A cross-platform recommendation method and device, equipment and storage medium

Abstract

Embodiments of the present application relate to the technical field of data processing, in particular, relate to a cross-platform recommendation method and device, equipment and storage medium, aiming to improve the effect of cross-platform recommendation. The method comprises: generating graph structure data through user data and item data in multiple different platforms; inputting the graph structure data into a pre-trained cross-platform recommendation model, extracting features of the graph structure data through a graph convolutional neural network, obtaining user representations of each user and item representations of each item in the multiple different platforms; inputting user representations of each overlapping user in the multiple different platforms into a multi-modal autoencoder network, obtaining reconstructed representations of each overlapping user in each platform of the multiple different platforms; and obtaining recommendation results of each overlapping user on each platform according to the reconstructed representations of each overlapping user in each platform and the item representations corresponding to each platform.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, in particular, to a cross-platform recommendation method and device, equipment and storage medium. BACKGROUND

[0002] With the emergence of a large number of online services and platforms, people have become accustomed to using various platforms at the same time to meet different needs, and such usage habits have led to more and more platforms being gradually linked, and the preferences of a user on different platforms being combined to make cross-platform recommendations of items on the current platform, which can better improve the user's experience. The existing cross-platform recommendation method is to extract the features of users on different platforms respectively, establish a connection between the user features, and make item recommendations, or to establish a model to uniformly learn the cross-platform associated user features and make item recommendations.

[0003] The existing cross-platform recommendation method cannot accurately and fully represent high-order nonlinear connection relationships, and cannot well solve the problem of cross-platform inconsistency, and no end-to-end recommendation method based on a performance excellent neural network has been proposed, so the effect of cross-platform recommendation needs to be improved. SUMMARY

[0004] Embodiments of the present application provide a cross-platform recommendation method, device, equipment and storage medium, aiming to improve the effect of cross-platform recommendation.

[0005] The first aspect of the embodiments of the present application provides a cross-platform recommendation method, which comprises:

[0006] generating graph structure data from user data and item data in a plurality of different platforms;

[0007] inputting the graph structure data into a pre-trained cross-platform recommendation model, and extracting features of the graph structure data through a graph convolutional neural network to obtain user representations of each user and item representations of each item in the plurality of different platforms;

[0008] inputting the user representations of each overlapping user in the plurality of different platforms into a multi-modal autoencoder network to obtain reconstructed representations of each overlapping user in each platform of the plurality of different platforms;

[0009] obtaining a recommendation result of each overlapping user on each platform according to the reconstructed representations of each overlapping user in each platform and the item representations corresponding to each platform.

[0010] Optionally, the training step of the cross-platform recommendation model comprises:

[0011] a loss function of the cross-platform recommendation model is composed of a loss function of the graph convolutional neural network and a reconstruction error value of the multi-modal autoencoder network;

[0012] inputting pre-labeled graph structure data into the cross-platform recommendation model;

[0013] training the cross-platform recommendation model iteratively by a back propagation algorithm using the loss function to obtain the trained cross-platform recommendation model.

[0014] Optionally, the graph structure data is generated by user data and item data in multiple different platforms, including:

[0015] collecting user data and item data on the multiple different platforms;

[0016] obtaining multiple user nodes and item nodes by taking the user data and item data as nodes;

[0017] connecting the user nodes and the item nodes according to a corresponding relationship to obtain graph structure data.

[0018] Optionally, inputting user representations of each overlapping user in the multiple different platforms into a multi-modal autoencoder network to obtain reconstruction representations of the each overlapping user in each platform of the multiple different platforms, including:

[0019] mapping the user representations of each overlapping user in the multiple different platforms into a common hidden layer space and fusing to obtain common representations of the each overlapping user;

[0020] reconstructing the common representations on the multiple different platforms to obtain reconstruction representations of the each overlapping user in each platform of the multiple different platforms.

[0021] Optionally, obtaining recommendation results of the each overlapping user on each platform according to the reconstruction representations of the each overlapping user in each platform and item representations corresponding to each platform, including:

[0022] calculating inner products of the reconstruction representations of the each overlapping user in each platform and item representations of the corresponding platform to obtain inner product calculation results;

[0023] ranking items of the corresponding platform according to recommendation degrees according to the inner product calculation results to obtain a ranked item list;

[0024] taking the ranked item list as the recommendation results of the user on the corresponding platform.

[0025] A second aspect of the embodiments of the present application provides a cross-platform recommendation device, the device comprising:

[0026] a graph structure data generation module configured to generate graph structure data from user data and item data in a plurality of different platforms;

[0027] a representation extraction module configured to input the graph structure data into a pre-trained cross-platform recommendation model, and extract features of the graph structure data by using a graph convolutional neural network to obtain user representations of each user in the plurality of different platforms and item representations of each item;

[0028] a representation reconstruction module configured to input the user representations of each overlapping user in the plurality of different platforms into a multi-modal autoencoder network to obtain reconstructed representations of the each overlapping user in each platform of the plurality of different platforms;

[0029] an item recommendation module configured to obtain a recommendation result of the each overlapping user on each platform according to the reconstructed representations of the each overlapping user in each platform and the item representations of each platform.

[0030] Optionally, the training step of the cross-platform recommendation model comprises:

[0031] a loss function of the cross-platform recommendation model is composed of a loss function of the graph convolutional neural network and a reconstruction error value of the multi-modal autoencoder network;

[0032] pre-labeled graph structure data is input into the cross-platform recommendation model;

[0033] the cross-platform recommendation model is trained iteratively by using the loss function and a back propagation algorithm to obtain the trained cross-platform recommendation model.

[0034] Optionally, the graph structure data generation module comprises:

[0035] a data collection sub-module configured to collect user data and item data on the plurality of different platforms;

[0036] a node generation sub-module configured to use the user data and the item data as nodes to obtain a plurality of user nodes and item nodes;

[0037] a graph structure data generation sub-module configured to connect the user nodes and the item nodes according to a corresponding relationship to obtain graph structure data.

[0038] Optionally, the representation reconstruction module comprises:

[0039] a representation fusion sub-module configured to map the user representations of each overlapping user in the plurality of different platforms to a common hidden layer space and fuse the user representations to obtain common representations of the each overlapping user.

[0040] a representation reconstruction sub-module, configured to reconstruct the common representation on the plurality of different platforms to obtain a reconstructed representation of each platform in the plurality of different platforms for the each overlapping user.

[0041] Optionally, the result recommendation module comprises:

[0042] an inner product calculation sub-module, configured to calculate an inner product of the reconstructed representation of each platform in the plurality of different platforms for the each overlapping user and an item representation of a corresponding platform to obtain an inner product calculation result;

[0043] an item ranking sub-module, configured to rank items of the corresponding platform according to the inner product calculation result to obtain a ranked item list;

[0044] a result recommendation sub-module, configured to take the ranked item list as a recommendation result of the user on the corresponding platform.

[0045] A third aspect of the embodiments of the present application provides a readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the steps in the method of the first aspect of the present application.

[0046] A fourth aspect of the embodiments of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method of the first aspect of the present application are implemented.

[0047] This application employs a cross-platform recommendation method that generates graph-structured data from user and item data across multiple platforms. This graph-structured data is then input into a pre-trained cross-platform recommendation model, where a graph convolutional neural network is used to extract features, yielding user representations for each user and item representations for each item across the multiple platforms. The user representations of each overlapping user across the multiple platforms are then input into a multimodal autoencoder network to obtain reconstructed representations of each overlapping user on each of the multiple platforms. Based on the reconstructed representations of each overlapping user on each platform and the corresponding item representations for each platform, the recommendation result for each overlapping user on each platform is obtained. In this application, graph-structured data is generated from user and item data across multiple platforms, and a graph convolutional network is used to extract features from this data, extracting and preserving high-dimensional nonlinear relationships, which helps improve recommendation performance. This application uses a multimodal autoencoder to map the features of overlapping users into a common hidden space through a nonlinear neural network, and reconstructs user representations from various platforms. The reconstructed user representations, combined with user preference information from each platform, provide a more comprehensive reflection of user preferences, serving as the basis for recommendations and resulting in more accurate recommendations. Furthermore, the multimodal autoencoder network can adaptively mine and correlate consistent parts of the platform data, ensuring cross-platform data consistency. This application combines graph convolutional networks and multimodal autoencoder networks, making the entire model a unified end-to-end neural network that can be trained and applied uniformly. This method fully mines and preserves the high-dimensional nonlinear connections in cross-platform user data, while effectively solving the problem of cross-platform inconsistency, thus improving cross-platform recommendation performance. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of a cross-platform recommendation method proposed in one embodiment of this application;

[0050] Figure 2 This is a flowchart illustrating a cross-platform recommendation process proposed in one embodiment of this application;

[0051] Figure 3 This is a schematic diagram of a cross-platform recommendation device proposed in an embodiment of this application. Detailed Implementation

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0053] The cross-platform recommendation method proposed in the embodiments of the present application is realized based on a cross-platform recommendation model, which includes a graph convolution network and a multi-modal autoencoder network. The training steps of the cross-platform recommendation model are as follows.

[0054] S11: The loss function of the cross-platform recommendation model is composed of the loss function of the graph convolution neural network and the reconstruction error value of the multi-modal autoencoder network.

[0055] In the embodiment, the graph convolution neural network is used to model and extract high-dimensional nonlinear connection relationships, and the multi-modal autoencoder network is used to model the association between cross-platforms. The two parts constitute the entire cross-platform recommendation model.

[0056] The loss function of the cross-platform recommendation model is composed of the loss function of the graph convolution neural network and the reconstruction error of the multi-modal autoencoder network. The reconstruction error is the difference between the user features generated by the multi-modal autoencoder network and the input user features.

[0057] For example, the loss function of the convolutional neural network can use BPR-loss, which can more effectively train the model.

[0058] S12: Input the pre-labeled graph structure data into the cross-platform recommendation model.

[0059] In the embodiment, the pre-labeled graph structure data is a bipartite graph data, which is composed of user data and item data. The labeled graph structure data has marked the associated items of the user on each platform.

[0060] For example, the user A is associated with item 1 and item 2 on platform A, and is associated with item 3 and item 4 on platform B. The user B is associated with item 1 and item 3 on platform A, and is associated with item 2 and item 4 on platform B.

[0061] S13: Use the loss function to train the cross-platform recommendation model iteratively through a back propagation algorithm to obtain the trained cross-platform recommendation model.

[0062] In this embodiment, the labeled graph structure data is input into the cross-platform recommendation model, the graph convolution network obtains user features and item features on each platform in multiple different platforms, and then the overlapping user features in multiple platforms are input into the multi-modal autoencoder network to reconstruct the overlapping user features. According to the input user features and the output reconstructed features, the loss function of the model obtains a loss value, which is fed back to the whole model, which is the back propagation algorithm. Through the algorithm, the cross-platform recommendation model is iteratively trained. When the parameters are adjusted to the optimal, the loss value gradually converges to a smaller level, and the trained cross-platform recommendation model is obtained.

[0063] Reference Figure 1 , Figure 1 is a flowchart of a cross-platform recommendation method according to an embodiment of the present application. As shown in Figure 1 , the method comprises the following steps:

[0064] S21: generating graph structure data from user data and item data in multiple different platforms.

[0065] In this embodiment, the graph structure data is a data composed of nodes and edges, and the nodes represent specific things, and the edges represent the relationship between things.

[0066] In this embodiment, the step of generating graph structure data from user data and item data in multiple different platforms comprises:

[0067] S21-1: collecting user data and item data on the multiple different platforms.

[0068] In this embodiment, the user data of multiple platforms refers to the account information of the user, which includes the identity information of the user and the historical information of the user's activities on the platform. The item data refers to each item on the platform. The user can perform corresponding operations on each item on the platform, and the user has a certain connection with the operated item.

[0069] For example, the multiple different platforms can be shopping platforms, video websites, etc. The items on the shopping platform are various goods such as clothes, mobile phones, etc. The items on the video website are various videos such as emotional videos, sports videos, etc.

[0070] S21-2: taking the user data and the item data as nodes to obtain multiple user nodes and item nodes.

[0071] S21-3: connecting the user nodes and the item nodes according to the corresponding relationship to obtain the graph structure data.

[0072] In this embodiment, the user data is taken as nodes, the item data is taken as nodes, and the relationship between the user and the item is taken as an edge to construct a user-item bipartite graph structure, wherein the user is connected only to the item with which the user has a connection history, there is no connection edge between the users, and there is no connection edge between the items.

[0073] For example, user A has purchased commodity 1 and commodity 2 on a shopping platform, and in the graph structure data, the node of user A is connected to the nodes of commodity 1 and commodity 2. User A has browsed video 3 and video 4 on a video website, and the node of user A is connected to video 3 and video 4.

[0074] In this embodiment, by collecting the user data and the item data on each platform, the user-item bipartite graph structure data is constructed, and analyzing the graph structure data helps to mine the user representation that retains the high-dimensional nonlinear connection relationship.

[0075] S22: inputting the graph structure data into a pre-trained cross-platform recommendation model, performing feature extraction on the graph structure data by a graph convolutional neural network, and obtaining the user representation of each user and the item representation of each item in the plurality of different platforms.

[0076] In this embodiment, the graph convolutional neural network extracts the high-dimensional nonlinear connection relationship, inputs the graph structure data into the graph convolutional neural network, and the graph convolutional neural network performs convolution iteration on the graph structure data. After multiple convolution iterations, the representation of each node on the bipartite graph can obtain useful information from the adjacent representation to improve its own representation, and then obtain the user representation and the item representation that retain the high-dimensional nonlinear connection relationship.

[0077] In this embodiment, the representation of the same user in the plurality of different platforms extracted by the graph convolutional neural network is different, that is, the user representation of the same user on different platforms is different.

[0078] For example, the graph convolutional neural network can adopt a GCN network.

[0079] S23: inputting the user representation of each overlapping user in the plurality of different platforms into a multi-modal autoencoder network to obtain the reconstruction representation of each overlapping user in each platform in the plurality of different platforms.

[0080] In this embodiment, the overlapping user refers to a user who uses a plurality of different platforms at the same time, and the reconstruction representation refers to the representation of the user reconstructed by the multi-modal autoencoder on each different platform. The reconstruction representation is reconstructed according to the representation of the user on each platform, and better represents the preference of the user.

[0081] In this embodiment, the multi-modal autoencoder network has a three-layer structure, including an input layer, a common hidden layer, and an output layer. There is also a fully connected layer between the common hidden layer and the output layer of each platform, which is used to reconstruct the representation of the overlapping user on each platform.

[0082] In this embodiment, the user representation of each overlapping user in the plurality of different platforms is input into the multi-modal autoencoder network to obtain the reconstructed representation of each overlapping user on each platform in the plurality of different platforms, including:

[0083] S23-1: Map the user representation of each overlapping user in the plurality of different platforms to the common hidden layer space and fuse to obtain the common representation of each overlapping user.

[0084] In this embodiment, the common hidden layer space is a layer in the multi-modal autoencoder network, and the user representation of each overlapping user in different platforms is mapped to the common hidden layer space and fused in the common hidden layer space to obtain the common representation, which integrates the user representation of the same user on each platform and can comprehensively reflect the user's preferences.

[0085] For example, user A has a user representation 1 on platform A and a user representation 2 on platform B. The user representation 1 and the user representation 2 are input into the multi-modal autoencoder network to obtain the common representation of user A.

[0086] S23-2: Reconstruct the common representation on the plurality of different platforms to obtain the reconstructed representation of each overlapping user on each platform in the plurality of different platforms.

[0087] In this embodiment, after obtaining the common representation, for each overlapping user, the common representation of the user is reconstructed based on each platform in the plurality of different platforms. The reconstruction process is performed in the fully connected layer. The trained multi-modal autoencoder network will reconstruct the representation of the user on each platform based on the parameters of the user representation of each platform learned during training and the common representation of the user. The reconstructed representation contains the preference information of the user on each platform.

[0088] For example, the overlapping user is user A, who has used a shopping platform and a video platform. The reconstructed feature of user A on the shopping platform combines the user feature of user A on the video platform, and the reconstructed feature of user A on the video platform combines the user feature of user A on the shopping platform.

[0089] In this embodiment, the multi-modal auto-encoder network does not require that the user representations of different platforms have a linear relationship, but instead relaxes the mapping to a common latent space by a nonlinear neural network, and constrains the user representation reconstructed from the common representation to be as consistent as possible with the input user representation. Therefore, the multi-modal auto-encoder network can adaptively mine the consistent part of the cross-platform data for association, ensuring the consistency of the representation of the cross-platform data and increasing the accuracy of cross-platform recommendation.

[0090] S24: obtaining a recommendation result of each overlapping user on each platform according to the reconstructed representation of each overlapping user on each platform and the item representation corresponding to each platform.

[0091] In this embodiment, after obtaining the reconstructed representation of the overlapping user on each platform, the item recommendation for the overlapping user on each platform can be performed according to the obtained reconstructed representation.

[0092] In this embodiment, the step of obtaining a recommendation result of each overlapping user on each platform according to the reconstructed representation of each overlapping user on each platform and the item representation corresponding to each platform comprises:

[0093] S24-1: calculating an inner product of the reconstructed representation of each overlapping user on each platform and the item representation corresponding to the platform to obtain an inner product calculation result.

[0094] S24-2: ranking the items on the corresponding platform according to the recommendation degree according to the inner product calculation result to obtain a ranked item list.

[0095] In this embodiment, the inner product calculation result is positively correlated with the recommendation degree of the item. When the inner product calculation result is larger, it means that the user is more interested in the item, and the corresponding item is more worthy of recommendation. When the inner product calculation result is smaller, it means that the user is less interested in the item, and the recommendation degree of the corresponding item is lower. The items on the platform are ranked according to the recommendation degree, the items with larger recommendation degree are ranked higher, and a ranked item list is obtained.

[0096] For example, platform A is a video website, and the reconstructed representation of user A on the platform is calculated with the representation corresponding to each video in the platform to obtain a ranked video list.

[0097] S24-3: taking the ranked item list as the recommendation result of the user on the corresponding platform.

[0098] For example, user A watches basketball videos and game videos on platform A (a video platform), and purchases paint and kitchenware on platform B (a shopping platform). Taking user A as a target user, the graph structure data of user data and item data on platform A is processed using a cross-platform recommendation model, and the reconstructed representation of user A on platform A and platform B can be obtained.

[0099] If platform A is the target platform, then home decoration videos (referring to the fact that user A purchased paint on platform B) and cooking videos (referring to the fact that user A purchased kitchenware on platform B) are recommended to user A on the target platform.

[0100] If platform B is the target platform, then basketball shoes (referring to the fact that user A watched basketball videos on platform A) and game controllers (referring to the fact that user A watched game videos on platform A) are recommended to user A on the target platform.

[0101] The above items are sorted according to the recommendation degree and pushed to user A.

[0102] In this embodiment, the reconstructed representation is obtained, and item recommendations are made to the user on each platform. In addition, the recommendation degree is sorted according to the recommendation degree, which facilitates the user experience.

[0103] In another case, the input graph structure data includes platform C, but user A has not used platform C. Taking platform C as the target platform, the representation of the user can be reconstructed on the target platform C in combination with the public representation of the user, and items are recommended to the user on platform C according to the reconstructed representation and the item representation on platform C.

[0104] For example, user A watches basketball videos and game videos on platform A (a video platform), and purchases paint and kitchenware on platform B (a shopping platform). After processing by the cross-platform recommendation model, the recommended items for user A on target platform C (a forum) are basketball discussion areas, game discussion areas, home decoration discussion areas, and the like.

[0105] In this embodiment, the cross-platform recommendation model is used to fully mine and retain the high-dimensional nonlinear connection relationship in the cross-platform user data, effectively solve the cross-platform inconsistency problem, and improve the cross-platform recommendation effect.

[0106] In another embodiment of the present application, the cross-platform recommendation process is further illustrated in combination with a cross-platform recommendation flowchart. As shown in Figure 2 Figure 2 is a cross-platform recommendation flowchart according to an embodiment of the present application, as shown in Figure 2 ​The purpose platform is a platform that needs to recommend a to-be-recommended item to a user, and the overlapping users of the source platform and the purpose platform are user A and user B. Then, the graph structure formed by the user data and the item data in the source platform and the purpose platform is input into a graph convolution network to obtain user representations of users ABCD and item representations of items 123456. The overlapping users are user A and user B. The representations of the overlapping user A and the overlapping user B on the source platform and the target platform are input into a multi-modal autoencoder network. First, fusion is performed to obtain a common vector. Then, a reconstruction vector is generated based on the source platform and the purpose platform.

[0107] Based on the same inventive concept, an embodiment of the present application provides a cross-platform recommendation device. Referring to Figure 3 , Figure 3 FIG. 1 is a schematic diagram of a cross-platform recommendation device 300 according to an embodiment of the present application. As shown in the figure, the device includes: Figure 3

[0108] A graph structure data generation module 301 is configured to generate graph structure data by using user data and item data in a plurality of different platforms.

[0109] A representation extraction module 302 is configured to input the graph structure data into a pre-trained cross-platform recommendation model, extract features of the graph structure data by using a graph convolution neural network, and obtain user representations of each user and item representations of each item in the plurality of different platforms.

[0110] A representation reconstruction module 303 is configured to input the user representations of each overlapping user in the plurality of different platforms into a multi-modal autoencoder network, and obtain reconstruction representations of the each overlapping user in each platform of the plurality of different platforms.

[0111] An item recommendation module 304 is configured to obtain a recommendation result of the each overlapping user on each platform according to the reconstruction representations of the each overlapping user in each platform and the item representations corresponding to each platform.

[0112] Optionally, the training step of the cross-platform recommendation model includes:

[0113] The loss function of the cross-platform recommendation model is composed of a loss function of the graph convolution neural network and a reconstruction error value of the multi-modal autoencoder network.

[0114] The pre-labeled graph structure data is input into the cross-platform recommendation model.

[0115] The cross-platform recommendation model is iteratively trained by using the loss function and a back propagation algorithm to obtain the trained cross-platform recommendation model.​

[0116] Optionally, the graph structure data generation module comprises:

[0117] a data collection sub-module, configured to collect user data and item data on the plurality of different platforms;

[0118] a node generation sub-module, configured to take the user data and the item data as nodes to obtain a plurality of user nodes and item nodes;

[0119] a graph structure data generation sub-module, configured to connect the user nodes and the item nodes according to a corresponding relationship to obtain graph structure data.

[0120] Optionally, the representation reconstruction module comprises:

[0121] a representation fusion sub-module, configured to map and fuse user representations of each overlapping user in the plurality of different platforms into a common hidden space to obtain a common representation of each overlapping user;

[0122] a representation reconstruction sub-module, configured to reconstruct the common representation on the plurality of different platforms to obtain a reconstructed representation of each platform of each overlapping user.

[0123] Optionally, the result recommendation module comprises:

[0124] an inner product calculation sub-module, configured to calculate an inner product of the reconstructed representation of each overlapping user in each platform and an item representation of a corresponding platform to obtain an inner product calculation result;

[0125] an item ranking sub-module, configured to rank items of the corresponding platform according to the inner product calculation result to obtain a ranked item list;

[0126] a result recommendation sub-module, configured to take the ranked item list as a recommendation result of the corresponding platform.

[0127] Based on the same inventive concept, another embodiment of the present application provides a readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps in the cross-platform recommendation method according to any one of the above embodiments of the present application.

[0128] Based on the same inventive concept, another embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor being executed to implement the steps in the cross-platform recommendation method according to any one of the above embodiments of the present application.

[0129] For apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts are referred to the part of the description of the method embodiments.

[0130] Each of the embodiments in the present specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other.

[0131] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, apparatus or computer program product. Therefore, the embodiments of the present application can be in the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. In addition, the embodiments of the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] The embodiments of the present application are described with reference to flowcharts and / or block diagrams according to the method, terminal device (system) and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal device to produce a machine, so that the instructions executed by the computer or other programmable data processing terminal device produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows and / or blocks.

[0133] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing terminal device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows and / or blocks.

[0134] These computer program instructions can also be loaded into a computer or other programmable data processing terminal device, so that a series of operation steps are performed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows and / or blocks.

[0135] While preferred embodiments of the application have been described, those skilled in the art will appreciate that other modifications than those specifically described can be made within the scope of the application. Accordingly, the appended claims are intended to embrace all such alternatives as well as the embodiments specifically described.

[0136] Finally, it should be noted that, in the description above, relative terms such as first and second, etc. are used merely to distinguish one entity or action from another, without necessarily implying any actual relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0137] The cross-platform recommendation method, device, equipment and storage medium provided by the present application are described in detail above, and the principles and implementation manners of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range can be changed, and the above description of the specification should not be understood as a limitation of the present application.

Claims

1. A cross-platform recommendation method, characterized by, The method comprises: generating graph structure data from user data and item data in multiple different platforms; inputting the graph structure data into a pre-trained cross-platform recommendation model, extracting features of the graph structure data by a graph convolutional neural network, obtaining user representations of each user in the multiple different platforms and item representations of each item; inputting user representations of each overlapping user in the multiple different platforms into a multi-modal autoencoder network, obtaining reconstructed representations of the each overlapping user in each platform of the multiple different platforms; the overlapping user refers to a user using multiple different platforms simultaneously; the multi-modal autoencoder network takes as a constraint that the user representation reconstructed by a common representation is as consistent as possible with the input user representation, reconstructs the representation of the user on each platform in combination with the common representation of the user, and the reconstructed representation contains preference information of the user on each platform; the common representation of the user integrates user representations of the same user on each platform to comprehensively reflect the preference of the user; obtaining a recommendation result of the each overlapping user on each platform according to the reconstructed representation of the each overlapping user in each platform and the item representation corresponding to each platform.

2. The method of claim 1, wherein, The training steps of the cross-platform recommendation model comprise: composing a loss function of the cross-platform recommendation model by a loss function of the graph convolutional neural network and a reconstruction error value of the multi-modal autoencoder network; inputting pre-labeled graph structure data into the cross-platform recommendation model; training the cross-platform recommendation model iteratively by a back propagation algorithm through the loss function, obtaining the trained cross-platform recommendation model.

3. The method of claim 1, wherein, Generating graph structure data from user data and item data in multiple different platforms comprises: collecting user data and item data on the multiple different platforms; obtaining multiple user nodes and item nodes by taking the user data and item data as nodes; connecting the user nodes and the item nodes according to a corresponding relationship to obtain graph structure data.

4. The method of claim 1, wherein, Inputting user representations of each overlapping user in the multiple different platforms into a multi-modal autoencoder network to obtain reconstructed representations of the each overlapping user in each platform of the multiple different platforms comprises: mapping the user representations of the each overlapping user in the multiple different platforms to a common hidden layer space and integrating them to obtain a common representation of the each overlapping user; reconstructing the common representation on the multiple different platforms to obtain the reconstructed representation of the each overlapping user in each platform of the multiple different platforms.

5. The method of claim 1, wherein, Obtaining a recommendation result of the each overlapping user on each platform according to the reconstructed representation of the each overlapping user in each platform and the item representation corresponding to each platform comprises: calculating an inner product of the reconstructed representation of the each overlapping user in each platform and the item representation of the corresponding platform to obtain an inner product calculation result; ranking items of the corresponding platform according to a recommendation degree according to the inner product calculation result to obtain a ranked item list; The sorted item list is taken as a recommendation result of the user on the corresponding platform. 6.A cross-platform recommendation apparatus, characterized by comprising: The device comprises: a graph structure data generation module configured to generate graph structure data from user data and item data in a plurality of different platforms; a representation extraction module configured to input the graph structure data into a pre-trained cross-platform recommendation model, and extract features of the graph structure data by a graph convolutional neural network to obtain user representations of each user and item representations of each item in the plurality of different platforms; a representation reconstruction module configured to input the user representation of each overlapping user in the plurality of different platforms into a multi-modal autoencoder network to obtain a reconstructed representation of the each overlapping user in each platform of the plurality of different platforms; the overlapping user refers to a user who uses a plurality of different platforms simultaneously; the multi-modal autoencoder network takes as a constraint that the user representation reconstructed by a common representation is as consistent as possible with the input user representation, reconstructs the representation of the user on each platform in combination with the common representation of the user, and the reconstructed representation contains preference information of the user on each platform; the common representation of the user integrates the user representations of the same user on each platform to comprehensively reflect the preference of the user; an item recommendation module configured to obtain a recommendation result of the each overlapping user on each platform according to the reconstructed representation of the each overlapping user in each platform and the item representation corresponding to each platform.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the steps in the method of any one of claims 1 to 5.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 5.

Images