Product recommendation method based on graph neural network and transfer learning

Abstract

The invention discloses a product recommendation method based on a graph neural network and transfer learning, and relates to the technical field of e-commerce recommendation. Performing embedded learning on the multimode network according to the graph neural network to obtain an embedded vector of each user; determining neighbor users of the target user; determining matching scores between the target user and all online product categories according to historical online product categories to which products purchased by the neighbor users of the target user belong and the embedded vector of the target user; migrating offline industries to online product categories through migration learning to obtain a plurality of target online product categories with strongest association obtained by mapping the offline industries corresponding to the neighbor users so as to update matching scores between the target users and all the online product categories; and for any online product category, recommending the online product category to a plurality of users with the highest matching scores. The method can improve the accuracy of e-commerce platform recommendation results.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce recommendation technology, and in particular to a product recommendation method based on graph neural network and transfer learning. Background Art

[0002] At present, with the continuous development of e-commerce, the scale of online shopping users will continue to grow, resulting in the application of recommendation technology in shopping websites, and the effect is becoming more and more obvious.

[0003] In the exploration of personalized recommendation methods, various e-commerce platforms are continuously increasing their investment and conducting in-depth exploration to recommend products of interest to users to promote orders. Existing recommendation methods mainly rely on users' online browsing, purchasing, and collection behaviors, and predict users' interests and preferences by analyzing these data, and then make product recommendations.

[0004] However, existing recommendation methods suffer from the problem that the recommendation results are not accurate enough. Summary of the invention

[0005] Based on this, it is necessary to provide a product recommendation method based on graph neural network and transfer learning to address the above technical problems. This method can improve the accuracy of recommendation results on e-commerce platforms.

[0006] The present invention adopts the following technical solutions: The present invention provides a product recommendation method based on graph neural network and transfer learning, comprising: Based on offline transaction and online shopping data, a multi-mode network of users, offline industries and online product categories is constructed; the multi-mode network includes the association between multiple users and the offline industries to which the products purchased offline belong, and the association between each user and the online product categories to which the products purchased online belong; The multi-mode network is embedded and learned based on the graph neural network to obtain the embedding vector of each user; the user's embedding vector includes the user's embedding vector in each offline industry and each online product category; According to the embedding vector of each user, determine the neighboring users of the target user; the target user is any user among all users; Determine the matching score between the target user and all online product categories based on the historical online product categories to which the products purchased by the target user's neighboring users belong and the target user's embedding vector; Through transfer learning, the offline industry is migrated to the online product category, and multiple target online product categories with the strongest correlation are obtained by mapping the offline industry corresponding to the neighboring users. The matching score between the target user and all online product categories is updated based on the alignment matching score between the offline industry corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedding vector; For any online product category, the online product category is recommended to multiple users with the highest matching scores.

[0007] Optionally, determining neighboring users of the target user according to the embedding vector of each user includes: Get the distance between the target user's embedding vector and other users' embedding vectors; The K users whose embedding vectors are closest to the target user are regarded as the target user’s neighbor users.

[0008] Optionally, according to the historical online product categories to which the products purchased by the neighboring users of the target user belong and the embedding vector of the target user, the matching scores between the target user and all the online product categories are determined, including: Determine the initial matching score between the target user and all online product categories to be 0; According to the distance between the embedding vector of the target user in the historical online product categories and the embedding vector of the neighboring users in the historical online product categories, the matching score between the target user and all online product categories is updated.

[0009] Optionally, according to the distance between the embedding vector of the target user in the historical online product categories and the embedding vector of the neighboring user in the historical online product categories, the matching score between the target user and all online product categories is updated, including: Traverse all neighboring users of the target user, and update the matching score between the target user and all online product categories based on the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighboring user in the historical online product category; The updated matching score is used as the initial matching score. According to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the second neighbor user in the historical online product category, the matching score between the target user and all online product categories is continued to be updated until all neighbor users are traversed and the matching score between the target user and all online product categories is obtained.

[0010] Optionally, according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category, the matching score between the target user and all online product categories is updated, including: For any historical online product category of the first neighbor user, update the matching score between the target user and the historical online product category according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category; Among them, the matching scores between the target user and all online product categories except the historical online product categories remain unchanged; the updated calculation formula for the matching scores between the target user and the historical online product categories is: ; in, is the updated matching score between the target user and the online product category. is the initial matching score between the target user and the online product category, It is the distance between the embedding vector of the target user in the online product category and the embedding vector of the neighboring user in the online product category.

[0011] Optionally, the process of determining multiple target online product categories with the strongest association obtained by mapping the offline industries corresponding to the neighboring users includes: For any neighbor user, obtain multiple online related product categories that are closest to the embedding vector of each historical offline industry that the neighbor user has purchased products from; Multiple online related product categories corresponding to each historical offline industry in which the neighboring user has purchased products are determined as multiple target online product categories.

[0012] Optionally, according to the alignment matching scores between the offline industries corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedding vectors, the matching scores between the target users and all online product categories are updated, including: Traverse all neighboring users of the target user, and update the matching scores between the target user and all online product categories based on the alignment matching scores between the offline industry corresponding to the first neighboring user and the corresponding multiple target online product categories based on the embedding vectors; The updated matching score is used as the initial matching score. According to the alignment matching scores between the offline industry corresponding to the second neighbor user and the corresponding multiple target online product categories based on the embedding vector, the matching scores between the target user and all online product categories are continued to be updated until all neighbor users are traversed.

[0013] Optionally, updating the matching scores between the target user and all online product categories according to the alignment matching scores between the offline industry corresponding to the first neighbor user and the corresponding multiple target online product categories based on the embedding vectors includes: For any target online product category, update the matching score between the target user and the target online product category based on the alignment matching score between the offline industry corresponding to the first neighbor user and the target online product category based on the embedding vector, and the distance between the embedding vector of the first neighbor user in the target online product category and the embedding vector of the target user in the target online product category; Among them, the matching scores between the target user and all online product categories except the target online product category remain unchanged; the calculation formula for the matching score between the target user and the target online product category is: ; in, is the updated matching score between the target user and the target online product category. is the initial matching score between the target user and the target online product category, is the distance between the embedding vector of the target user in the target online product category and the embedding vector of the neighboring user in the target online product category, It is the alignment matching score between the offline industry corresponding to the neighboring users and the target online product category based on the embedding vector.

[0014] Optionally, before recommending the online product category to a plurality of users with the highest matching scores, the method further includes: For any user, the matching scores between the user and all online product categories are normalized, and the normalized matching scores are used as the matching scores between the user and all online product categories.

[0015] The present invention provides a product recommendation device based on graph neural network and transfer learning, comprising: A construction module is used to construct a multi-mode network of users, offline industries and online product categories based on offline transaction and online shopping data; the multi-mode network includes associations between multiple users and offline industries to which products purchased offline belong, and between each user and online product categories to which products purchased online belong; The learning module is used to embed the multi-mode network according to the graph neural network to obtain the embedding vector of each user; the user's embedding vector includes the user's embedding vector in each offline industry and each online product category; The first determination module is used to determine the neighboring users of the target user according to the embedding vector of each user; the target user is any user among all users; The second determination module is used to determine the matching scores between the target user and all online product categories based on the historical online product categories to which the products purchased by the target user's neighboring users belong and the target user's embedding vector; An update module is used to migrate offline industries to online product categories through transfer learning, obtain multiple target online product categories with the strongest association obtained by mapping the offline industries corresponding to the neighboring users, and update the matching scores between the target users and all online product categories based on the alignment matching scores between the offline industries corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedding vectors; The recommendation module is used to recommend any online product category to multiple users with the highest matching scores.

[0016] The present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned product recommendation method based on graph neural network and transfer learning.

[0017] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned product recommendation method based on graph neural network and transfer learning when executing the program.

[0018] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects: First, a multi-mode network including users, offline industries and online product categories is constructed. The network represents the association between users and offline industries, and users and online product categories. Then, a graph neural network model is used to perform random walks in the network to learn the embedding vector of each user. Furthermore, the offline industry is effectively migrated to the online e-commerce recommendation through transfer learning. Specifically, based on the learned embedding vector, the strength of the association between offline industries and online product categories is determined, that is, which offline industries have a close association with which online product categories. In this way, when generating recommendations for users, not only the user's historical online product categories are referenced, but also their offline physical purchases are comprehensively considered. According to the strength of the association between offline industries and online products, the user's matching score is further updated, thereby providing more comprehensive and accurate information support for online e-commerce recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 A flowchart of a product recommendation method based on graph neural network and transfer learning provided by the present invention; Figure 2 A schematic diagram of the structure of a multimode network provided by the present invention; Figure 3 A schematic diagram of another process of a product recommendation method based on graph neural network and transfer learning provided by the present invention; Figure 4 A schematic diagram of a product recommendation device based on graph neural network and transfer learning provided by the present invention; Figure 5 A schematic diagram of a computer device for implementing a product recommendation method based on graph neural network and transfer learning provided by the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0021] As consumer shopping behaviors become more multi-channel and complex, it is difficult to fully capture users’ consumption habits and potential needs by relying solely on online data. For example, users’ offline consumption behaviors in industries such as catering, retail, and beauty salons are often related to online product purchases, but traditional recommendation systems cannot effectively utilize these offline data.

[0022] Existing recommendation methods can usually only use the accumulated data of online e-commerce for recommendation. The available information is limited and easily affected by the cold start problem, that is, the recommendation effect is not ideal due to the lack of sufficient historical data. At the same time, traditional recommendation methods based on single platform data also have certain limitations in dealing with complex user behaviors and cross-platform data fusion. They cannot fully explore the consumption correlation between users on different platforms, thus limiting the performance of the recommendation system and user experience. In this case, traditional recommendation methods may not be able to provide accurate user interest estimation, especially when dealing with high-dimensional data and complex user behaviors. In addition, existing recommendation methods mainly focus on improving recommendation efficiency, but lack targeted solutions when dealing with cross-platform data. Although some studies have proposed some data fusion methods, these methods are still limited to specific situations and cannot be widely applied to all recommendation problems, especially in the case of cross-platform data fusion.

[0023] Therefore, the present invention aims to propose a product recommendation method based on graph neural network and transfer learning. This method combines the complex relationship capture capability of graph neural network and the knowledge transfer advantage of transfer learning, and can effectively deal with problems such as high-dimensional data, heterogeneous user behavior, and cross-platform data fusion. Through this method, not only can the accuracy of recommendations be improved, but also the application scope of recommendation systems in complex data environments can be expanded, providing more reliable personalized recommendation support.

[0024] The execution subject of the method provided in the present invention may be a server arranged on a business platform, or a device such as a desktop computer, a notebook computer, etc. that can execute the solution of the present invention.

[0025] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0026] Figure 1 The following is a flow chart of a product recommendation method based on graph neural network and transfer learning in the present invention, which specifically includes the following steps: S101, based on offline transaction and online shopping data, construct a multi-mode network of users, offline industries and online product categories; the multi-mode network includes the association relationship between multiple users and the offline industries to which the products purchased offline belong, and each user and the online product category to which the products purchased online belong.

[0027] Among them, the offline industries to which all products purchased online by users belong, as well as the online product categories to which products purchased online belong, can be obtained from the transaction system. The transaction system is a cross-platform system that includes transaction data from multiple payment platforms.

[0028] like Figure 2 As shown, Figure 2 This is a structural diagram of a multi-mode network. For this multi-mode network, each offline industry, online product category and all users can be regarded as nodes of the multi-mode network.

[0029] S102, embedding learning is performed on the multi-mode network according to the graph neural network to obtain an embedding vector for each user; the user's embedding vector includes the user's embedding vector in each offline industry and each online product category.

[0030] Based on the graph neural network, the multi-modal network is embedded and learned to obtain the embedding vectors of all nodes, that is, a low-dimensional vector representing the local network structure information is obtained for each node. Among them, the embedding vectors of offline industries and online product categories can be used to explore the correlation information between online industries and online product categories. By using the similarity of the embedding vectors, the most similar (strongly correlated) online product categories can be determined for each offline industry. In this way, the purchase information of the online industry can be directly used to obtain the user's interest preference for online products.

[0031] S103, determining neighboring users of the target user according to the embedding vector of each user; the target user is any user among all users.

[0032] Optionally, according to the embedding vector of each user, determining the neighboring users of the target user includes: obtaining the distance between the embedding vector of the target user and the embedding vectors of other users; K users as the neighboring users of the target user.

[0033] The number of neighboring users of the target user can be set according to actual conditions and is not limited in this implementation.

[0034] S104, determining a matching score between the target user and all online product categories based on the historical online product categories to which the products purchased by the target user's neighboring users belong and the target user's embedding vector.

[0035] Optionally, the matching score between the target user and all online product categories is determined based on the historical online product categories to which the products purchased by the target user's neighboring users belong and the target user's embedding vector, including: determining that the initial matching score between the target user and all online product categories is 0; and updating the matching score between the target user and all online product categories based on the distance between the embedding vector of the target user in the historical online product categories and the embedding vector of the neighboring users in the historical online product categories.

[0036] Specifically, the offline industries and online product categories are recorded as and ,in, Indicates offline industries, Indicates the online product category. represents the number of offline industries, Represents the number of online product categories. Based on graph neural network embedding learning, for each user i , and obtain an embedding vector of fixed dimension v i ; For each user i, initialize the user i About the matching scores of all online product categories , and all matching scores are 0. The matching score is used to measure the user's interest preference in a certain online product category. The higher the score, the more suitable it is to recommend this type of product to him.

[0037] Taking user i as the target user, the set of neighboring users of the target user is D i ,for D iEvery user in j , , get user j Online product categories purchased by users j Online product categories purchased as historical online product categories , for each historical online product category, update the matching score between the target user and the historical online product category.

[0038] Specifically, according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the neighboring user in the historical online product category, the matching score between the target user and all online product categories is updated, including: traversing all the neighboring users of the target user, and updating the matching score between the target user and all online product categories according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighboring user in the historical online product category; using the updated matching score as the initial matching score, and continuing to update the matching score between the target user and all online product categories according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the second neighboring user in the historical online product category, until all neighboring users are traversed, and the matching score between the target user and all online product categories is obtained.

[0039] Among them, according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category, the matching score between the target user and all online product categories is updated, including: for any historical online product category of the first neighbor user, according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category, the matching score between the target user and all online product categories except the historical online product categories remains unchanged.

[0040] Specifically, the update calculation formula for the matching score between the target user and the historical online product category is: (1); in, is the updated matching score between the target user and the online product category. is the initial matching score between the target user and the online product category, It is the distance between the embedding vector of the target user in the online product category and the embedding vector of the neighboring user in the online product category.

[0041] S105, migrate offline industries to online product categories through transfer learning, obtain multiple target online product categories with the strongest correlation obtained by mapping the offline industries corresponding to the neighboring users, and update the matching scores between the target users and all online product categories based on the alignment matching scores between the offline industries corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedding vectors.

[0042] for D i Every user in j , get user j Purchased offline industries , for each offline industry, according to The embedding vector of the corresponding node finds the closest one in vector distance M Online product categories, thereby mapping to obtain several target online product categories with the strongest correlation , for each associated target online product category , update the matching scores between target users and all online product categories.

[0043] Specifically, the process of determining multiple target online product categories with the strongest associations obtained by mapping the offline industries corresponding to the neighboring users includes: for any neighboring user, obtaining multiple online related product categories that are closest in distance to the embedding vectors of each historical offline industry in which the neighboring user has purchased products; and determining the multiple online related product categories corresponding to each historical offline industry in which the neighboring user has purchased products as multiple target online product categories.

[0044] It should be noted that the number of target online product categories can be set according to actual needs, and this embodiment does not limit this.

[0045] Optionally, the matching score between the target user and all online product categories is updated according to the alignment matching score between the offline industry corresponding to the neighboring user and the corresponding multiple target online product categories based on the embedded vector, including: traversing all neighboring users of the target user, and updating the matching score between the target user and all online product categories according to the alignment matching score between the offline industry corresponding to the first neighboring user and the corresponding multiple target online product categories based on the embedded vector; using the updated matching score as the initial matching score, and continuing to update the matching score between the target user and all online product categories according to the alignment matching score between the offline industry corresponding to the second neighboring user and the corresponding multiple target online product categories based on the embedded vector, until all neighboring users are traversed.

[0046] Among them, according to the alignment matching score between the offline industry corresponding to the first neighbor user and the corresponding multiple target online product categories based on the embedding vector, the matching score between the target user and all online product categories is updated, including: for any target online product category, according to the alignment matching score between the offline industry corresponding to the first neighbor user and the target online product category based on the embedding vector, and the distance between the embedding vector of the first neighbor user in the target online product category and the embedding vector of the target user in the target online product category, the matching score between the target user and all online product categories except the target online product category remains unchanged.

[0047] The calculation formula for the matching score between the target user and the target online product category is: (2); in, is the updated matching score between the target user and the target online product category. is the initial matching score between the target user and the target online product category, is the distance between the embedding vector of the target user in the target online product category and the embedding vector of the neighboring user in the target online product category, It is the alignment matching score between the offline industry corresponding to the neighboring users and the target online product category based on the embedding vector.

[0048] The alignment matching score between the offline industry corresponding to the user and the target online product category based on the embedding vector can be determined according to the cosine similarity of the embedding vector or other similarity measurement methods.

[0049] Based on the above method, the matching scores between all users and all online product categories are obtained.

[0050] S106, for any online product category, recommend the online product category to multiple users with the highest matching scores.

[0051] In one embodiment, before recommending online product categories to multiple users with the highest matching scores, the embodiment includes: for any user, normalizing the matching scores between the user and all online product categories, and using the normalized matching scores as the matching scores between the user and all online product categories. The calculation formula for normalization can be: (3); in, For users after standardization With The matching scores between online product categories, For users With The matching scores between online product categories, is the number of online product categories.

[0052] The standardized matching score is used as the final matching score between the user and all online product categories. Based on the final matching score, for each online product category, the matching scores can be sorted from bottom to top, and the top products are selected. X Users, recommend this online product category to this X Users can set according to actual needs X The specific value of .

[0053] In one embodiment, Figure 3 As shown, Figure 3 This is a flowchart of a product recommendation method based on graph neural network and transfer learning. First, users can build a multi-mode network based on online transactions and offline shopping, and then embed the multi-mode network through graph neural network to obtain the embedding vector of each user; based on the embedding vector, determine the neighboring users of each user; for any target user, according to the online product categories of the target user and the neighboring users, calculate the matching score between the target user and all online product categories; then through transfer learning, offline consumption information is effectively transferred to online e-commerce recommendations. Specifically, for each offline industry corresponding to the neighboring user, based on the correlation strength between the offline industry and the online product category, the target online product category associated with each offline industry is obtained to update the matching score between the target user and all online product categories, so as to complete the online product recommendation of the e-commerce platform based on the matching score.

[0054] The present invention adopts a product recommendation method based on graph neural network and transfer learning for cross-platform data fusion recommendation, and combines a variety of technical means to ensure that accurate recommendation results can be provided even in a complex data environment. The key technical points include building a multimodal network, learning node embedding vectors, and using transfer learning for knowledge transfer. Specifically, first, a multimodal network containing users, offline industries of consumption, and online product categories is constructed. The network represents the relationship between users and offline industries of consumption, and users and online product categories through edges. Then, a graph neural network (such as the deepwalk model) is used to perform random walks in the network to capture the local neighborhood information of the node and learn the embedding vector of each node. These embedding vectors not only contain the characteristics of the node itself, but also integrate the relationship information of the surrounding nodes, so that the user's behavior pattern in the multimodal network can be fully represented.

[0055] Furthermore, offline consumption information is effectively transferred to online e-commerce recommendations through transfer learning. Specifically, based on the learned embedding vectors, the similarity between the offline industries of consumption and the online product categories is calculated to determine the strength of the association between them. For example, by calculating the cosine similarity between the embedding vectors, it can be found which offline industries are closely associated with which online product categories. When generating recommendations for users, the system not only refers to the user's historical online shopping data, but also comprehensively considers their offline physical consumption. According to the strength of the association between the offline industry and the online product category, the user's matching score is further updated, thereby providing more comprehensive and accurate information support for online e-commerce recommendations and advertising delivery.

[0056] At each step, graph neural networks and transfer learning methods provide a flexible framework that adaptively divides the covariate space by building a tree structure, so that the recommendation results can be locally optimized according to different user behaviors.

[0057] The innovation of this method lies in the combination of graph neural network and transfer learning. By constructing a multimodal network and learning node embedding vectors, the accuracy and personalization of the recommendation system are significantly improved. Graph neural network captures the complex behavior patterns of users in multimodal networks through adaptive structures, so that the recommendation results of each small area can be optimized according to the local characteristics of the data, thus overcoming the dependence of traditional models on single platform data. In addition, transfer learning has the ability to effectively transfer offline consumption information to online e-commerce recommendations. Compared with traditional methods, it can better cope with complex cross-platform data structures, especially in recommendation tasks. Provide more accurate user interest estimation. By calculating node embedding vectors and association strengths, this method uses graph neural networks and transfer learning to accurately estimate multiple parameters, further enhancing the stability and robustness of the model when dealing with multi-platform data fusion.

[0058] The present invention adopts a cross-platform data fusion recommendation model based on graph neural network and transfer learning, combined with multi-mode network construction and node embedding vector learning. In the case of cross-platform data fusion, even if some model assumptions are not true, the recommendation results can still maintain accuracy and personalization. The recommendation effect of this method is also enhanced, reducing the dependence on single platform data, especially suitable for complex data environments, overcoming the shortcomings of the existing technology.

[0059] When applying the product recommendation method based on graph neural network and transfer learning provided by the present invention, it is not necessary to Figure 1 The steps are executed in the order shown. The specific execution order of the steps can be determined according to needs, and the present invention does not limit this.

[0060] The above is a product recommendation method based on graph neural network and transfer learning provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding product recommendation device based on graph neural network and transfer learning, such as Figure 4 shown.

[0061] Figure 4 A schematic diagram of a product recommendation device based on graph neural network and transfer learning provided by the present invention, the device 400 includes: Construction module 401 is used to construct a multi-mode network of users, offline industries and online product categories based on offline transaction and online shopping data; the multi-mode network includes the association relationship between multiple users and the offline industries to which the products purchased offline belong, and each user and the online product category to which the products purchased online belong.

[0062] The learning module 402 is used to perform embedding learning on the multi-mode network according to the graph neural network to obtain the embedding vector of each user; the user's embedding vector includes the embedding vector of the user in each offline industry and each online product category.

[0063] The first determination module 403 is used to determine the neighboring users of the target user according to the embedding vector of each user; the target user is any user among all the users.

[0064] The second determination module 404 is used to determine the matching scores between the target user and all online product categories according to the historical online product categories to which the products purchased by the neighboring users of the target user belong and the embedding vector of the target user.

[0065] Update module 405 is used to migrate offline industries to online product categories through transfer learning, obtain multiple target online product categories with the strongest association obtained by mapping the offline industries corresponding to the neighboring users, and update the matching scores between the target users and all online product categories based on the alignment matching scores between the offline industries corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedded vectors.

[0066] The recommendation module 406 is used to recommend any online product category to multiple users with the highest matching scores.

[0067] For the specific limitations of the product recommendation device based on graph neural networks and transfer learning, please refer to the limitations of the product recommendation method based on graph neural networks and transfer learning above, which will not be repeated here. Each module in the above-mentioned product recommendation device based on graph neural networks and transfer learning can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0068] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 The product recommendation method provided is based on graph neural network and transfer learning.

[0069] The present invention also provides Figure 5 The structural diagram of the computer device shown in FIG. Figure 5 As shown in the figure, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The product recommendation method provided is based on graph neural network and transfer learning.

[0070] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0071] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0072] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A product recommendation method based on graph neural network and transfer learning, characterized in that: include: Based on offline transaction and online shopping data, a multi-mode network of users, offline industries and online product categories is constructed; the multi-mode network includes the association relationship between multiple users and the offline industries to which the products purchased offline belong, and each user and the online product category to which the products purchased online belong; The multi-mode network is embedded and learned based on the graph neural network to obtain the embedding vector of each user; the user's embedding vector includes the user's embedding vector in each offline industry and each online product category; Determine the neighboring users of the target user according to the embedding vector of each user; the target user is any user among all users; Determine a matching score between the target user and all online product categories based on the historical online product categories to which the products purchased by the neighboring users of the target user belong and the embedding vector of the target user; Migrate offline industries to online product categories through transfer learning to obtain multiple target online product categories with the strongest associations obtained by mapping the offline industries corresponding to the neighboring users, and update the matching scores between the target user and all online product categories based on the alignment matching scores between the offline industries corresponding to the neighboring users and the corresponding multiple target online product categories based on the embedding vectors; For any online product category, the online product category is recommended to multiple users with the highest matching scores.

2. The method according to claim 1, characterized in that The step of determining the neighboring users of the target user according to the embedding vector of each user includes: Obtaining the distance between the embedding vector of the target user and the embedding vectors of other users; The K users whose embedding vectors are closest to the target user are regarded as the neighboring users of the target user.

3. The method according to claim 1, characterized in that The determining, based on the historical online product categories to which the products purchased by the neighboring users of the target user belong and the embedding vector of the target user, a matching score between the target user and all online product categories includes: Determine that the initial matching score between the target user and all online product categories is 0; According to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the neighboring user in the historical online product category, the matching score between the target user and all online product categories is updated.

4. The method according to claim 3, characterized in that The updating of the matching scores between the target user and all online product categories according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the neighboring user in the historical online product category includes: Traversing all neighboring users of the target user, and updating the matching scores between the target user and all online product categories according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighboring user in the historical online product category; The updated matching score is used as the initial matching score. According to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the second neighbor user in the historical online product category, the matching score between the target user and all online product categories is continued to be updated until all neighbor users are traversed, thereby obtaining the matching score between the target user and all online product categories.

5. The method according to claim 4, characterized in that The updating of the matching scores between the target user and all online product categories according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category includes: For any historical online product category of the first neighbor user, updating the matching score between the target user and the historical online product category according to the distance between the embedding vector of the target user in the historical online product category and the embedding vector of the first neighbor user in the historical online product category; The matching scores between the target user and other online product categories except the historical online product categories remain unchanged; the updated calculation formula for the matching scores between the target user and the historical online product categories is: ； in, is the updated matching score between the target user and the online product category. is the initial matching score between the target user and the online product category, It is the distance between the embedding vector of the target user in the online product category and the embedding vector of the neighboring user in the online product category.

6. The method according to claim 1, characterized in that The process of determining the multiple target online product categories with the strongest association obtained by mapping the offline industries corresponding to the neighboring users includes: For any neighbor user, obtain multiple online related product categories that are closest to the embedding vector of each historical offline industry that the neighbor user has purchased products from; Multiple online related product categories corresponding to each historical offline industry in which the neighboring user has purchased products are determined as multiple target online product categories.

7. The method according to claim 1, characterized in that The updating of the matching scores between the target user and all online product categories according to the alignment matching scores between the offline industry corresponding to the neighboring user and the corresponding multiple target online product categories based on the embedded vectors includes: Traversing all neighboring users of the target user, and updating the matching scores between the target user and all online product categories according to the alignment matching scores between the offline industry corresponding to the first neighboring user and the corresponding multiple target online product categories based on the embedding vectors; The updated matching score is used as the initial matching score, and the matching scores between the target user and all online product categories are continuously updated according to the alignment matching scores between the offline industry corresponding to the second neighbor user and the corresponding multiple target online product categories based on the embedded vectors, until all neighbor users are traversed.

8. The method according to claim 7, characterized in that The updating of the matching scores between the target user and all online product categories according to the alignment matching scores between the offline industry corresponding to the first neighbor user and the corresponding multiple target online product categories based on the embedding vectors includes: For any target online product category, update the matching score between the target user and the target online product category based on the alignment matching score between the offline industry corresponding to the first neighbor user and the target online product category based on the embedding vector, and the distance between the embedding vector of the first neighbor user in the target online product category and the embedding vector of the target user in the target online product category; The matching scores between the target user and other online product categories except the target online product category among all online product categories remain unchanged; the calculation formula for the matching score between the target user and the target online product category is: ； in, is the updated matching score between the target user and the target online product category. is the initial matching score between the target user and the target online product category, is the distance between the embedding vector of the target user in the target online product category and the embedding vector of the neighboring user in the target online product category, It is the alignment matching score between the offline industry corresponding to the neighboring users and the target online product category based on the embedding vector.

9. The method according to claim 1, characterized in that: Before recommending the online product category to the plurality of users with the highest matching scores, the method further includes: For any user, the matching scores between the user and all online product categories are normalized, and the normalized matching scores are used as the matching scores between the user and all online product categories.

Images