E-commerce customer service method and system based on Al

Abstract

The invention discloses an E-commerce customer service method and system based on Al, and the method comprises the steps: collecting the interaction data of a user on an E-commerce platform, and constructing a user portrait and user interest model; extracting commodity features in the e-commerce platform, carrying out digital processing on the commodity features to obtain digital commodity features, and calculating the similarity between the user portrait and each digital commodity feature; generating an adjacent matrix and a user feature matrix; according to the user feature matrix and the adjacent matrix, calculating a user interest vector by using a user interest model; and calculating the interest score of the user for each commodity by using the user interest vector and the digital commodity characteristics, and retrieving and pushing the user interest commodity in the customer service platform according to the similarity and the interest score, thereby realizing the customer service automatic pushing service. According to the embodiment of the invention, the method can achieve the efficient retrieval and pushing of the interested commodities of the user, improves the recommendation accuracy, can effectively improve the shopping experience of the user, and improves the competitiveness of an e-commerce platform.

Description

Technical Field

[0001] The present invention belongs to the technical field of data mining, and in particular to an e-commerce customer service method and system based on AI. Background Art

[0002] With the rapid development of the Internet, e-commerce has become one of the most important ways for people to shop. E-commerce platforms have attracted a large number of users through rich product selection and convenient online shopping experience. However, the massive amount of products and information that comes with it often makes users face the problem of information overload when shopping, resulting in difficulty in choosing and a reduced shopping experience. Therefore, how to efficiently and accurately meet user needs and improve user satisfaction has become a key challenge that e-commerce platforms need to solve urgently.

[0003] Traditional e-commerce customer service mainly relies on manual customer service, answering questions and recommending products to users through telephone, online chat, etc. However, this method has many limitations: 1. Slow response speed: The response time of manual customer service is usually long, especially during peak hours, the waiting time of users may increase significantly, thus affecting the user experience; 2. Low service efficiency: The number of requests that manual customer service can handle is limited, which cannot meet the immediate needs of a large number of users and easily causes user loss; 3. Insufficient personalized recommendations: Traditional customer service usually provides services based on users' immediate inquiries, lacks in-depth analysis of users' historical behaviors, preferences and purchasing habits, and is difficult to make personalized product recommendations; 4. Lack of data mining capabilities: E-commerce platforms have accumulated a large amount of user interaction data, and the existing manual customer service system is difficult to fully tap the potential value of this data, and cannot provide users with intelligent services based on deep learning and data analysis. Summary of the invention

[0004] The purpose of the present invention is to provide an e-commerce customer service method and system based on AI to address the deficiencies in the prior art. By constructing a dynamically updated user interest model, combined with product features in the e-commerce platform, and through deep mining of user behavior data and portraits, efficient retrieval and push of products of user interest can be achieved, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform.

[0005] An embodiment of the present application provides an e-commerce customer service method based on AI, the method comprising: Collect user interaction data on e-commerce platforms to build user portraits and user interest models based on graph convolutional networks; Extracting product features from the e-commerce platform and digitizing the product features to obtain digitized product features, and calculating the similarity between the user portrait and each digitized product feature; Generate an adjacency matrix representing the interaction between users and products, and a user feature matrix representing the basic attributes of users; Calculating a user interest vector reflecting the user's preference and interest in the product by using the user interest model according to the user feature matrix and the adjacency matrix; The user interest vector and the digital product features are used to calculate the user's interest score for each product, and the user's interest products in the e-commerce platform are retrieved and pushed according to the similarity and the interest score, thereby realizing automatic push service of customer service.

[0006] Optionally, the calculation formula of the user interest vector is:

[0007] Among them, the is the user interest vector, is the adjacency matrix, is the user feature matrix, is a learnable weight matrix, is the activation function.

[0008] Optionally, the calculation formula of the interest score is:

[0009] Among them, the Score user i's interest in product j, is the inner product of the digital product feature vector and the user interest vector. is the digital product feature vector of product j, , To adjust the parameters.

[0010] Optionally, the retrieving and pushing user-interested products on the e-commerce platform according to the similarity and the interest score includes: Retrieving a first product corresponding to a digital product feature having a similarity greater than a preset similarity threshold, and a second product having an interest score greater than a preset score; The intersection of the first product and the second product is used as a user interest product and pushed to the user.

[0011] Optionally, the method further includes: Receive the demand feedback information input by the user for the pushed interest products, include the demand feedback information in the interaction data, and return to execute the step of collecting the user's interaction data on the e-commerce platform to retrieve and push new user interest products on the e-commerce platform.

[0012] Another embodiment of the present application provides an e-commerce customer service system based on AI, the system comprising: The collection module is used to collect user interaction data on the e-commerce platform to build user portraits and user interest models based on graph convolutional networks; An extraction module is used to extract product features from the e-commerce platform and digitize the product features to obtain digitized product features, and calculate the similarity between the user portrait and each digitized product feature; A generation module, used to generate an adjacency matrix representing the interaction relationship between users and products, and a user feature matrix representing basic attributes of users; A calculation module, configured to calculate a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; The retrieval module is used to calculate the user's interest score for each product by using the user interest vector and the digital product features, and retrieve and push the user's interest products in the e-commerce platform according to the similarity and the interest score, so as to realize the automatic push service of the customer service.

[0013] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0014] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0015] Compared with the prior art, the present invention provides an e-commerce customer service method based on AI, which collects user interaction data on an e-commerce platform to construct a user portrait and a user interest model based on a graph convolutional network; extracts commodity features in the e-commerce platform and digitizes the commodity features to obtain digitized commodity features, and calculates the similarity between the user portrait and each digitized commodity feature; generates an adjacency matrix representing the interaction relationship between the user and the commodity, and a user feature matrix representing the basic attributes of the user; calculates a user interest vector reflecting the user's preference and interest in the commodity based on the user feature matrix and the adjacency matrix and the user interest model; calculates the user's interest score for each commodity based on the user interest vector and the digitized commodity features, and retrieves and pushes the user's interest commodities in the e-commerce platform based on the similarity and interest score, thereby being able to construct a dynamically updated user interest model, combined with the commodity features in the e-commerce platform, and through deep mining of user behavior data and portraits, to achieve efficient retrieval and push of user interest commodities, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A hardware structure block diagram of a computer terminal for an e-commerce customer service method based on Al provided in an embodiment of the present invention; Figure 2 A schematic diagram of a flow chart of an e-commerce customer service method based on Al provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an Al-based e-commerce customer service system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0018] The embodiment of the present invention first provides an e-commerce customer service method based on AI, which can be applied to electronic devices such as computer terminals, specifically ordinary computers, etc.

[0019] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal of an e-commerce customer service method based on Al provided by an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0020] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / modules corresponding to the e-commerce customer service method based on Al in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0021] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0022] See also Figure 2 The embodiment of the present invention provides an e-commerce customer service method based on AI, which may include the following steps: S201, collecting user interaction data on the e-commerce platform to build a user portrait and a user interest model based on a graph convolutional network; On modern e-commerce platforms, user behavior data is the key to understanding their interests and needs. By collecting user interaction data on the platform, a dynamic user profile can be constructed to make more accurate personalized recommendations. These interaction data include but are not limited to user browsing history, purchase history, product collections, evaluations, search behavior, etc. Through these data, user preference information for different product categories, brands, and features can be obtained. Building a user interest model based on graph convolutional networks (GCN) will further enhance the ability to capture user interests, because GCN can effectively parse the complex relationship between users and products. Role and significance: 1. Improved personalized recommendations: Through accurate user portraits and interest models, the relevance of recommendations can be significantly improved, increasing user satisfaction and repurchase rates. 2. Dynamic adaptability: User interests change with time and environment. Timely collection of interaction data enables user portraits to be dynamically updated to better meet users' immediate needs. 3. Information mining and insights: In-depth analysis of interactive data can not only help the recommendation system learn automatically, but also provide merchants with insights into user preferences, thereby formulating more effective marketing strategies and promotional activities. One implementation method may include: 1. User behavior logging: Design an efficient logging system to capture all user interactions on the e-commerce platform in real time, including page browsing, product clicks, add-to-cart, purchases, etc. This can be done through WebSocket, API calls, etc. for data transmission. 2. Behavioral data structuring: The captured raw log data is structured to form a data table containing fields such as user ID, product ID, behavior type, and timestamp.

[0023] Step 2: Build a user profile 1. Feature extraction and statistics: - Extract user feature information from structured data, such as: - Frequency characteristics: the frequency of browsing, adding to cart and purchasing product categories. - Timeliness feature: The time of the user's most recent purchase or browsing, used to calculate user activity. - Behavior sequence: Record the user's behavior pattern based on their behavior sequence. 2. User portrait construction: - Combine the extracted information into user profiles.

[0024] Step 3: User interest model based on graph convolutional network 1. Adjacency matrix construction: - Generate an adjacency matrix based on the interaction between users and products. The elements of the matrix can be the interaction frequency between users and products, indicating the connection strength between users and products. 2. User feature matrix construction: - Encode each user’s basic attributes (such as age, gender, points level, etc.) into a user feature matrix, with each row corresponding to one user. 3. GCN model design: - Define the graph convolution layer, and transfer information and transform features through the adjacency matrix and user feature matrix. The characteristic of the model is to calculate the user interest vector through network learning.

[0025] Step 4: Training and updating of interest models 1. Loss function setting: - Use loss functions based on interactive behavior, such as collaborative filtering loss function framework, to train the model through users’ positive interactions (such as purchases) and negative interactions (such as non-purchases). 2. Dynamic update mechanism: - Set a regular model training cycle (for example, daily or hourly) to train on new interaction data so that the user interest model is always up to date. Through the above steps, the system will be able to effectively collect user interaction data on the e-commerce platform, establish user portraits and user interest models based on graph convolutional networks, and lay a solid foundation for subsequent product recommendations. This process not only improves the user experience, but also provides a strong competitive advantage for the e-commerce platform.

[0026] S202, extracting product features from the e-commerce platform and digitizing the product features to obtain digitized product features, and calculating the similarity between the user portrait and each digitized product feature; In e-commerce platforms, there are many kinds of goods, and each product has multi-dimensional features, such as brand, type, price, specification, color, shelf time, etc. These features play a vital role in users' purchasing decisions. By extracting product features and digitizing them, this information can be effectively converted into a form that can be understood by computers, laying the foundation for the subsequent similarity calculation between user portraits and product features. The process of digitizing product features not only includes feature extraction and encoding, but also requires considering the relationship between features and how to combine them into a highly representative and structured data model. This process helps build a feature-based recommendation system, making it easier for users to find products that meet their preferences. 1. Improve recommendation accuracy: By accurately extracting and digitizing product features, it can provide richer information for the user interest model, making the recommendation results more relevant, thereby improving user satisfaction. 2. Multi-dimensional feature representation: Different commodities may have features of different importance. Through proper digital processing, the priority and importance of these features can be effectively reflected. 3. Support personalized needs: By calculating the similarity between user portraits and product features, it is possible to quickly respond to user personalized needs and help users find products that meet their needs more easily. One implementation method may include: Step 1: Product feature extraction 1. Data source identification: Determine the available product data sources on the e-commerce platform, including databases, API interfaces, product details pages, etc., to obtain basic information about the products.

[0027] 2. Feature definition and selection: A set of criteria for determining product features, which may include the following: - Category feature: The category to which the product belongs (e.g. electronics, clothing, furniture, etc.).

[0028] - Text features: text data such as product descriptions, brand names, and user reviews.

[0029] - Numerical features: numerical information such as price, inventory, and number of reviews.

[0030] - Symbol characteristics: applicability marks (such as seasonality, purpose, suitable for age groups, etc.).

[0031] 3. Feature extraction: - Write a data extraction script to extract the above features from product information and construct a product feature data table.

[0032] Step 2: Digital processing of product features 1. Feature encoding: Encode the extracted features for digitization: - Category Encoding: Use One-Hot Encoding to convert categorical features into binary vectors.

[0033] - Text feature processing: Use methods such as TF-IDF (Term Frequency-Inverse Document Frequency) or Word2Vec to convert text features into vector representation.

[0034] - Numerical normalization: For numerical features, apply minimum-maximum normalization or Z-score normalization so that the values ​​fall into the same range, thereby balancing the impact of different numerical features.

[0035] 2. Feature vector integration: Integrate various feature codes into a comprehensive digital product feature vector. For example, category features, text features, and numerical features can be combined into a complete feature vector.

[0036] Step 3: Calculate the similarity between user portrait and digital product features 1. User portrait vectorization: Convert user portraits into vector form to express user interests and preferences. These vectors should have the same dimensions as the feature vectors of digital products. 2. Similarity calculation method: Use measurement methods such as cosine similarity, Euclidean distance or Manhattan distance to calculate the similarity between the user portrait vector and the feature vector of each digital product.

[0037] 3. Similarity matrix construction: The calculated similarity results are structured to form a similarity matrix between the user and all products, which can be used in the subsequent recommendation stage. Through the above steps, the e-commerce platform can efficiently extract and digitize product features, and calculate the similarity between user portraits and digitized product features. This process not only provides basic data support for subsequent recommendation algorithms, but also improves the personalized recommendation capabilities of e-commerce platforms to meet users' diverse shopping needs.

[0038] S203, generating an adjacency matrix representing the interaction relationship between users and commodities, and a user feature matrix representing basic attributes of users; In the recommendation system, the adjacency matrix and the user feature matrix are important data structures that describe the relationship between users and products. The adjacency matrix is ​​used to represent the interaction between users and products, and can capture the user's preference pattern for products. The user feature matrix contains the user's basic attribute information and can provide an additional dimension about the user's personal characteristics. - Adjacency matrix: This matrix not only reflects the direct interaction between users and products, but also models complex interaction patterns through graph convolutional networks (GCNs), allowing the recommendation system to better understand the deep structure of user interests. - User feature matrix: This matrix integrates the user's static features (such as age, gender, occupation, etc.) as important inputs to the usage model, supplements the user's background information, and helps the model make personalized recommendations at the feature level. By generating these two matrices, complex user behaviors and features can be structured, making subsequent calculations and model training possible. 1. Information structuring: The complex relationship between users and products is structured through the adjacency matrix to provide clear input for subsequent model learning. 2. Enrich user information: The user feature matrix integrates a variety of user background information, enabling the model to perform more in-depth personalized analysis based on simple behavioral data. 3. Improve model effect: Combining the adjacency matrix and the user feature matrix enables the model to capture user preferences more comprehensively and provide guarantees for the accuracy of recommendation results. One implementation method may include: Step 1: Generation of adjacency matrix 1. Define the set of users and products: Determine the set of all users and products on the platform, and mark m users and n products. 2. Initialize the adjacency matrix: Create an adjacency matrix A of size m*n. Initially, all elements are 0, indicating no interaction.

[0039] 3. Fill in the adjacency matrix: For each piece of user interaction data, traverse and update the corresponding position of the adjacency matrix. For example, if user i purchases product j, then: A[i][j] = A[i][j] + 1. This records the number of interactions between the user and the product, which can reflect the strength of the association between the user and the product.

[0040] Step 2: Generation of user feature matrix 1. Feature definition: - Determine the user attributes that need to be included in the user feature matrix, such as age, gender, occupation, geographic location, account creation time, etc. 2. Feature standardization and encoding: Use appropriate encoding methods to standardize user features: - Normalize numerical features. For example, age can be normalized using Z-score. - Convert categorical variables (such as gender, occupation) to one-hot encoding. 3. Construct user feature matrix: - By integrating the feature information of each user into a matrix, a user feature matrix H is formed with a dimension of m*k, where k is the number of user features.

[0041] Step 3: Matrix Integration and Output 1. Merge operation: After generating the adjacency matrix A and the user feature matrix H, these two matrices can be used as input to provide the basic data structure for subsequent model training. 2. Preprocess the output: Ensure that the adjacency matrix and user feature matrix are in a good format and perform necessary conversions (such as converting them to sparse matrix format) for efficient operation. 3. Data storage: Save the generated adjacency matrix and user feature matrix in an efficient format to facilitate the subsequent training of the graph convolutional network and the calculation of the user interest model. Through the above steps, the system can effectively generate the adjacency matrix of the interaction relationship between users and products, as well as the user feature matrix representing the basic attributes of users. This not only provides the necessary data structure for the subsequent recommendation algorithm, but also enhances the recommendation system's ability to understand and respond to user interest-driven.

[0042] S204, calculating a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; In the recommendation system, user interest vector is a core concept, which comprehensively reflects the user's preference and interest in products. By combining the user feature matrix and the adjacency matrix, the user's behavior patterns and characteristics can be effectively modeled, thereby predicting the products that the user may be interested in in the future. - User feature matrix: contains the static attributes of each user, such as age, gender, geographic location, etc. These features provide basic background information of the user and help understand the user's personalized needs and preferences. - Adjacency matrix: represents the interaction between users and products, and can quantify the connection strength between users and products. This matrix can capture the user's behavior on the platform, such as browsing, purchasing, etc. By combining the two, the graph convolutional network (GCN) can process this type of structured data to calculate the user interest vector, reflecting the user's interest in various products. Functional significance: 1. Personalized recommendations: The calculation of user interest vectors enables the recommendation system to generate a personalized recommendation list for each user, thereby improving user satisfaction and platform activity. 2. Behavioral pattern mining: By analyzing user interest vectors, developers can deeply understand user behavior patterns and optimize product display, marketing strategies and user experience. 3. Dynamic adaptability: User interest vectors can be dynamically updated as user behavior changes, enabling the recommendation system to agilely adapt to users’ real-time needs.

[0043] Specifically, a calculation formula for a user interest vector may be:

[0044] Among them, the is the user interest vector, It is an adjacency matrix that reflects the interaction between users and products and can provide comprehensive information about user behavior. Through the adjacency matrix, the system can understand the degree of connection between users and each product, providing a basis for the formation of user interests. It is a user feature matrix, which contains the user's basic attributes and background information, and can help the model more accurately capture the source of user preferences. User features provide contextual support for the calculation of interest vectors, enabling the model to better understand user behavior. The weight matrix is ​​a learnable weight matrix that is used to adjust the influence of each feature in the final user interest vector, which ensures that the model can automatically optimize the importance of features during training. The weight learning process enables the user interest vector to reflect changes in user needs and pay more attention to specific features, thereby improving the recommendation effect of the model. This matrix is ​​used to adjust the influence of different features on the final interest vector. The initial value of the weight can be randomly generated and will be optimized later during the training process. is the activation function. The activation function can be implemented by selecting common activation functions such as ReLU (rectified linear unit) or Sigmoid. The activation function is used for nonlinear transformation to help the model capture more complex feature relationships. Design significance: 1. Linear transformation: It expresses the fusion of user features and the interaction relationship between them. First, the adjacency matrix A is multiplied by the user feature matrix H to obtain the feature representation based on user interaction data. This step can effectively combine the feature information of the products that users interact with. 2. Feature enhancement: The linear transformation of the weight matrix W is added to optimize and learn the weights of different features, so that the model can adaptively adjust the influence of each feature according to the training data. This mechanism makes the model more expressive.

[0045] 3. Non-linear mapping: activation function Usually, nonlinear functions such as ReLU or Sigmoid are used. Their main function is to introduce nonlinear characteristics, allowing the model to learn more complex user interest patterns. Through this activation function, the model can avoid the problem of insufficient expression ability that may be caused by linear transformation, thereby improving the flexibility and accuracy of the recommendation system.

[0046] From the above explanation, we can see that the calculation and design of user interest vectors is not only a key step in the recommendation system, but also an important tool for understanding user behavior and preferences. By effectively combining the adjacency matrix and the user feature matrix, and using activation functions and learnable weights, the system can achieve accurate personalized recommendations, greatly improving user experience and platform value.

[0047] S205, using the user interest vector and the digital product features, calculating the user's interest score for each product, and based on the similarity and the interest score, retrieving and pushing the user's interest products in the e-commerce platform to implement automatic customer service push service.

[0048] The core of this step is to calculate the user's interest score for each product by combining the user's interest vector and the digital product features. The user interest vector is constructed by analyzing the user's behavior and preferences, and can effectively reflect the user's personalized needs; the digital product features are a quantitative description of the product attributes, covering information such as price, category, and brand. By combining the two, the system can quantify the degree of match between the user and the product, and then generate a specific interest score. Next, the system will retrieve and push products based on the calculated interest score and the similarity between the user portrait and the product features. This process not only ensures that the products recommended to the user are consistent with their interests, but also improves the accuracy and relevance of the recommendation, thereby better meeting the user's shopping needs.

[0049] Specifically, a calculation formula for interest score can be:

[0050] The formula uses the form of logistic regression, so that the calculated interest score is mapped to between (0, 1), which is suitable for expressing the user's interest in a certain product. This nonlinear mapping can better adapt to the changes in user interest in the product and avoid the limitations of simple linearity.

[0051] Among them, the Score user i's interest in product j, It is the inner product of the digital product feature vector and the user interest vector, which can be used to measure the degree of matching between the user and the product features. The larger this value is, the higher the fit between the user and the product is, and the higher the interest score is. is the digital product feature vector of product j. The product of the user's interest vector and the digital product feature vector represents the user's overall interest in the product. This design takes into account the performance of user interest in specific products and further refines the scoring. , For the corresponding adjustment parameters, you can set them yourself and optimize them according to the needs of different scenarios.

[0052] Specifically, according to the similarity and the interest score, the user's interest products in the e-commerce platform are retrieved and pushed, and the first product corresponding to the digital product feature with a similarity greater than a preset similarity threshold and the second product with an interest score greater than a preset score can be retrieved; Filter out products with high similarity to the user profile from all products on the e-commerce platform. This is a preliminary screening based on the similarity between the calculated user profile and product features. The preset similarity threshold is a set standard, the purpose of which is to ensure that the selected products can effectively match the user's interests, preferences and needs. Only products with a similarity higher than this threshold will be considered as potential recommended products.

[0053] The calculated interest score is compared with a preset score threshold to filter out products with scores above the score threshold. The setting of this score threshold can be adjusted based on historical data and user behavior to ensure that the recommended products have a certain attractiveness and potential purchase possibility.

[0054] The intersection of the first product and the second product is used as a user-interested product and pushed to the user. Alternatively, the intersection of the first product and the second product is used as a user-interested product, which is also reasonable and feasible.

[0055] The selected products are intersected to extract the product set that the user is most likely to be interested in. By combining the features of the first product (high similarity product) and the second product (high interest score product), the system ensures that the products pushed to the user are the best matches that have been strictly screened. The intersection operation ensures that the recommended products not only match the user's interest vector, but also match the user's overall behavior characteristics. This comprehensive recommendation can significantly improve user satisfaction.

[0056] The above steps together constitute a systematic product recommendation mechanism, which emphasizes the cross-analysis of similarity and interest score to improve the effect of personalized recommendation. This not only enhances user experience and improves satisfaction, but also helps e-commerce platforms achieve higher sales efficiency and user stickiness, providing strong support for overall business development.

[0057] Furthermore, in actual applications, it is also possible to receive demand feedback information input by users on the pushed products of interest, include the demand feedback information in the interaction data, and return to execute the step of collecting the user's interaction data on the e-commerce platform to retrieve and push new user interest products on the e-commerce platform.

[0058] The core of this step is to continuously optimize and adjust the product recommendations of e-commerce customer service through an active user feedback mechanism. Specifically, the system will receive user feedback on the pushed products of interest (such as satisfaction, interest, and whether they want to view other similar products, etc.), and incorporate this information into the user's interaction data. This feedback can not only update the user portrait, but also help the system record the user's real-time preference changes. 1. User feedback collection: After receiving the recommended products, users can provide feedback to the system through simple interface operations (such as clicking, rating, commenting, etc.). This feedback information can be specific product evaluations, satisfaction with the recommendations, or other types of products that users would like to see. 2. Update interaction data: Store user feedback information in interaction data and update user behavior records. This dynamic update can ensure that the system has a more accurate grasp of user interests and ensure the effectiveness of subsequent recommendations. 3. Re-execute the interaction data collection step: The system will re-execute the user interaction data collection step based on the updated interaction data, re-evaluate the user's interests and needs, and then generate a new user profile. 4. Retrieve and push new products: After obtaining the updated user portrait, the system will recalculate the similarity and interest score with the product based on the new user interest, and then retrieve new user interest products for push. - Personalized service improvement: The access to user feedback ensures that the recommendation system can flexibly adjust the recommendation strategy according to the ever-changing user needs and provide more personalized and accurate product recommendation services. - Improve user satisfaction: By actively collecting user feedback, the system can better understand the real needs of users and respond in a timely manner, thereby improving the overall shopping experience and satisfaction of users. Users feel that their opinions are valued, which in turn strengthens their trust in the platform. - Dynamically optimize the recommendation algorithm: Including demand feedback information in the interaction data can not only be used to adjust the user portrait, but also provide richer training data for the recommendation algorithm, promote the continuous optimization and self-learning ability of the algorithm, and enhance the accuracy of the recommendation. - Enhance user stickiness: Through continuous interaction and feedback mechanisms, companies can better maintain relationships with users, improve user stickiness, reduce user churn, and thus improve the platform's user retention rate. - Quickly respond to market changes: User needs and market trends change rapidly. Through an effective feedback mechanism, the system can quickly capture these changes and adjust product recommendation strategies in a timely manner to help the platform adapt to market dynamics and maintain competitive advantages. Through the above steps and function descriptions, the effective collection and processing of user feedback provides good dynamic adaptability for the AI-based e-commerce customer service method, so that the system can always be in an optimized state and truly realize intelligent personalized recommendations.

[0059] It can be seen that the user's interaction data on the e-commerce platform is collected to build a user portrait and a user interest model based on a graph convolutional network; the product features in the e-commerce platform are extracted and digitized to obtain the digital product features, and the similarity between the user portrait and each digital product feature is calculated; an adjacency matrix representing the interaction relationship between the user and the product, and a user feature matrix representing the basic attributes of the user are generated; based on the user feature matrix and the adjacency matrix, the user interest model is used to calculate the user's interest vector that reflects the user's preference and interest in the product; the user interest vector and the digital product features are used to calculate the user's interest score for each product, and based on the similarity and interest score, the user's interest products in the e-commerce platform are retrieved and pushed, so that by building a dynamically updated user interest model, combined with the product features in the e-commerce platform, and through deep mining of user behavior data and portraits, efficient retrieval and push of user interest products can be achieved, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform.

[0060] Another embodiment of the present invention provides an e-commerce customer service system based on AI, see Figure 3 , the system may include: The collection module 301 is used to collect the interaction data of users on the e-commerce platform to build user portraits and user interest models based on graph convolutional networks; An extraction module 302 is used to extract product features from the e-commerce platform and digitize the product features to obtain digitized product features, and calculate the similarity between the user portrait and each digitized product feature; A generation module 303 is used to generate an adjacency matrix representing the interaction relationship between users and commodities, and a user feature matrix representing basic attributes of users; A calculation module 304 is used to calculate a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; The retrieval module 305 is used to calculate the user's interest score for each product by using the user interest vector and the digital product features, and retrieve and push the user's interest products in the e-commerce platform according to the similarity and the interest score, so as to realize the automatic push service of customer service.

[0061] It can be seen that the user's interaction data on the e-commerce platform is collected to build a user portrait and a user interest model based on a graph convolutional network; the product features in the e-commerce platform are extracted and digitized to obtain the digital product features, and the similarity between the user portrait and each digital product feature is calculated; an adjacency matrix representing the interaction relationship between the user and the product, and a user feature matrix representing the basic attributes of the user are generated; based on the user feature matrix and the adjacency matrix, the user interest model is used to calculate the user's interest vector that reflects the user's preference and interest in the product; the user interest vector and the digital product features are used to calculate the user's interest score for each product, and based on the similarity and interest score, the user's interest products in the e-commerce platform are retrieved and pushed, so that by building a dynamically updated user interest model, combined with the product features in the e-commerce platform, and through deep mining of user behavior data and portraits, efficient retrieval and push of user interest products can be achieved, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform.

[0062] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0063] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps: S201, collecting user interaction data on the e-commerce platform to build a user portrait and a user interest model based on a graph convolutional network; S202, extracting product features from the e-commerce platform and digitizing the product features to obtain digitized product features, and calculating the similarity between the user portrait and each digitized product feature; S203, generating an adjacency matrix representing the interaction relationship between users and commodities, and a user feature matrix representing basic attributes of users; S204, calculating a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; S205, using the user interest vector and the digital product features, calculating the user's interest score for each product, and based on the similarity and the interest score, retrieving and pushing the user's interest products in the e-commerce platform to implement automatic customer service push service.

[0064] It can be seen that the user's interaction data on the e-commerce platform is collected to build a user portrait and a user interest model based on a graph convolutional network; the product features in the e-commerce platform are extracted and digitized to obtain the digital product features, and the similarity between the user portrait and each digital product feature is calculated; an adjacency matrix representing the interaction relationship between the user and the product, and a user feature matrix representing the basic attributes of the user are generated; based on the user feature matrix and the adjacency matrix, the user interest model is used to calculate the user's interest vector that reflects the user's preference and interest in the product; the user interest vector and the digital product features are used to calculate the user's interest score for each product, and based on the similarity and interest score, the user's interest products in the e-commerce platform are retrieved and pushed, so that by building a dynamically updated user interest model, combined with the product features in the e-commerce platform, and through deep mining of user behavior data and portraits, efficient retrieval and push of user interest products can be achieved, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform.

[0065] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0066] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0067] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program: S201, collecting user interaction data on the e-commerce platform to build a user portrait and a user interest model based on a graph convolutional network; S202, extracting product features from the e-commerce platform and digitizing the product features to obtain digitized product features, and calculating the similarity between the user portrait and each digitized product feature; S203, generating an adjacency matrix representing the interaction relationship between users and commodities, and a user feature matrix representing basic attributes of users; S204, calculating a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; S205, using the user interest vector and the digital product features, calculating the user's interest score for each product, and based on the similarity and the interest score, retrieving and pushing the user's interest products in the e-commerce platform to implement automatic customer service push service.

[0068] It can be seen that the user's interaction data on the e-commerce platform is collected to build a user portrait and a user interest model based on a graph convolutional network; the product features in the e-commerce platform are extracted and digitized to obtain the digital product features, and the similarity between the user portrait and each digital product feature is calculated; an adjacency matrix representing the interaction relationship between the user and the product, and a user feature matrix representing the basic attributes of the user are generated; based on the user feature matrix and the adjacency matrix, the user interest model is used to calculate the user's interest vector that reflects the user's preference and interest in the product; the user interest vector and the digital product features are used to calculate the user's interest score for each product, and based on the similarity and interest score, the user's interest products in the e-commerce platform are retrieved and pushed, so that by building a dynamically updated user interest model, combined with the product features in the e-commerce platform, and through deep mining of user behavior data and portraits, efficient retrieval and push of user interest products can be achieved, which can not only improve the accuracy of recommendations, but also effectively improve the user's shopping experience and enhance the competitiveness of the e-commerce platform.

[0069] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. An e-commerce customer service method based on AI, characterized in that: The method comprises: Collect user interaction data on e-commerce platforms to build user portraits and user interest models based on graph convolutional networks; Extracting product features from the e-commerce platform and digitizing the product features to obtain digitized product features, and calculating the similarity between the user portrait and each digitized product feature; Generate an adjacency matrix representing the interaction between users and products, and a user feature matrix representing the basic attributes of users; Calculating a user interest vector reflecting the user's preference and interest in the product by using the user interest model according to the user feature matrix and the adjacency matrix; The user interest vector and the digital product features are used to calculate the user's interest score for each product, and the user's interest products in the e-commerce platform are retrieved and pushed according to the similarity and the interest score, so as to realize the automatic push service of customer service.

2. The method according to claim 1, characterized in that The calculation formula of the user interest vector is: Among them, the is the user interest vector, is the adjacency matrix, is the user feature matrix, is a learnable weight matrix, is the activation function.

3. The method according to claim 2, characterized in that The calculation formula of the interest score is: Among them, the Score user i's interest in product j, is the inner product of the digital product feature vector and the user interest vector. is the digital product feature vector of product j, , To adjust the parameters.

4. The method according to claim 3, characterized in that The retrieving and pushing user-interested products in the e-commerce platform according to the similarity and the interest score includes: Retrieving a first product corresponding to a digital product feature having a similarity greater than a preset similarity threshold, and a second product having an interest score greater than a preset score; The intersection of the first product and the second product is used as a user-interested product and pushed to the user.

5. The method according to claim 4, characterized in that The method further comprises: Receive the demand feedback information input by the user for the pushed interest products, include the demand feedback information in the interaction data, and return to execute the step of collecting the user's interaction data on the e-commerce platform to retrieve and push new user interest products on the e-commerce platform.

6. An e-commerce customer service system based on AI, characterized in that: The system comprises: The collection module is used to collect user interaction data on the e-commerce platform to build user portraits and user interest models based on graph convolutional networks; An extraction module is used to extract product features from the e-commerce platform and digitize the product features to obtain digitized product features, and calculate the similarity between the user portrait and each digitized product feature; A generation module, used to generate an adjacency matrix representing the interaction relationship between users and products, and a user feature matrix representing basic attributes of users; A calculation module, configured to calculate a user interest vector reflecting the user's preference and interest in the product based on the user feature matrix and the adjacency matrix and using the user interest model; The retrieval module is used to calculate the user's interest score for each product by using the user interest vector and the digital product features, and retrieve and push the user's interest products in the e-commerce platform according to the similarity and the interest score, so as to realize the automatic push service of the customer service.

7. The system according to claim 6, characterized in that The calculation formula of the user interest vector is: Among them, the is the user interest vector, is the adjacency matrix, is the user feature matrix, is a learnable weight matrix, is the activation function.

8. The system according to claim 7, characterized in that The calculation formula of the interest score is: Among them, the Score user i's interest in product j, is the inner product of the digital product feature vector and the user interest vector. is the digital product feature vector of product j, , To adjust the parameters.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images