button, allowing users to directly trigger processing operation instructions based on the query results, improving operational efficiency. The process of extracting product selling points and generating marketing content begins with the following steps: First, selling points are extracted by receiving product information (such as title, attributes, description, and images). Key features (such as material, function, and design highlights) are identified through analysis of product text attributes or by using computer vision technology to analyze image content, and core selling point keywords (such as "pure cotton," "ultra-long battery life," and "waterproof") are extracted. Next, based on the extracted selling point keywords and specific user needs (such as target audience, advertising platform, and style requirements), generative text models (such as GPT series, Claude, and other large language models) are used to generate suitable marketing copy (such as short product titles, detail page descriptions, and advertising slogans). Perform marketing image processing: Provides multiple functions, including receiving user-uploaded original images and editing instructions (such as "change the background to a beach" or "remove the clothes hanger behind the model"), and calling image processing algorithms (such as image segmentation and image restoration) to perform operations such as cropping, stretching, or object erasure; receiving text descriptions and style parameters (such as "summer coolness" or "e-commerce promotional style"), and calling image generation models (such as Stable Diffusion and DALL-E) to generate marketing poster images that meet the requirements; specifically for clothing products, it receives flat lay or hanging photos and parameters (such as model body shape and pose), and calls image generation models (such as virtual try-on technology based on GAN or Diffusion) to generate virtual model display images to enhance product appeal. Perform business analysis report generation: Integrates multi-source data and uses data analysis technology to generate periodic or thematic reports covering multiple dimensions such as sales, traffic, users, and services. Perform rule monitoring and risk warning: Continuously monitors changes in platform rules and business indicators, and combines proactive inspection results to identify potential violations or business risks.It's worth noting that the product selling point extraction and marketing content generation functions introduce a user feedback-driven continuous optimization mechanism: the system records the user's final actions on the generated text or images (adopting for direct use, modifying for use, or completely discarding); it collects samples of the content ultimately adopted by users, and associates and stores them with the initial input instructions, original product information, selling point keywords extracted by the system, and prompt word templates used during generation, forming an optimization dataset; using these collected positive sample datasets, the selling point keyword extraction model (such as the text feature extraction model) is fine-tuned to better meet actual business needs and user preferences; the differences between the user-adopted content and the system's initial generated content are analyzed to extract specific aspects of user preferences, such as language style (more colloquial / more professional), text structure (selling point pre-positioning / scenario-based description), and image style (simple / rich); based on these difference analysis results and adopted samples, the content and structure of the prompt word template library used to call the generation model are adjusted and optimized, so that the templates can better guide the model to generate results that satisfy users; finally, the updated keyword extraction model and optimized prompt word templates are used to process subsequent requests, forming a closed-loop optimization.
[0021] The results generation and proactive output module is responsible for integrating the results of all the above processing. For results actively requested by users (such as query data, generated text and images, and analysis reports), they are clearly presented through a unified intelligent interaction interface. For anomalies or risks identified by the proactive inspection engine, the system generates structured early warning information: First, based on the specific type of identification (such as logistics delays, after-sales timeout risks, and a sudden increase in complaints), it accesses the knowledge base to obtain relevant knowledge, and combines it with specific affected business data details (such as the current sorting center of delayed orders, the remaining processing time of overdue after-sales orders, and the specific content of sudden increases in complaints) to conduct impact analysis and risk assessment, and classifies different risk levels (such as urgent, high, medium, and low) according to preset rules (such as the amount of the affected order, the level of the customer involved, and the urgency of the timeout). Then, generate early warning information containing key elements: a clear description of the anomaly / risk (e.g., "The average transit time for orders shipped from the South China warehouse to East China has exceeded the 48-hour warning threshold"), a specific range of affected data (e.g., a list of order numbers involved), relevant rule knowledge references (e.g., "According to the 'Logistics Timeliness Commitment,' standard express delivery should be sent within 48 hours"), the determined risk level (e.g., "High"), and feasible handling suggestions (e.g., "Contact the logistics carrier to verify the cause and send batch delay reassurance notices"). Finally, based on the risk level field marked in the early warning information, strictly select the output method according to the preset push rules: for urgent or high-risk levels (e.g., those that may cause customer complaints or fines), immediately trigger a strong pop-up reminder, push mobile notification, and send a detailed email; for medium- and low-risk levels (e.g., missing but non-critical data fields), use a gentler notification or display it in the information flow of the interactive interface to avoid excessive disturbance. All early warning information includes links that can directly jump to the relevant business interface (e.g., a list of problematic orders, a page of pending after-sales orders) or a knowledge base details page, making it easy for users to quickly locate and handle the issue.In addition, the system will automatically convert proactive warning information or important business suggestions (such as "A hot-selling product is expected to sell out in 3 days, it is recommended to replenish the stock") into to-do tasks: parsing the information content, extracting the task type (such as "logistics verification", "after-sales processing", "inventory replenishment"), associated object identifiers (such as order number, after-sales tracking number, product SKU), suggested operation steps, priority based on risk level, and suggested completion time (such as setting a deadline based on overdue risk); matching the task type with preset task templates (these templates predefine the rules for generating task titles, which fields the description should include, and default assignments). The rules specify which person is responsible and how the deadline is calculated based on urgency. The extracted elements are used to populate the matched template, automatically generating structured to-do items that include a standardized task title, detailed description, designated responsible person, clear deadline, and initial status (e.g., "Pending"). These generated to-do items are then synchronized via API to the merchant's existing workflow management system (e.g., DingTalk, WeChat Work, Jira) and integrated into their daily task management process. A unified interaction interface also provides links to specific tasks within the workflow system and synchronizes task status updates (e.g., completed, closed) for easy user tracking.
[0022] Reference Figure 2 As shown, an e-commerce operation support system based on an active AI agent is the specific implementation of the above-mentioned methods. Its core components include: The data access and fusion module is responsible for establishing connections with internal and external data sources, accessing e-commerce platform operation data (orders, products, user behavior), logistics tracking information, customer service and after-sales data, user reviews, platform rule updates, etc. in real time or on a regular schedule. At the same time, it maintains and synchronizes the internal knowledge base (product knowledge base, rule base, FAQ base, and best practice case base) to provide a unified and timely data foundation for the entire system.
[0023] The perception and recognition module is the core of the system's intelligence, comprising two key engines: a natural language processing engine, responsible for parsing the text or speech-to-text input by the user through the interactive interface, accurately identifying the true intent behind the user's query, command, or question; and an active inspection engine, which operates independently, scanning business data according to a preset cycle or in response to specific events (data updates, rule changes), using preset rules and integrated machine learning models (such as time series prediction models, classification models, and clustering models) to identify abnormal states (logistics delays, data gaps) or risk opportunities (after-sales timeouts, sudden increases in complaints, rule impacts), and relating them to specific business data.
[0024] The core processing module is the system's "execution brain." Based on the output of the perception and recognition module, it dynamically schedules and executes various functions: intelligent question answering utilizes a knowledge base to quickly respond to user queries; high-frequency business operation functions perform data queries (directly accessing the business database and formatting results) and business processing (updating system status via API calls); the selling point extraction and marketing content generation functions analyze product information to extract keywords and call large language models to generate copy; the marketing image processing functions provide image editing, text-to-image, and virtual model image capabilities; the business analysis report function integrates data to generate analytical insights; and the rule monitoring and risk warning functions conduct continuous risk assessment based on data and rules.
[0025] The proactive service module focuses on proactive output: it receives the results of the proactive inspection engine, generates structured early warning information (including description, impact, suggestions, and level), triggers different levels of reminder pushes, and automatically converts early warning information or business suggestions into to-do tasks and synchronizes them to the external workflow system.
[0026] The unified interaction interface serves as a bridge between users and the system, supporting multimodal interaction: it integrates a speech recognition engine, allowing users to issue commands via voice; it integrates a speech synthesis engine, converting text results (such as report summaries and warning notifications) into speech broadcasts; it provides a graphical interface for users to upload and edit images and view system-generated image results; and when processing high-risk warning information, this interface can also trigger sound alarm prompts to ensure that critical information is not overlooked. All modules work collaboratively, receiving user input through this interface and outputting processing results, warning information, task links, etc., providing users with a seamless intelligent assistance experience.
[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. An e-commerce operation assistance method based on proactive AI intelligent agents, characterized in that, Includes the following steps: S1. Multi-source data access and status monitoring: Real-time or scheduled access to e-commerce operation-related data, synchronous updates to the internal knowledge base, and monitoring of the status of preset key business indicators. S2. Proactive perception and intent recognition: User input is parsed through natural language processing to identify user intent. At the same time, the proactive inspection engine scans the accessed business data and identifies abnormal statuses or risk opportunities based on preset rules. S3. Task Scheduling and Intelligent Processing: Based on the identified user intent or the results of proactive inspections, dynamically schedule and execute at least one of the following functional processes: intelligent question answering based on a knowledge base; executing preset high-frequency business operations; extracting product selling points and generating marketing content; and processing marketing images. Generate business analysis reports; monitor and warn against violations of rules; S4. Result Generation and Active Output: Integrate the processing results and output them through a unified intelligent interactive interface; For anomalies or risks identified through proactive inspections, early warning information containing a problem description and handling suggestions is generated, and different levels of alerts are issued according to preset rules.
2. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, In step S2, the specific method for identifying abnormal states or risk / opportunity points based on preset rules is as follows: The proactive inspection engine periodically scans key business data tables in timed mode, and initiates scanning in response to specific business events or knowledge update events in event-triggered mode. During the scanning process, the current business data is compared with preset threshold rules to identify at least one of the following abnormal states or risk opportunities: Logistics timeliness anomaly: The duration of the calculated transportation status exceeds a preset threshold; After-sales processing timeout risk: Identify after-sales orders that are in the pending status and have less than a threshold before the preset processing deadline; A surge in complaints: The number of specific types of complaints exceeds the historical baseline or a preset threshold within a specific time period; Incomplete data: Key information fields are missing in the detected business data; Impact of the rules: Detect platform rule updates and identify potentially affected businesses or products; The identification results are associated with the corresponding business data details and trigger subsequent processing.
3. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, In step S2, the proactive inspection engine uses a machine learning model trained based on historical business data to identify state anomalies or risk opportunity points: Collect historical business data and corresponding result labels, and train at least one of the following models: A time-series anomaly detection model is used to predict the probability of logistics and transportation delays or abnormal nodes. A classification model is used to assess the risk level of after-sales order timeout. Statistical process control or clustering models are used to detect sudden and abnormal increases in complaint data; A rule impact assessment model is used to predict the risk score of product violations; The trained model is integrated into the proactive inspection engine to replace or enhance the preset rule judgment logic, calculate the current business data and output the predicted abnormal state label or risk score as the basis for identification.
4. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, In step S3, the execution of the preset high-frequency service operation specifically includes: When data query operations are involved, the corresponding business system database is accessed directly, relevant data is retrieved according to the query conditions, and then converted into a structured format for output. When business processing operations are involved, receive user operation instructions, complete status updates or data modifications by calling the application programming interface of the business system, and record operation logs. The unified interaction interface associates the structured data obtained from the query with the corresponding business processing operation functions and outputs them, allowing users to directly trigger processing operation instructions based on the query results.
5. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, Step S3, which involves refining product selling points and generating marketing content, as well as processing marketing images, specifically includes: Selling point extraction: Receive product information, identify key features by analyzing product attributes or image content, and extract core selling point keywords; Marketing content generation: Based on the extracted selling point keywords and user needs, a generative text model is used to generate marketing copy; Marketing image processing: Provides at least one of the following functions: It receives user-uploaded images and editing instructions, and calls image processing algorithms to perform cropping, stretching, or object erasure operations. Receive text description and parameters, and call the image and text generation model to generate marketing poster images; Receive images and parameters of clothing products, and call an image generation model to generate virtual model display images.
6. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, In step S3, the extraction of product selling points and generation of marketing content are carried out, and continuous optimization is performed using user feedback data: Record user actions regarding the generated text or images, including acceptance, modification, or rejection. Collect the content samples that users ultimately adopt, and store them in association with the initial input instructions, product information, extracted selling point keywords, and the prompt word templates used; The selling point keyword extraction model was fine-tuned using the collected sample dataset; Analyze the differences between user-adopted content and the initially generated content to extract user-preferred language styles, structures, or image styles; Based on the results of the difference analysis and the adopted samples, the content and structure of the prompt word template library were adjusted and optimized. Use the updated model and template to process subsequent requests.
7. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, In step S4, the step of generating early warning information for anomalies or risks identified by proactive inspection and providing different levels of alerts according to preset rules specifically includes: Access the knowledge base based on the identified status anomalies or risk types, combine specific business data details to conduct impact analysis and risk assessment, and classify different risk levels based on preset rules. Generate structured early warning information, including anomaly / risk descriptions, affected data, association rule knowledge references, risk levels, and handling suggestions; Based on the risk level field contained in the warning information, the output method is selected according to the preset push rules, including: for high risk levels, triggering pop-ups, push notifications and emails; for lower risk levels, displaying via notifications or in the information flow of the interactive interface; The warning message includes a link that redirects to the relevant business interface or knowledge details.
8. The e-commerce operation assistance method based on an active AI agent according to claim 1, characterized in that, Also includes: The generated proactive warning information or business suggestions will be automatically converted into to-do tasks, specifically including: Analyze the content of early warning information or business suggestions to extract task type, related object identifier, operation steps, priority and suggested completion time; Match a preset task template according to the task type. The template defines the task title rules, description field, assignment rules and deadline rules. Use the extracted elements to populate the template and generate a structured to-do task item that includes task title, description, responsible person, deadline and status fields; Synchronize the generated to-do items to the merchant's workflow management system; Provide a link to the task in the interactive interface and synchronize status updates.
9. An e-commerce operation support system based on an active AI intelligent agent, characterized in that: The method for assisting e-commerce operations based on an active AI agent as described in any one of claims 1-5 includes: The data access and fusion module is configured to access e-commerce operation-related data and internal knowledge base; The perception and recognition module includes a natural language processing engine for parsing user input and recognizing user intent, and an active inspection engine for scanning business data and recognizing abnormal statuses or risk opportunities. The core processing module is configured to perform at least one of the following functions: intelligent question answering; high-frequency business operations; selling point extraction and marketing content generation; marketing image processing; business analysis report generation; rule monitoring and risk warning; The proactive service module is used to generate early warning information or trigger tasks based on proactive inspection results, and to integrate the processing results. A unified interaction interface is used to receive user input and output processing results and warning information.
10. The e-commerce operation assistance system based on an active AI intelligent agent according to claim 9, characterized in that, The unified interaction interface supports multimodal interaction, specifically including: An integrated speech recognition engine converts received voice commands into text input for the perception and recognition module; An integrated speech synthesis engine converts text-based results into speech signals for output. Provides a graphical interface for receiving, editing, and displaying images; When processing high-risk warning information, trigger an audible alarm.
Citation Information
Cited By
AI native ERP enterprise management software system and management method
CN121526101A
