Cross-domain data recommendation method, device, electronic device and storage medium

By identifying potential overlapping users and establishing implicit bridges, inter-domain knowledge migration is carried out, and the accuracy of cross-domain data recommendation in the existing technology is solved, achieving efficient data recommendation between non-overlapping domains.

CN117874355BActive Publication Date: 2024-11-29BEIJING JINFENG-BI SOFTWARE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410175562.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-07
Publication Date
2024-11-29
Estimated Expiration
2044-02-07

AI Technical Summary

Technical Problem

Existing recommendation systems are difficult to accurately recommend data between different domains, especially objects without overlapping data, because cross-domain data comparison and recommendation cannot be effectively performed.

Method used

By determining the source and target domains of the user-item two-part graph, the graph convolution model is used to identify potential overlapping users, and inter-domain knowledge migration is carried out based on implicit bridges to achieve mutual mapping of user features, thereby between non-overlapping domains. Make data recommendations.

Benefits of technology

It improves the accuracy of cross-domain data recommendations, enables two-way knowledge migration between different domains, and enhances the accuracy and user experience of recommendation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117874355B_ABST
    Figure CN117874355B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a cross-domain data recommendation method, device, electronic device and storage medium, which determine a source domain and a target domain respectively including a user-item bipartite graph, and there are no overlapping users or items between the two different domains. The source domain and the target domain are respectively input into the trained graph convolution model, and potential overlapping users are obtained through graph convolution. An implicit bridge between the source domain and the target domain is determined based on the potential overlapping users, and inter-domain knowledge transfer is performed between the source domain and the target domain based on the implicit bridge to achieve mutual mapping of user features. Finally, the items corresponding to the users in the source domain in the target domain and the items corresponding to the users in the target domain in the source domain are determined based on the user features after the inter-domain knowledge transfer. The present disclosure improves the accuracy of the data recommendation results by building an implicit bridge between domains without overlapping data for bidirectional knowledge transfer, and realizing cross-domain data recommendation based on the features after the knowledge transfer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a cross-domain data recommendation method, device, electronic device and storage medium. Background Art

[0002] With the explosive development of global information technologies such as big data, the Internet of Things, and cloud computing in recent years, we are entering an era of exponential information growth. The amount of information that users can get on the Internet is exploding, and it is difficult for users to find valuable information among the many messages. It may be difficult for information producers to display the information they produce to users who are interested in it. This is the problem of information overload. Recommendation systems provide a solution to the problem of information overload, but current recommendation systems can only make relatively accurate data recommendations for objects with overlapping data in the same domain, and it is difficult to make accurate data recommendations based on objects that do not have overlapping data between different domains. Summary of the invention

[0003] In view of this, the present disclosure proposes a cross-domain data recommendation method, device, electronic device and storage medium, aiming to perform cross-domain data recommendation for objects with no overlapping data.

[0004] According to a first aspect of the present disclosure, a cross-domain data recommendation method is provided, the method comprising:

[0005] Determine a source domain and a target domain respectively including a user-item bipartite graph, wherein there are no overlapping users or items in the source domain and the target domain;

[0006] Inputting the source domain and the target domain into the trained graph convolution model respectively, and obtaining potential overlapping users in the source domain and the target domain through graph convolution;

[0007] determining an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain, wherein the implicit bridge represents a plurality of potential user pairs formed by potential overlapping users between the source domain and the target domain;

[0008] Based on the implicit bridge, inter-domain knowledge transfer is performed between the source domain and the target domain to achieve mutual mapping of user features between the source domain and the target domain;

[0009] The items corresponding to the users in the source domain in the source domain are determined according to the user characteristics after the inter-domain knowledge transfer, and the items corresponding to the users in the target domain in the target domain are determined.

[0010] In a possible implementation manner, inputting the source domain and the target domain into the trained graph convolution model respectively, and obtaining potential overlapping users in the source domain and the target domain through graph convolution, includes:

[0011] Performing multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain respectively to obtain a graph convolution result corresponding to each graph convolution, wherein the graph convolution result includes a user feature vector of each user and an item feature vector of each item;

[0012] Classifying users in the source domain and the target domain respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each subgraph having a corresponding category;

[0013] Users in the subgraphs of the same category in the source domain and the target domain are determined as potential overlapping users.

[0014] In a possible implementation, the user in the source domain and the target domain are classified respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, including:

[0015] The user initial embedding vector obtained by encoding the user feature vectors in the source domain and the target domain through the embedding layer and the user feature vector obtained after the first graph convolution on the user-item bipartite graph are input into the classifier to obtain a preset number of subgraphs and the category corresponding to each of the subgraphs.

[0016] In a possible implementation, the classifier is:

[0017]

[0018] U o =W2U h +b2

[0019] Among them, W1 and W2 are the weight matrices of the classifier, b1 and b2 are the bias vectors of the classifier, σ(·) is the nonlinear activation function LeakyReLU, U o is the predicted category, is the initial embedding vector of user u, is the user feature vector corresponding to user u in the first graph convolution result;

[0020] The method further comprises: classifying users in the source domain and the target domain based on the graph convolution result by using an unsupervised subgraph classifier to obtain a preset number of subgraphs;

[0021] The InfoNCE loss of the classification result is calculated, and the weight matrix and bias vector in the classifier are adjusted.

[0022] In a possible implementation manner, determining an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain includes:

[0023] creating an implicit bridge between the source domain and the target domain according to the potential overlapping users;

[0024] Calculating similarities of potential overlapping users in the source domain and the target domain;

[0025] The implicit bridge is modified according to the similarity.

[0026] In a possible implementation manner, the calculating the similarity of potential overlapping users in the source domain and the target domain includes:

[0027] pass Calculate the similarity of potential overlapping users in the source domain and the target domain, where u is the user feature vector of the user in the source domain, u′ is the user feature vector of the user in the target domain, l is the number of graph convolution results, L is the total number of graph convolution layers, is the user feature vector obtained by the l-th graph convolution.

[0028] In a possible implementation manner, the modifying the implicit bridge according to the similarity includes:

[0029] Deleting irrelevant bridges corresponding to potential overlapping users in the source domain and potential overlapping users in the target domain according to the similarity, wherein the irrelevant bridges are potential overlapping user pairs whose similarity is lower than a preset threshold;

[0030] A preset number of potential overlapping users in the target domain with the highest similarity to users in the source domain and a preset number of potential overlapping users in the source domain with the highest similarity to users in the target domain are retained.

[0031] In a possible implementation, the potential overlapping users are (u s ,u t ),u s For users in the source domain, u t is a user in the target domain, and the implicit bridge is , i is the item corresponding to the user;

[0032] The deleting irrelevant bridges corresponding to the potential overlapping users in the source domain and the potential overlapping users in the target domain according to the similarity includes:

[0033] Determine the implicit bridge from the target domain to the source as Determine the implicit bridge from the source domain to the target domain as

[0034] In a possible implementation, retaining a preset number of potential overlapping users in the target domain having the highest similarity with users in the source domain, and a preset number of potential overlapping users in the source domain having the highest similarity with users in the target domain, includes:

[0035] Determine the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as k is the preset number,

[0036] In a possible implementation, the performing inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge to achieve mutual mapping of user features between the source domain and the target domain includes:

[0037] Performing multiple graph convolutions on each type of subgraph in the source domain and the target domain respectively, to obtain a subgraph convolution result corresponding to each graph convolution, wherein the subgraph convolution result includes a user feature vector corresponding to each user and an item feature vector corresponding to each item;

[0038] Performing layer fusion on the user feature vector of each of the potential overlapping users in the source domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain a source domain user vector of each of the potential overlapping users in the source domain;

[0039] Performing layer fusion on the user feature vector of each of the potential overlapping users in the target domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain the target domain user vector of each of the potential overlapping users in the target domain;

[0040] By formula and Perform inter-domain knowledge transfer between the source domain and the target domain respectively, wherein: f(·) is the similarity, λ is the preset hyperparameter, W tr is a preset conversion matrix used to map the user features of the target domain to the source domain, u s is the user feature vector of potential overlapping users in the source domain, u t is the user feature vector of potential overlapping users in the target domain, l is the number of graph convolution results, For user u s The corresponding source domain user vector, For user u t The corresponding target domain user vector.

[0041] In a possible implementation, determining the items corresponding to the users in the source domain in the source domain according to the user features after the inter-domain knowledge transfer, and determining the items corresponding to the users in the target domain in the target domain, includes:

[0042] For each potential overlapping user in the source domain and the target domain, the user feature vector after the inter-domain knowledge transfer between the source domain and the target domain is Performing sub-graph fusion on each of the sub-graph convolution results corresponding to the corresponding graph convolution layer, and for non-potentially overlapping users and each item in the source domain and the target domain, performing sub-graph fusion on the user / item feature vector in each sub-graph and the user / item feature vector in each sub-graph convolution result to obtain a first fusion result;

[0043] Respectively fusing the first fusion result, the initial embedding vector, and each of the graph convolution results corresponding to the source domain and the target domain to obtain a second fusion result;

[0044] Determine the items in the source domain corresponding to the potential overlapping users in the source domain according to the feature vector corresponding to each of the users in the second fusion result corresponding to the source domain and the feature vector of each of the items;

[0045] According to the feature vector corresponding to each of the users in the second fusion result corresponding to the target domain and the feature vector of each of the items, the items corresponding to the potential overlapping users in the target domain in the target domain are determined.

[0046] According to a second aspect of the present disclosure, a cross-domain data recommendation device is provided, the device comprising:

[0047] A domain determination module, used to determine a source domain and a target domain respectively including a user-item bipartite graph, wherein there are no overlapping users or items in the source domain and the target domain;

[0048] A graph convolution module, used to input the source domain and the target domain into the trained graph convolution model respectively, and obtain potential overlapping users in the source domain and the target domain through graph convolution;

[0049] a bridge building module, configured to determine an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain;

[0050] A knowledge transfer module, used for performing inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge, so as to realize mutual mapping of user features between the source domain and the target domain;

[0051] The data recommendation module is used to determine the items corresponding to the users in the source domain in the source domain according to the user characteristics after the inter-domain knowledge transfer, and to determine the items corresponding to the users in the target domain in the target domain.

[0052] In a possible implementation, the graph convolution module is further used to:

[0053] Performing multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain respectively to obtain a graph convolution result corresponding to each graph convolution, wherein the graph convolution result includes a user feature vector of each user and an item feature vector of each item;

[0054] Classifying users in the source domain and the target domain respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each subgraph having a corresponding category;

[0055] Users in the subgraphs of the same category in the source domain and the target domain are determined as potential overlapping users.

[0056] In a possible implementation, the graph convolution module is further used to:

[0057] The user initial embedding vector obtained by encoding the user feature vectors in the source domain and the target domain through the embedding layer and the user feature vector obtained after the first graph convolution on the user-item bipartite graph are input into the classifier to obtain a preset number of subgraphs and the category corresponding to each of the subgraphs.

[0058] In a possible implementation, the classifier is:

[0059]

[0060] U o =W2U h +b2

[0061] Among them, W1 and W2 are the weight matrices of the classifier, b1 and b2 are the bias vectors of the classifier, σ(·) is the nonlinear activation function LeakyReLU, U o is the predicted category, is the initial embedding vector of user u, is the user feature vector corresponding to user u in the first graph convolution result;

[0062] The graph convolution module is further used to:

[0063] The InfoNCE loss of the classification result is calculated, and the weight matrix and bias vector in the classifier are adjusted.

[0064] In a possible implementation, the bridge construction module is further used to:

[0065] creating an implicit bridge between the source domain and the target domain according to the potential overlapping users;

[0066] Calculating similarities of potential overlapping users in the source domain and the target domain;

[0067] The implicit bridge is modified according to the similarity.

[0068] In a possible implementation, the bridge construction module is further used to:

[0069] pass Calculate the similarity of potential overlapping users in the source domain and the target domain, where u is the user feature vector of the user in the source domain, u′ is the user feature vector of the user in the target domain, l is the number of graph convolution results, L is the total number of graph convolution layers, is the user feature vector obtained by the l-th graph convolution.

[0070] In a possible implementation, the bridge construction module is further used to:

[0071] Deleting irrelevant bridges corresponding to potential overlapping users in the source domain and potential overlapping users in the target domain according to the similarity, wherein the irrelevant bridges are potential overlapping user pairs whose similarity is lower than a preset threshold;

[0072] A preset number of potential overlapping users in the target domain with the highest similarity to users in the source domain and a preset number of potential overlapping users in the source domain with the highest similarity to users in the target domain are retained.

[0073] In a possible implementation, the potential overlapping users are (u s ,u t ),u s For users in the source domain, u t is a user in the target domain, and the implicit bridge is i is the item corresponding to the user;

[0074] The bridge building module is further used to:

[0075] Determine the implicit bridge from the target domain to the source as Determine the implicit bridge from the source domain to the target domain as

[0076] In a possible implementation, the bridge construction module is further used to:

[0077] Determine the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as k is the preset number,

[0078] In a possible implementation, the knowledge transfer module is further used to:

[0079] Performing multiple graph convolutions on each type of subgraph in the source domain and the target domain respectively, to obtain a subgraph convolution result corresponding to each graph convolution, wherein the subgraph convolution result includes a user feature vector corresponding to each user and an item feature vector corresponding to each item;

[0080] Performing layer fusion on the user feature vector of each of the potential overlapping users in the source domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain a source domain user vector of each of the potential overlapping users in the source domain;

[0081] Performing layer fusion on the user feature vector of each of the potential overlapping users in the target domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain the target domain user vector of each of the potential overlapping users in the target domain;

[0082] By formula and Perform inter-domain knowledge transfer between the source domain and the target domain respectively, wherein: f(·) is the similarity, λ is the preset hyperparameter, W tr is a preset conversion matrix used to map the user features of the target domain to the source domain, u s is the user feature vector of potential overlapping users in the source domain, u t is the user feature vector of potential overlapping users in the target domain, l is the number of graph convolution results, For user u s The corresponding source domain user vector, For user u t The corresponding target domain user vector.

[0083] In a possible implementation, the data recommendation module is further used to:

[0084] For each potential overlapping user in the source domain and the target domain, the user feature vector after the inter-domain knowledge transfer between the source domain and the target domain is Performing sub-graph fusion on each of the sub-graph convolution results corresponding to the corresponding graph convolution layer, and for non-potentially overlapping users and each item in the source domain and the target domain, performing sub-graph fusion on the user / item feature vector in each sub-graph and the user / item feature vector in each sub-graph convolution result to obtain a first fusion result;

[0085] Respectively fusing the first fusion result, the initial embedding vector, and each of the graph convolution results corresponding to the source domain and the target domain to obtain a second fusion result;

[0086] Determine the items in the source domain corresponding to the potential overlapping users in the source domain according to the feature vector corresponding to each of the users in the second fusion result corresponding to the source domain and the feature vector of each of the items;

[0087] According to the feature vector corresponding to each of the users in the second fusion result corresponding to the target domain and the feature vector of each of the items, the items corresponding to the potential overlapping users in the target domain in the target domain are determined.

[0088] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0089] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions implement the above method when executed by a processor.

[0090] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0091] In an embodiment of the present disclosure, a source domain and a target domain are determined, each including a user-item bipartite graph, and there are no overlapping users or items between the two different domains. The source domain and the target domain are respectively input into the trained graph convolution model, and potential overlapping users are obtained through graph convolution. An implicit bridge between the source domain and the target domain is determined based on the potential overlapping users, and inter-domain knowledge transfer is performed between the source domain and the target domain based on the implicit bridge to achieve mutual mapping of user features. Finally, based on the user features after the inter-domain knowledge transfer, the items corresponding to the users in the source domain in the target domain, as well as the items corresponding to the users in the target domain in the source domain, are determined. The present disclosure improves the accuracy of data recommendation results by building an implicit bridge between domains without overlapping data for bidirectional knowledge transfer, and realizing cross-domain data recommendation based on the features after knowledge transfer.

[0092] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0094] Figure 1 A flow chart showing a cross-domain data recommendation method according to an embodiment of the present disclosure;

[0095] Figure 2 A schematic diagram showing a cross-domain data recommendation process according to an embodiment of the present disclosure;

[0096] Figure 3 A schematic diagram showing a cross-domain data recommendation device according to an embodiment of the present disclosure;

[0097] Figure 4 A schematic diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0098] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0099] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0100] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0101] The cross-domain data recommendation method of the embodiment of the present disclosure can be performed by an electronic device such as a terminal device or a server. Among them, the terminal device can be any fixed or mobile terminal such as a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The server can be a single server or a server cluster composed of multiple servers. Any electronic device can implement the cross-domain data recommendation method of the embodiment of the present disclosure by a processor calling a computer-readable instruction stored in a memory.

[0102] Figure 1 A flow chart of a cross-domain data recommendation method according to an embodiment of the present disclosure is shown. Figure 1 As shown, the cross-domain data recommendation method of the embodiment of the present disclosure may include the following steps S10-S50.

[0103] Step S10: determine a source domain and a target domain which respectively include a user-item bipartite graph.

[0104] In a possible implementation, the source domain and the target domain can be determined by an electronic device, and the source domain and the target domain respectively include a user-item bipartite graph representing the corresponding relationship between users and items in the domain, which is used to characterize the association relationship between users and items in the domain. "Domain" can represent a collection of data with the same data type. Among them, the user-item bipartite graph can be predetermined based on the user's historical behavior data. For example, for an e-commerce platform, the user-item bipartite graph can be determined based on the user's historical purchase information. For an information search platform, the user-item bipartite graph can be determined based on the user's historical search record for items. Optionally, there are no overlapping users or items in the source domain and the target domain. That is, each user included in the source domain is not in the target domain, and there is no corresponding item in the target domain. At the same time, each user included in the target domain is not in the source domain, and there is no corresponding item in the source domain. Among them, users in the source domain and the target domain can be represented by user feature vectors, and items can be represented by item feature vectors.

[0105] Exemplarily, for an e-commerce shopping platform, the source domain and the target domain can be determined based on the users who purchase items of different categories. For example, the source domain and the target domain can correspond to the electrical goods category and the beauty and cosmetics category, respectively, wherein the user-item bipartite graph included in the source domain is used to characterize the relationship between the electrical goods category users who have not purchased beauty and cosmetics category items among the users who purchase electrical goods category items, and the electrical goods category items purchased by the electrical goods category users. The user-item bipartite graph included in the target domain is used to characterize the relationship between the beauty and cosmetics category users who have not purchased electrical goods category items among the users who purchase beauty and cosmetics category items, and the beauty and cosmetics category items purchased by the beauty and cosmetics category users.

[0106] Step S20: input the source domain and the target domain into the trained graph convolution model respectively, and obtain potential overlapping users in the source domain and the target domain through graph convolution.

[0107] In a possible implementation, the electronic device can identify potential overlapping users in the source domain and the target domain, that is, users in the source domain and the target domain who may have similar needs or behaviors, by performing graph convolution on the user-item bipartite graphs in the source domain and the target domain respectively through a pre-trained graph convolution model. The graph convolution process may include first convolving the user-item bipartite graph in each domain as a whole, then dividing the user-item bipartite graph in each domain into subgraphs corresponding to different feature dimensions to identify potential overlapping users, and further performing graph convolution on subgraphs of different categories to further perform cross-domain knowledge transfer.

[0108] Specifically, the electronic device can first perform multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain, respectively, to obtain the graph convolution results corresponding to each graph convolution. The graph convolution results include the user feature vector of each user and the item feature vector of each item. Then, the users in the source domain and the target domain are classified based on the graph convolution results by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each of which has a corresponding category. Users in the subgraphs of the same category in the source domain and the target domain are determined to be potential overlapping users. Then, multiple graph convolutions are performed on each type of subgraph in the source domain and the target domain, respectively, to obtain the subgraph convolution results corresponding to each graph convolution. Optionally, the user-item bipartite graph in each domain After inputting the graph convolution model, the initial embedding vector corresponding to each user u and item i is first obtained through the embedding layer encoding. and Where d is the dimension of the embedding vector. Then, the user-item bipartite graph in the domain Perform multiple convolutions to obtain the graph convolution results after each convolution. The graph convolution results include the user feature vector of each user obtained by convolving the user's embedding vector and the item feature vector of each item obtained by convolving the item's embedding vector. The user feature vector in the graph convolution result corresponding to the first convolution layer can be obtained by the formula Determine that the item feature vector can be obtained by the formula Sure, and For user u and item i in the figure The first-order neighbors on .

[0109] Furthermore, in order to prevent the introduction of noise in high-order graph convolution, the electronic device can divide the user-item bipartite graph in each domain based on different dimensions such as user interests to construct multiple subgraphs, so that only the information of similar neighbor nodes can be aggregated, further improving the accuracy of the determined potential overlapping users. Since the potential overlapping user group reflects the common parts of the source domain and the target domain, and the specific user group in the domain reflects the difference between the two domains, the embodiment of the present disclosure classifies the potential overlapping users through an unsupervised subgraph classifier, and the subgraph generation result directly affects the extraction of potential overlapping users.

[0110] Optionally, the initial embedding vectors of users in the source domain and the target domain and the feature vector obtained after the first graph convolution can be input into the classifier to obtain a preset number of subgraphs and the category corresponding to each subgraph. The user feature vector of the user and the item feature vector of the corresponding item can be respectively and l+1 is the corresponding convolutional layer number, on the subgraph The node after the item passes through the l-layer convolution layer for graph convolution can be

[0111] The unsupervised subgraph classifier of the embodiment of the present disclosure may be a two-layer classifier including two layers of neural networks, wherein the two layers of neural networks are respectively U o =W2U h +b2, and are the weight matrices of the two-layer classifiers, and are the bias vectors of the two-layer classifiers, σ(·) is the nonlinear activation function LeakyReLU, is the predicted category, is the initial embedding vector of user u, is the user feature vector corresponding to user u in the first graph convolution result. Based on the two-layer classifier, the electronic device can obtain a preset number of subgraphs and the category corresponding to each subgraph. Among them, the users in each subgraph are a group of similar users, and the electronic device can determine that the users in the subgraphs corresponding to the same category in different domains are potential overlapping users.

[0112] Optionally, in order to further improve the accuracy of the classification result, the electronic device may adjust the weight matrix and the bias vector in the classifier by calculating the loss of the classification result output by the classifier. The loss of the classification result may be the InfoNCE loss, which is calculated by the formula Calculated. Among them, and For positive samples, the users in the source domain and the target domain (u s ,u t )Contrastive loss in both directions. is the set of potential overlapping users in the source domain, is a set of potential overlapping users in the target domain, τ is a temperature parameter used to control the contribution of negative sampling nodes, and f(u s ,u t ) is user u in the source domain s With user u in the target domain t , N is the number of sub-graphs.

[0113] In a possible implementation, the similarity in the embodiment of the present disclosure can be determined based on the preference differences between different layers, so as to highlight the specificity between the layers and enhance the ability of the graph convolution model to perceive the subtle differences between different users. Calculate the similarity between two users in the source domain and the target domain, where u is the user feature vector of the user in the source domain, and u′ is the user feature vector of the user in the target domain. l is the number of graph convolution results, L is the total number of graph convolution layers, is the user feature vector obtained by the l-th graph convolution.

[0114] Furthermore, after determining the potential overlapping users, the electronic device performs multiple graph convolutions on each type of subgraph in the source domain and the target domain respectively, and obtains the corresponding subgraph convolution results after each graph convolution, which is used for later inter-domain knowledge transfer between the target domain and the source domain.

[0115] Step S30: determining an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain.

[0116] In a possible implementation, after determining the potential overlapping users in the source domain and the target domain, the electronic device can create an implicit bridge between the source domain and the target domain based on the potential overlapping users. Then, the similarity of the potential overlapping users in the source domain and the target domain is calculated, and then the implicit bridge between the source domain and the target domain is corrected based on the similarity. The similarity of the potential overlapping users in the source domain and the target domain can also be calculated by Calculated, u is the user in the source domain, u′ is the user in the target domain, l is the number of graph convolution results, L is the total number of graph convolution layers, is the user feature vector obtained by the l-th graph convolution.

[0117] Optionally, according to the potential overlapping users (u s ,u t )Build an implicit bridge i is the item corresponding to the user, u s For users in the source domain, u t For users in the target domain, after calculating the similarity between the potential overlapping users in the source domain and the target domain, irrelevant bridges corresponding to the potential overlapping users in the source domain and the potential overlapping users in the target domain can be deleted according to the similarity. Then, a preset number of potential overlapping users in the target domain with the highest similarity to the users in the source domain and a preset number of potential overlapping users in the source domain with the highest similarity to the users in the target domain are retained to correct the implicit bridges. Irrelevant bridges are potential overlapping user pairs whose similarity is lower than a preset threshold.

[0118] The process of removing irrelevant bridges can be achieved by determining the implicit bridges from the target domain to the source as Determine the implicit bridge from the source domain to the target domain as The process of retaining a preset number of potential users with the highest similarity between the source domain and the target domain can be achieved by determining the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as Implementation. Where j is the preset number, Represents from the set The j highest values ​​obtained in .

[0119] Step S40: performing inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge to achieve mutual mapping of user features between the source domain and the target domain.

[0120] In a possible implementation, after the implicit bridge between the source domain and the target domain is determined and corrected, inter-domain knowledge transfer can be performed between the source domain and the target domain based on the implicit bridge to map the user features of potential overlapping users in the source domain to the target domain, and map the user features of potential overlapping users in the target domain to the source domain, so as to achieve mutual mapping of user features between the source domain and the target domain. Optionally, the basis for inter-domain knowledge transfer between the source domain and the target domain is the result obtained by layer fusion of multiple subgraphs of the source domain and the corresponding subgraph convolution results, and the result obtained by layer fusion of multiple subgraphs of the target domain and the corresponding subgraph convolution results. That is, the electronic device can first layer-fuse the user feature vectors of the potential overlapping users in each subgraph in the source domain with the user feature vectors in the corresponding multiple subgraph convolution results, that is, fuse the user feature vectors corresponding to different convolution layers, and obtain the source domain user vector of each potential overlapping user, and layer-fuse the user feature vectors of the potential overlapping users in each subgraph in the target domain with the user feature vectors in the corresponding multiple subgraph convolution results, and obtain the target domain user vector. Furthermore, inter-domain knowledge transfer is performed based on the source domain fusion information and the target domain fusion information.

[0121] Alternatively, the formula and The inter-domain knowledge transfer between the source domain and the target domain is performed respectively, where f(·) is the similarity, λ is the preset hyperparameter, W tr is the preset transformation matrix used to map the user features of the target domain to the source domain, u s are potential overlapping users in the source domain, u t The source domain user u s Potential overlapping users in the target domain, l is the number of graph convolution results, The source domain user u s The user vector of For the target domain user u t This method of knowledge transfer between domains can reduce the difficulty of model training and prevent the model from being over-parameterized. At the same time, it can also establish a more general and stable bidirectional mapping relationship between domains to achieve accurate feature mapping.

[0122] Step S50: determining the items corresponding to the users in the source domain in the source domain according to the user characteristics after the inter-domain knowledge transfer, and determining the items corresponding to the users in the target domain in the target domain.

[0123] In a possible implementation, after completing the bidirectional inter-domain knowledge transfer, the users in the source domain and the target domain realize feature fusion, and the electronic device can further realize the fusion of all features of each potential overlapping user and item in the domain by performing subgraph fusion and feature fusion on different subgraphs, subgraph convolution results, initial embedding vectors, and graph convolution results in each domain, and then search for items corresponding to the user in the domain based on the user feature vector of each user obtained after fusion. That is, the corresponding items can be searched in the source domain according to the user feature vector of each user obtained after feature fusion in the source domain, and the corresponding items can be searched in the target domain according to the user feature vector of each user obtained after feature fusion in the target domain.

[0124] Optionally, for each potential overlapping user in the source domain and the target domain, the user feature vector after the inter-domain knowledge transfer between the source domain and the target domain is obtained. Subgraph fusion is performed on each of the subgraph convolution results corresponding to the corresponding graph convolution layer. For non-potentially overlapping users and each item in the source domain and the target domain, subgraph fusion is performed on the user / item feature vector in each subgraph and the user / item feature vector in each subgraph convolution result to obtain a first fusion result. For each potential overlapping user in the source domain and the target domain, the electronic device can perform subgraph fusion on the user feature vector after knowledge transfer. The user feature vectors in each subgraph convolution result and the user feature vectors in each subgraph are fused by calculating the weighted sum. For example, for a potential overlapping user in the source domain, the user feature vectors in the same layer of knowledge transfer, the user feature vectors in the subgraph convolution result, and the user feature vectors in the subgraph are first weighted summed, and then the weighted sum results of different layers are weighted summed again to complete the subgraph fusion for the user. For non-potential overlapping users and each item in the source domain and the target domain, the electronic device can perform subgraph fusion on the user / item feature vectors in each subgraph and the user / item feature vectors in each subgraph convolution result by calculating the weighted sum.

[0125] For example, for a non-potentially overlapping user, its user feature vector in the subgraph can be weighted and summed with the user feature vector in the subgraph convolution results of each layer, and then the weighted sum of the weighted sum results of each layer can be weighted and summed. That is, the first fusion result obtained includes the feature vector corresponding to each user (including non-potentially overlapping users and potential overlapping users) and the item subgraph in the source domain and the target domain after fusion. Then, the first fusion result corresponding to the source domain and the target domain, the initial embedding vector, and each graph convolution result are layer-fused to obtain a second fusion result. For example, for each user in the source domain and the target domain, the electronic device can complete the layer fusion by calculating the weighted sum of the corresponding user feature vector in the first fusion result, the initial embedding vector, and the corresponding user feature vector in each graph convolution result, and then perform weighted summation between the weighted sums of each layer to obtain the second fusion result of the user. For each item in the source domain and the target domain, the electronic device can calculate the weighted sum to complete the layer fusion through the item feature vector corresponding to the first fusion result, the initial embedding vector, and the item feature vector corresponding to each graph convolution result, and then perform the weighted sum between the weighted sums of each layer to obtain the second fusion result of the item.

[0126] After obtaining the second fusion result of each domain, the item corresponding to the user in the source domain is determined according to the feature vector corresponding to each user in the second fusion result corresponding to the source domain and the feature vector of each item in the second fusion result corresponding to the source domain. The item corresponding to the user in the target domain is determined according to the feature vector corresponding to each user in the second fusion result corresponding to the target domain and the feature vector of each item in the second fusion result corresponding to the target domain.

[0127] Specifically, for each user's corresponding item in the same domain, the preference scores of the feature vector of the user in the second fusion result and the feature vector of each item in the same domain in the second fusion result can be calculated, and then the item corresponding to the user can be selected from multiple items according to the preference scores. For example, multiple items with the highest preference scores can be selected as the items corresponding to each user. The preference scores can be obtained by the formula Calculated, where e i When e is the feature vector of the item in the source domain in the second fusion result, u is the feature vector of the user in the source domain in the second fusion result. i When e is the feature vector of the item in the target domain in the second fusion result, u is the feature vector of the user in the target domain in the second fusion result.

[0128] After determining the corresponding item of each potential overlapping user in the source domain in the target domain and the corresponding item of each potential overlapping user in the target domain in the source domain in the above manner, the corresponding item can be recommended to each potential overlapping user.

[0129] Figure 2 A schematic diagram of a cross-domain data recommendation process according to an embodiment of the present disclosure is shown. Figure 2 As shown, in the cross-domain data recommendation process of the embodiment of the present disclosure, the source domain and the target domain, which respectively include the user-item bipartite graph, are first determined, and then the user-item bipartite graphs in the source domain and the target domain are respectively input into the trained graph convolution model, and multiple subgraphs corresponding to each domain are generated by the subgraph generator after multiple graph convolutions, and potential overlapping users in the source domain and the target domain are obtained. Among them, the subgraph generator (i.e., the unsupervised classifier) ​​is also back-propagated based on the InfoNCE loss to correct the parameters of the subgraph generator. Further, the implicit bridge between the source domain and the target domain is determined based on the potential overlapping users in the source domain and the target domain, and then the subgraph is convolved multiple times, and layer fusion is performed based on each subgraph and the user feature vector in the convolution result, and then inter-domain knowledge transfer is performed between the source domain and the target domain based on the implicit bridge, so as to realize the mutual mapping of user features of potential overlapping users between the source domain and the target domain. Finally, based on the user features after inter-domain knowledge transfer, each subgraph, the subgraph convolution results, the initial embedding vector in the user-item bipartite graph, and each graph convolution result, the feature vectors of the final users and the feature vectors of the items in each domain are determined by subgraph fusion and layer fusion, and then the corresponding items of the users in the source domain are calculated, and the corresponding items of the users in the target domain are calculated.

[0130] Based on the above technical features, the embodiment of the present disclosure identifies potential overlapping users between two different domains by means of graph convolution, constructs an implicit bridge between domains based on the potential overlapping users, and performs cross-domain bidirectional feature migration based on the implicit bridge. Further, according to the results after feature migration, data recommendations are implemented for users based on the features of different domains, thereby improving the accuracy of the recommendation results. Among them, in the process of determining potential overlapping users, the present disclosure introduces a new flow-aware similarity metric, which highlights the specificity between layers and enhances the model's ability to perceive subtle differences between users. At the same time, potential overlapping users are extracted through an unsupervised two-layer subgraph classifier, and InfoNCE is used to constrain the classifier loss to improve the classifier effect and optimize the classification results. Furthermore, in the process of building an implicit bridge, the embodiment of the present disclosure corrects the implicit bridge through similarity, reduces the difficulty of model training, prevents the model from being over-parameterized, and establishes a more general and stable bidirectional mapping relationship between domains.

[0131] Figure 3 A schematic diagram of a cross-domain data recommendation device according to an embodiment of the present disclosure is shown. Figure 3 As shown, the cross-domain data recommendation device of the embodiment of the present disclosure may include:

[0132] A domain determination module 30, configured to determine a source domain and a target domain respectively including a user-item bipartite graph, wherein there are no overlapping users or items in the source domain and the target domain;

[0133] A graph convolution module 31, configured to input the source domain and the target domain into the trained graph convolution model respectively, and obtain potential overlapping users in the source domain and the target domain through graph convolution;

[0134] A bridge building module 32, configured to determine an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain;

[0135] A knowledge transfer module 33, configured to perform inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge, so as to achieve mutual mapping of user features between the source domain and the target domain;

[0136] The data recommendation module 34 is used to determine the items corresponding to the users in the source domain in the source domain according to the user characteristics after the inter-domain knowledge transfer, and to determine the items corresponding to the users in the target domain in the target domain.

[0137] In a possible implementation, the graph convolution module 31 is further used to:

[0138] Performing multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain respectively to obtain a graph convolution result corresponding to each graph convolution, wherein the graph convolution result includes a user feature vector of each user and an item feature vector of each item;

[0139] Classifying users in the source domain and the target domain respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each subgraph having a corresponding category;

[0140] Users in the subgraphs of the same category in the source domain and the target domain are determined as potential overlapping users.

[0141] In a possible implementation, the graph convolution module 31 is further used to:

[0142] The user initial embedding vector obtained by encoding the user feature vectors in the source domain and the target domain through the embedding layer and the user feature vector obtained after the first graph convolution on the user-item bipartite graph are input into the classifier to obtain a preset number of subgraphs and the category corresponding to each of the subgraphs.

[0143] In a possible implementation, the classifier is:

[0144]

[0145] U o =W2U h +b2

[0146] Among them, W1 and W2 are the weight matrices of the classifier, b1 and b2 are the bias vectors of the classifier, σ(·) is the nonlinear activation function LeakyReLU, U o is the predicted category, is the initial embedding vector of user u, is the user feature vector corresponding to user u in the first graph convolution result;

[0147] The graph convolution module 31 is further used for:

[0148] The InfoNCE loss of the classification result is calculated, and the weight matrix and bias vector in the classifier are adjusted.

[0149] In a possible implementation, the bridge construction module 32 is further used to:

[0150] creating an implicit bridge between the source domain and the target domain according to the potential overlapping users;

[0151] Calculating similarities of potential overlapping users in the source domain and the target domain;

[0152] The implicit bridge is modified according to the similarity.

[0153] In a possible implementation, the bridge construction module 32 is further used to:

[0154] pass Calculate the similarity of potential overlapping users in the source domain and the target domain, where u is the user feature vector of the user in the source domain, u′ is the user feature vector of the user in the target domain, l is the number of graph convolution results, L is the total number of graph convolution layers, is the user feature vector obtained by the l-th graph convolution.

[0155] In a possible implementation, the bridge construction module 32 is further used to:

[0156] Deleting irrelevant bridges corresponding to potential overlapping users in the source domain and potential overlapping users in the target domain according to the similarity, wherein the irrelevant bridges are potential overlapping user pairs whose similarity is lower than a preset threshold;

[0157] A preset number of potential overlapping users in the target domain with the highest similarity to users in the source domain and a preset number of potential overlapping users in the source domain with the highest similarity to users in the target domain are retained.

[0158] In a possible implementation, the potential overlapping users are (u s ,u t ),u s For users in the source domain, u t is a user in the target domain, and the implicit bridge is , i is the item corresponding to the user;

[0159] The bridge construction module 32 is further used to:

[0160] Determine the implicit bridge from the target domain to the source as Determine the implicit bridge from the source domain to the target domain as

[0161] In a possible implementation, the bridge construction module 32 is further used to:

[0162] Determine the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as k is the preset number,

[0163] In a possible implementation, the knowledge transfer module 33 is further used to:

[0164] Performing multiple graph convolutions on each type of subgraph in the source domain and the target domain respectively, to obtain a subgraph convolution result corresponding to each graph convolution, wherein the subgraph convolution result includes a user feature vector corresponding to each user and an item feature vector corresponding to each item;

[0165] Performing layer fusion on the user feature vector of each of the potential overlapping users in the source domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain a source domain user vector of each of the potential overlapping users in the source domain;

[0166] Performing layer fusion on the user feature vector of each of the potential overlapping users in the target domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain the target domain user vector of each of the potential overlapping users in the target domain;

[0167] By formula and Perform inter-domain knowledge transfer between the source domain and the target domain respectively, wherein: f(·) is the similarity, λ is the preset hyperparameter, W tr is a preset conversion matrix used to map the user features of the target domain to the source domain, u sis the user feature vector of potential overlapping users in the source domain, u t is the user feature vector of potential overlapping users in the target domain, l is the number of graph convolution results, For user u s The corresponding source domain user vector, For user u t The corresponding target domain user vector.

[0168] In a possible implementation, the data recommendation module 34 is further configured to:

[0169] For each potential overlapping user in the source domain and the target domain, the user feature vector after the inter-domain knowledge transfer between the source domain and the target domain is Performing sub-graph fusion on each of the sub-graph convolution results corresponding to the corresponding graph convolution layer, and for non-potentially overlapping users and each item in the source domain and the target domain, performing sub-graph fusion on the user / item feature vector in each sub-graph and the user / item feature vector in each sub-graph convolution result to obtain a first fusion result;

[0170] Respectively fusing the first fusion result, the initial embedding vector, and each of the graph convolution results corresponding to the source domain and the target domain to obtain a second fusion result;

[0171] Determine the items in the source domain corresponding to the potential overlapping users in the source domain according to the feature vector corresponding to each of the users in the second fusion result corresponding to the source domain and the feature vector of each of the items;

[0172] According to the feature vector corresponding to each of the users in the second fusion result corresponding to the target domain and the feature vector of each of the items, the items corresponding to the potential overlapping users in the target domain in the target domain are determined.

[0173] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0174] The embodiment of the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor. The computer-readable storage medium can be a volatile or non-volatile computer-readable storage medium.

[0175] An embodiment of the present disclosure further proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0176] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0177] Figure 4 1 is a schematic diagram showing an electronic device 1900 according to an embodiment of the present disclosure. For example, the electronic device 1900 may be provided as a server or a terminal device. Figure 4 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0178] The electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output interface 1958 (I / O interface). The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2003. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM or similar.

[0179] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to perform the above method.

[0180] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0181] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0182] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0183] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0184] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0185] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0186] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0187] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0188] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A cross-domain data recommendation method, characterized in that: The method comprises: Determine a source domain and a target domain respectively including a user-item bipartite graph, wherein there are no overlapping users or items in the source domain and the target domain; Inputting the source domain and the target domain into the trained graph convolution model respectively, and obtaining potential overlapping users in the source domain and the target domain through graph convolution; determining an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain, wherein the implicit bridge represents a plurality of potential user pairs formed by potential overlapping users between the source domain and the target domain; Based on the implicit bridge, inter-domain knowledge transfer is performed between the source domain and the target domain to achieve mutual mapping of user features between the source domain and the target domain; Determine the items corresponding to the users in the source domain in the source domain according to the user characteristics after the inter-domain knowledge transfer, and determine the items corresponding to the users in the target domain in the target domain; The step of inputting the source domain and the target domain into the trained graph convolution model respectively, and obtaining potential overlapping users in the source domain and the target domain through graph convolution, comprises: Performing multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain respectively to obtain a graph convolution result corresponding to each graph convolution, wherein the graph convolution result includes a user feature vector of each user and an item feature vector of each item; Classifying users in the source domain and the target domain respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each subgraph having a corresponding category; Users in the subgraphs of the same category in the source domain and the target domain are determined as potential overlapping users.

2. The method according to claim 1, characterized in that The user in the source domain and the target domain are classified by the unsupervised subgraph classifier based on the graph convolution result to obtain a preset number of subgraphs including: The user initial embedding vector obtained by encoding the user feature vectors in the source domain and the target domain through the embedding layer and the user feature vector obtained after the first graph convolution on the user-item bipartite graph are input into the classifier to obtain a preset number of subgraphs and the category corresponding to each of the subgraphs.

3. The method according to claim 2, characterized in that The classifier is: <h2 style=";text-align:left;direction:ltr">U<h2 style=";text-align:left;direction:ltr"> o <h2 style=";text-align:left;direction:ltr"> =W2U<h2 style=";text-align:left;direction:ltr"> h <h2 style=";text-align:left;direction:ltr"> +b2 Among them, W1 and W2 are the weight matrices of the classifier, b1 and b2 are the bias vectors of the classifier, σ(·) is the nonlinear activation function LeakyReLU, U o is the predicted category, is the initial embedding vector of user u, is the user feature vector corresponding to user u in the first graph convolution result; The method further comprises: classifying users in the source domain and the target domain based on the graph convolution result by using an unsupervised subgraph classifier to obtain a preset number of subgraphs; The InfoNCE loss of the classification result output by the classifier is calculated, and the weight matrix and the bias vector in the classifier are adjusted.

4. The method according to any one of claims 1 to 3, characterized in that: The determining an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain comprises: creating an implicit bridge between the source domain and the target domain according to the potential overlapping users; Calculating similarities of potential overlapping users in the source domain and the target domain; The implicit bridge is modified according to the similarity.

5. The method according to claim 4, characterized in that The calculating the similarity of potential overlapping users in the source domain and the target domain includes: pass Calculate the similarity of potential overlapping users in the source domain and the target domain, where u * For users in the source domain, u ′ For users in the target domain, is the number of graph convolution results, L is the total number of graph convolution layers, is the user u obtained by the first graph convolution * The corresponding user feature vector, is the user u obtained by the l-th graph convolution * The corresponding user feature vector, is the user u obtained by the l-1th graph convolution * The corresponding user feature vector; is the user u obtained by the first graph convolution ′ The corresponding user feature vector, is the user u obtained by the l-th graph convolution ′ The corresponding user feature vector, is the user u obtained by the l-1th graph convolution ′ The corresponding user feature vector.

6. The method according to claim 5, characterized in that The step of correcting the implicit bridge according to the similarity comprises: Deleting irrelevant bridges corresponding to potential overlapping users in the source domain and potential overlapping users in the target domain according to the similarity, wherein the irrelevant bridges are potential overlapping user pairs whose similarity is lower than a preset threshold; A preset number of potential overlapping users in the target domain with the highest similarity to users in the source domain and a preset number of potential overlapping users in the source domain with the highest similarity to users in the target domain are retained.

7. The method according to claim 6, characterized in that The potential overlapping users are (u s ,u t ),u s For users in the source domain, u t is a user in the target domain, and the implicit bridge is i is the item corresponding to the user; The deleting irrelevant bridges corresponding to the potential overlapping users in the source domain and the potential overlapping users in the target domain according to the similarity includes: Determine the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as Among them, f(u s ,u t ) is user u in the source domain s With user u in the target domain t The similarity of f(u s ,u t )according to Perform calculations; is the set of potential overlapping users in the source domain, is a set of potential overlapping users in the target domain.

8. The method according to claim 7, characterized in that The retaining of a preset number of potential overlapping users in the target domain having the highest similarity to users in the source domain, and a preset number of potential overlapping users in the source domain having the highest similarity to users in the target domain, comprises: Determine the implicit bridge from the target domain to the source domain as Determine the implicit bridge from the source domain to the target domain as k is the preset number, 9. The method according to any one of claims 5 to 8, characterized in that: The inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge to achieve mutual mapping of user features between the source domain and the target domain includes: Performing multiple graph convolutions on each type of subgraph in the source domain and the target domain respectively, to obtain a subgraph convolution result corresponding to each graph convolution, wherein the subgraph convolution result includes a user feature vector corresponding to each user and an item feature vector corresponding to each item; Performing layer fusion on the user feature vector of each of the potential overlapping users in the source domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain a source domain user vector of each of the potential overlapping users in the source domain; Performing layer fusion on the user feature vector of each of the potential overlapping users in the target domain in the subgraph and the user feature vectors in the convolution results of multiple subgraphs corresponding to the subgraph, respectively, to obtain the target domain user vector of each of the potential overlapping users in the target domain; By formula and Perform inter-domain knowledge transfer between the source domain and the target domain respectively, wherein: λ is the preset hyperparameter, W tr is a preset conversion matrix used to map the user features of the target domain to the source domain, u s are potential overlapping users in the source domain, u t For user u in the source domain s The corresponding potential overlapping users in the target domain, f(u s ,u t ) is user u in the source domain s With user u in the target domain t The similarity of f(u s ,u t )according to Calculate, l is the number of graph convolution results, For user u s The corresponding source domain user vector, For user u t The corresponding target domain user vector.

10. The method according to claim 9, characterized in that The determining, based on the user characteristics after the inter-domain knowledge transfer, the items corresponding to the users in the source domain in the source domain, and determining the items corresponding to the users in the target domain in the target domain, includes: For each potential overlapping user in the source domain and the target domain, the user feature vector after the inter-domain knowledge transfer between the source domain and the target domain is Performing sub-graph fusion on each of the sub-graph convolution results corresponding to the corresponding graph convolution layer, and for non-potentially overlapping users and each item in the source domain and the target domain, performing sub-graph fusion on the user / item feature vector in each sub-graph and the user / item feature vector in each sub-graph convolution result to obtain a first fusion result; Respectively fusing the first fusion result, the initial embedding vector, and each of the graph convolution results corresponding to the source domain and the target domain to obtain a second fusion result; Determine the items in the source domain corresponding to the potential overlapping users in the source domain according to the feature vector corresponding to each of the users in the second fusion result corresponding to the source domain and the feature vector of each of the items; According to the feature vector corresponding to each of the users in the second fusion result corresponding to the target domain and the feature vector of each of the items, the items corresponding to the potential overlapping users in the target domain in the target domain are determined.

11. A cross-domain data recommendation device, characterized in that: The device comprises: A domain determination module, used to determine a source domain and a target domain respectively including a user-item bipartite graph, wherein there are no overlapping users or items in the source domain and the target domain; A graph convolution module, used to input the source domain and the target domain into the trained graph convolution model respectively, and obtain potential overlapping users in the source domain and the target domain through graph convolution; a bridge building module, configured to determine an implicit bridge between the source domain and the target domain according to potential overlapping users in the source domain and the target domain; A knowledge transfer module, used for performing inter-domain knowledge transfer between the source domain and the target domain based on the implicit bridge, so as to realize mutual mapping of user features between the source domain and the target domain; A data recommendation module, used to determine the items corresponding to the users in the source domain in the source domain according to the user characteristics after the inter-domain knowledge transfer, and to determine the items corresponding to the users in the target domain in the target domain; The graph convolution module is also used for: Performing multiple graph convolutions on the user-item bipartite graphs in the source domain and the target domain respectively to obtain a graph convolution result corresponding to each graph convolution, wherein the graph convolution result includes a user feature vector of each user and an item feature vector of each item; Classifying users in the source domain and the target domain respectively based on the graph convolution result by an unsupervised subgraph classifier to obtain a preset number of subgraphs, each subgraph having a corresponding category; Users in the subgraphs of the same category in the source domain and the target domain are determined as potential overlapping users.

12. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 10 when executing the instructions stored in the memory.

13. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.