Intelligent customer service method for point shopping mall based on large star model

Abstract

Along with continuous development of technologies such as deep learning and natural language processing, the AI model shows powerful capabilities in the aspects of understanding user intentions, processing complex information, realizing accurate matching and the like. The large star model is a large AI model which is developed by China Telecom and is based on an advanced Transform architecture. The Transformer architecture obtains significant results in the fields of natural language processing, image recognition and the like due to good expandability and strong performance of the Transformer architecture. When the method is applied to a customer service scene, the large star model can deeply understand the appeal of the user. Through natural language processing and semantic understanding technologies, the model can accurately identify and analyze questions of a user and quickly give corresponding answers or suggestions. In addition, the large star model also has strong copywriting generation ability and knowledge question-answering ability, can provide personalized exchange recommendation and shopping preferential information for the user, and further improves the user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of electronic commerce, customer service, artificial intelligence, multi-modal fusion, data processing, etc., in particular to a method for intelligent customer service of an integral mall based on a star model. BACKGROUND

[0002] In recent years, integral malls are facing huge user traffic and complex user demands, and the traditional customer service mode has been difficult to meet the efficient and accurate service needs. Therefore, China Telecom's demand for intelligent customer service is increasingly urgent. Intelligent customer service not only can provide 24-hour uninterrupted service, but also can realize accurate user portrait and customer service demand prediction through big data analysis and artificial intelligence technology, and provide more personalized service experience for users.

[0003] With the progress of science and technology, the vertical application of artificial intelligence in the search field is becoming more and more extensive and in-depth, especially in recent years, with the continuous development of deep learning, natural language processing and other technologies, AI models have shown strong ability in understanding user intent, processing complex information and realizing accurate matching. Star model is a large AI model developed by China Telecom based on advanced Transformer architecture. Transformer architecture has good scalability and strong performance, and has achieved remarkable results in natural language processing, image recognition and other fields. Applied to the customer service scene, the star model can deeply understand the user's appeal. Through natural language processing and semantic understanding technology, the model can accurately identify and analyze the user's problem and quickly give the corresponding answer or suggestion. In addition, the star model also has strong copywriting generation ability and knowledge question and answer ability, which can provide personalized exchange recommendation and shopping discount information for users, further improving the user experience.

[0004] Although the star model shows significant advantages in understanding and processing user problems, the response time of intelligent customer service is still a key challenge to be overcome. Simply relying on performance stacking is difficult to fundamentally alleviate the waiting delay problem of intelligent customer service. This patent aims to solve this pain point and rely on the star model to innovatively build an integral mall intelligent customer service system. SUMMARY

[0005] Module 1: User portrait construction and label management module (attached Figure 1 : S01) This module focuses on building and managing user profiles, providing solid data support for intelligent customer service judgment and response. Based on basic user information such as gender and age, and through UTM parameter tracking, it collects data on user points changes in the points mall, browsing behavior, search behavior, purchase behavior, and historical customer service requests. Preprocessing methods such as data cleaning, summarization, and standardization ensure the quality and consistency of user behavior data. Feature extraction is then performed on the processed user behavior data. Based on the Starry Sky Model, and combined with multiple features for comprehensive analysis, users are divided into different groups or categories. These groups can be divided according to different characteristics, behavioral patterns, purchasing preferences, etc., to determine "high probability of user consumption", "medium probability of user consumption", and "low probability of user consumption". Based on this, user profiles are formed and matched with relevant user tags, such as "high frequency redemption user", "high redemption demand user", and "frequent customer complaint demander". The aforementioned tags and weights need to be dynamically adjusted based on actual data and business needs. Furthermore, tag generation should be a dynamic process; as user data is updated and characteristics change, the tags should also be adjusted and updated accordingly.

[0006] Module 2: Multimodal Understanding of User Needs Module (Appendix) Figure 1 (S02) This module processes multi-modal information (such as text, images, and audio) uploaded by users during interaction. It integrates RTC (Real-Time Communication) technology, efficiently fusing and transmitting multi-modal information through streaming, shortening transmission time. Leveraging the multi-modal understanding capabilities of the Starry Sky Big Data model, it performs in-depth analysis and understanding of the information, identifying problems, scenarios, and user needs within the multi-modal information, thus laying the foundation for providing more accurate and personalized solutions.

[0007] Simultaneously, by combining a sentiment dictionary and a sentiment classifier, semantic analysis is performed on the text input by users. By identifying sentiment words and user sentiment features, feature extraction is performed based on TF-IDF. The model is then trained using a Logistic Regression model. The importance of words is evaluated by the frequency of the text submitted by users and the inverse document frequency in the entire corpus. Through cross-validation, the model is continuously optimized to determine the user's current sentiment tendency and understand and analyze the user's needs.

[0008] Here is a simple example code for sentiment analysis using Python and the scikit-learn library: In audio interaction, by analyzing the pitch changes, speaking speed, and emotional expression in the user's voice, and based on the existing emotional speech database RAVDESS, we ensure that the user's audio corresponds to the correct emotion label. We train the SVM vector machine model and continuously optimize it through cross-validation, so that intelligent customer service can judge the user's emotional state based on the audio. By sensing user emotions, personalized services can be provided. When customers encounter problems, their emotions can be perceived and soothed, preventing negative emotions from escalating and making them feel valued and cared for. Simultaneously, issues can be assigned to the most suitable customer service personnel, providing the customer service team with precise service suggestions.

[0009] Module 3 Customer Response Preprocessing and Optimization Module (with appendix) Figure 1 (S03) This module aims to leverage the deep learning and analysis capabilities of the Starry Sky model, combining the core user needs and emotional states extracted from previous modules with customer service history records to predict upcoming user demands and potential questions, generating preliminary responses covering multiple dimensions, and providing users with a comprehensive and detailed initial response.

[0010] As users respond, NLP technology is used for semantic matching and contextual analysis. By integrating the preliminary responses generated in the previous preprocessing stage, the response that best meets the user's needs is matched, and the content of the response is quickly optimized to form the final response, ensuring that the response is not only coherent but also highly relevant to the user's actual needs. This process will greatly enhance the automation of handling common problems, significantly reduce user waiting time, effectively alleviate the burden on customer service personnel, and ensure a fast and efficient response to user needs.

[0011] Module 4: Human Customer Service Response Module (with appendix) Figure 1 (S04) This module focuses on the integration of intelligent customer service and human customer service. In the intelligent customer service system, the human customer service response module plays a crucial role, ensuring that complex issues or important customers are promptly and effectively transferred to professional human agents for handling. This module operates based on user tags generated by previous modules, including key information such as the user's consumption probability, historical customer complaint records, current emotional state, and tags formed from search history.

[0012] When a user's needs trigger conditions for human customer service intervention, such as exhibiting high-value user behavior without a completed purchase, multiple complaint records, or expressing anger or anxiety in their requests, the intelligent customer service system will automatically and seamlessly transfer the conversation to a human customer service representative, leveraging available resources. During this transfer, the system uses a comprehensive model to display the user's basic information, spending potential score, historical complaint details, current sentiment analysis results, search history tags, and a summary of the conversation between the intelligent customer service representative and the user, providing the human representative with a complete and concise overview of the user's needs. To further improve the response efficiency and service quality of human customer service, the Xingchen Big Data Model provides a series of pre-processing suggestions based on user profiles and conversation content, including processing priorities, communication strategies, possible solutions, and related resources. These suggestions help human customer service representatives understand the user's situation quickly and formulate appropriate response strategies.

[0013] After receiving a transfer request, a human customer service representative will briefly review the user's information and pre-processing suggestions. Based on this information, they will adjust their communication style and begin a conversation with the user to resolve their issues or needs. After the conversation ends, the representative will record the key points, the outcome, and the user's feedback. This information will be used to update the user profile and optimize the intelligent customer service system's learning model and triggering conditions, thus forming a continuous improvement service loop.

[0014] Module 5: Multi-cloud Architecture Distributed Deployment Module (with appendix) Figure 1 (S05) This module implements a distributed deployment of the intelligent customer service system through a multi-cloud architecture. The system is decomposed into multiple microservice components, each undertaking a different and specific functional role, thus modularizing the system's functionality. Through distributed processing, these microservices can process user requests in parallel, significantly improving the system's concurrency capabilities. This parallel processing mechanism ensures that the system can respond quickly and efficiently even when faced with a large number of simultaneous user requests. After parallel processing is complete, the system intelligently aggregates and merges the results from each microservice. This process ensures the integrity and accuracy of information, thereby providing users with a final and precisely calculated solution.

[0015] Meanwhile, this intelligent customer service approach fully leverages the elastic scalability of the multi-cloud architecture. By dynamically adjusting computing resources, the system can easily handle high-concurrency access and sudden traffic spikes, ensuring that user requests receive timely and effective responses. Furthermore, the multi-cloud architecture guarantees failover and load balancing, further improving system stability and availability. Even if a cloud environment experiences a failure, the system can quickly switch to another healthy cloud environment, ensuring continuous service operation.

[0016] Module 6: Intelligent Customer Service Product Recommendation Module (Appendix) Figure 1 (S06) This module analyzes the user profiles generated by the Starry Sky Big Data Model, and recommends products to users based on multi-dimensional data such as recent browsing preferences, redemption behavior, historical redemption records, and points usage.

[0017] This module plays a key touchpoint role in the operation of the points mall. It not only intelligently recommends suitable points-based products based on users' points balance and redemption history, but also guides users to actively participate in points redemption activities through targeted marketing messages, thereby increasing the activity level and user stickiness of the points mall.

[0018] Finally, this module is continuously optimized in conjunction with the Starry Sky model. Based on user feedback and redemption data, the recommendation strategy is constantly adjusted to ensure that the recommended products always match the users' actual needs and preferences. Attached Figure Description

[0019] Figure 1 The structure and flowchart of a method for intelligent customer service in an points mall based on a star-shaped big data model.

Claims

1. A method for intelligent customer service in an points mall based on a star-centric model, characterized in that: The system collects basic user information and tracks user behavior data in the points mall (points changes, browsing, searching, purchasing, and historical customer service requests) through UTM parameter tracking. This data is then cleaned and mined to create user profiles, which are then categorized with tags. Integrated RTC technology enables rapid transmission of multimodal information related to user requests, and the system uses a comprehensive data model to understand the intent behind these multimodal messages. For text, TF-IDF feature extraction and a Logistic Regression model combined with a sentiment dictionary are used to analyze user text sentiment. For audio, the RAVDESS database and SVM model are used to analyze the emotional state of user voices. The system utilizes the comprehensive data model to analyze core user requests and emotions, combined with historical records, to predict user needs and generate initial responses. Based on subsequent user responses, NLP techniques are used for semantic matching and contextual analysis to optimize response content. When a request triggers human intervention, the system automatically transfers the conversation and provides a dialogue summary to human customer service through the comprehensive data model. Personalized product recommendations are made by analyzing user profiles and multidimensional data (browsing preferences, redemption behavior, historical records, and points usage).

2. The method for intelligent customer service in an points mall based on a star-shaped large model as described in claim 1, characterized in that: User profile building and tagging management module: This module collects user data based on basic user information and using UTM parameter tracking, including data on user points changes, browsing, searching, purchasing, and historical customer service requests in the points mall. After ensuring data quality and consistency through data preprocessing, features are extracted and comprehensively analyzed using the Xingchen big data model. Users are then divided into different groups or categories based on characteristics, behavioral patterns, and purchasing preferences, and assigned corresponding tags.

3. A method for intelligent customer service in an points mall based on a star-centric model, as described in claim 1, characterized in that: The multimodal understanding module for user requests utilizes RTC technology to efficiently integrate and rapidly transmit various modalities of information, including text, images, and audio, during user interactions. Leveraging the multimodal understanding capabilities of the Xingchen big data model, it deeply analyzes this information to accurately identify problems, scenarios, and user requests. In text interactions, it combines a sentiment dictionary and classifier, using TF-IDF feature extraction and Logistic Regression model training to determine user sentiment and gain a deeper understanding of their requests. For audio interactions, it analyzes the user's voice tone, speech rate, and emotional expression, referencing the RAVDESS sentiment speech database, and uses an SVM vector machine model trained and optimized through cross-validation to enable intelligent customer service to judge the user's emotional state based on audio.

4. A method for intelligent customer service in an points mall based on a star-shaped large model, as described in claim 1, characterized in that: Customer Response Pre-processing and Optimization Module: When a user's situation triggers conditions for human customer service intervention (such as a high-value user who hasn't made a purchase, multiple complaints, or emotional agitation), the intelligent customer service system seamlessly transfers the conversation to a human agent, leveraging available resources. During the transfer, the Starry Sky Model displays a user overview to the human agent, including basic information, spending potential score, complaint details, sentiment analysis, search tags, and a conversation summary. To improve efficiency, the Starry Sky Model also provides pre-processing suggestions, such as processing priorities, communication strategies, solutions, and related resources. The human agent adjusts their communication style accordingly, engaging with the user and resolving the issue. After the conversation, the human agent records key points, results, and feedback to update the user profile and optimize the system, forming a continuous improvement service loop.

5. A method for intelligent customer service in an points mall based on a star-centric model, as described in claim 1, characterized in that: Human customer service response module: Collects and records user behavior data during the search process, including keywords, click behavior, dwell time, etc.; mines multi-dimensional data through the Starry Sky Big Data Model to provide information for building user profiles and optimizing search algorithms.

6. A method for intelligent customer service in an points mall based on a star-centric model, as described in claim 1, characterized in that: Purchase prediction module: Intelligent customer service product recommendation module. When users inquire about products, it intelligently recommends products based on the user's recent browsing preferences, points balance, and redemption history. It can also improve user engagement and mall activity through precise marketing information. By combining user feedback and redemption data, the module continuously optimizes its recommendation strategy using the Starry Sky Big Data Model to ensure that recommendations are accurate and meet user needs.

Images