Cross-domain recommendation processing method and device

Abstract

The embodiments of the present specification provide a cross-domain recommendation processing method and device, wherein a cross-domain recommendation processing method includes: performing feature encoding on clothing data of virtual clothing of a virtual image of a user in a virtual world to obtain virtual clothing features; performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain real clothing features of the user; determining clothing recommendation features of the user based on the real clothing features and a user connection relationship network; and performing clothing recommendation processing for the user according to the clothing recommendation features.

Description

Technical Field

[0001] This document relates to the field of virtualization technology, and in particular to a cross-domain recommendation processing method and device. Background Art

[0002] The virtual world provides a simulation of the real world, and can even provide scenes that are difficult to achieve in the real world. Therefore, the virtual world is increasingly used in various scenarios. In the virtual world scenario, users log in to the three-dimensional virtual world through a specific identity ID and carry out activities through virtual user roles in the virtual world. Usually, there are different user roles in the virtual world, and these user roles each carry out different activities. Summary of the invention

[0003] One or more embodiments of the present specification provide a cross-domain recommendation processing method, including: encoding the clothing data of virtual clothing of a virtual image of a user in a virtual world to obtain virtual clothing features. Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain real clothing features of the user. Determining clothing recommendation features of the user based on the real clothing features and a user connection network. Performing clothing recommendation processing for the user according to the clothing recommendation features.

[0004] One or more embodiments of the present specification provide a cross-domain recommendation processing device, including: a feature encoding module, configured to perform feature encoding on clothing data of virtual clothing of a user's virtual image in a virtual world to obtain virtual clothing features. A feature mapping module, configured to perform feature mapping from the virtual world to the real world for the virtual clothing features to obtain the user's real clothing features. A clothing recommendation feature determination module, configured to determine the user's clothing recommendation features based on the real clothing features and a user connection relationship network. A clothing recommendation processing module, configured to perform clothing recommendation processing for the user based on the clothing recommendation features.

[0005] One or more embodiments of the present specification provide a cross-domain recommendation processing device, including: a processor; and a memory configured to store computer executable instructions, wherein when the computer executable instructions are executed, the processor: performs feature encoding on the clothing data of the virtual clothing of the virtual image of the user in the virtual world to obtain virtual clothing features. Performs feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user. Determines the clothing recommendation features of the user based on the real clothing features and the user connection relationship network. Performs clothing recommendation processing for the user according to the clothing recommendation features.

[0006] One or more embodiments of the present specification provide a storage medium for storing computer executable instructions, which implement the following process when executed by a processor: feature encoding the clothing data of virtual clothing of a virtual image of a user in a virtual world to obtain virtual clothing features. Feature mapping from the virtual world to the real world is performed on the virtual clothing features to obtain real clothing features of the user. Based on the real clothing features and the user connection relationship network, clothing recommendation features of the user are determined. Clothing recommendation processing is performed to the user according to the clothing recommendation features. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate one or more embodiments of the present specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

[0008] Figure 1 A processing flow chart of a cross-domain recommendation processing method provided by one or more embodiments of this specification;

[0009] Figure 2 A processing flow chart of a cross-domain recommendation processing method applied to a cross-domain recommendation scenario provided by one or more embodiments of this specification;

[0010] Figure 3 A schematic diagram of a cross-domain recommendation processing device provided for one or more embodiments of this specification;

[0011] Figure 4 A schematic diagram of the structure of a cross-domain recommendation processing device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0012] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0013] This specification provides an embodiment of a cross-domain recommendation processing method:

[0014] The cross-domain recommendation processing method provided by the present application uses the clothing data of the virtual clothing of the virtual image of the user in the virtual world as the source feature, cross-domain maps the source feature to the clothing recommendation feature of the real clothing of the real image of the user in the real world, and finally uses the clothing recommendation feature of the real clothing to perform clothing recommendation processing on the user. Specifically, the virtual clothing features of the virtual clothing of the virtual image of the user in the virtual world are mapped from the virtual world to the real world to obtain the real clothing features of the user in the real world; based on the real clothing features, the clothing recommendation features of the user are determined in the user connection relationship network, and clothing recommendation processing is performed on the user according to the clothing recommendation features, so that the clothing data of the user in the virtual world is used as the basic data for recommending clothing to the user in the real world, thereby improving the effectiveness of clothing recommendations to the user in the real world, and further improving the user's perception of the recommended clothing.

[0015] Step S102, feature encoding is performed on clothing data of virtual clothing of the user's avatar in the virtual world to obtain virtual clothing features.

[0016] The virtual world described in this embodiment refers to a virtual simulation world based on decentralized collaboration and with an open economic system. Optionally, decentralized transactions are performed in the virtual world by generating non-homogeneous identifiers, and ownership of virtual assets is obtained through transactions. Specifically, users in the real world can access the virtual world through access devices to conduct decentralized transactions and other behaviors in the virtual world, such as a game virtual world where users interact with games through access devices, a conference virtual world where users conduct online meetings through access devices, or a transaction virtual world where users conduct online transactions through access devices; the access device of the virtual world can be a VR (Virtual Reality) device connected to the virtual world, an AR (Augmented Reality) device, etc., such as a head-mounted VR device connected to the virtual world. The server refers to a server or service platform that provides corresponding services for access devices that access the virtual world, or a server or service platform that maintains the operation of the virtual world.

[0017] In this embodiment, the virtual image refers to a three-dimensional digital image (three-dimensional digital image) obtained by reconstructing the user's image using image reconstruction technology based on the user's image, which is used to display the user's image in the virtual world, or a digital twin image in the virtual world that has a digital twin relationship or identity mapping with the user in the real world obtained by reconstructing the user's image using image reconstruction technology based on the user's image. For example, based on the user's images from multiple perspectives in the real world, the user images from multiple perspectives are input into the Neural Radiance Fields (NeRF) model for image reconstruction to obtain a three-dimensional digital image, digital twin image, or identity mapping image in the virtual world.

[0018] The virtual image can be dressed in clothing, and the specific clothing dressed in the virtual image is the clothing data; the style characteristics of the clothing are virtual clothing characteristics of the clothing data; for example, the color of a virtual clothing is yellow casual short-sleeved, then the virtual clothing characteristics of the virtual clothing are yellow, casual, and short-sleeved.

[0019] In specific implementation, since the clothing data of virtual clothing in the virtual world is large, directly using the clothing data of virtual clothing for processing affects the processing efficiency. In order to improve the processing efficiency and save processing resources, in this embodiment, the clothing data of the virtual image of the user in the virtual world is feature encoded to obtain the virtual clothing features;

[0020] In order to ensure the accuracy of the obtained virtual clothing features, in an optional implementation provided by this embodiment, the following method is used to implement feature encoding of clothing data to obtain virtual clothing data:

[0021] The clothing data is input into a virtual coding network for feature coding to obtain the virtual clothing features.

[0022] Specifically, the clothing data is input into a virtual encoder for encoding processing to obtain virtual clothing features.

[0023] In order to improve the effectiveness of the virtual clothing features encoded by the virtual coding network, a virtual data reconstruction model for reconstructing clothing data in the virtual world is trained, and the encoding network in the virtual data reconstruction model is used as the virtual coding network for encoding the features of clothing data in the virtual world;

[0024] In practical applications, the virtual data reconstruction model can be pre-trained, for example, the virtual data reconstruction model is trained on a cloud server. Specifically, in the training process of the virtual data reconstruction model, in an optional implementation manner provided in this embodiment, the virtual data reconstruction model is trained in the following manner:

[0025] Inputting clothing data samples in the virtual world into the encoding network in the virtual data reconstruction model to be trained for feature encoding, and obtaining sample clothing features of the clothing data samples;

[0026] Inputting the sample clothing features into a decoding network in the virtual data reconstruction model to perform data decoding, and obtaining sample reconstruction data of the sample clothing features;

[0027] The training losses of the sample reconstruction data and the clothing data samples are calculated, and parameters of the virtual data reconstruction model are adjusted based on the training losses.

[0028] The virtual data reconstruction model to be trained can adopt semantic segmentation, such as UNET architecture; the virtual data reconstruction model to be trained includes an encoding network (encoder) and a decoding network (decoder); the input of the encoder is a clothing data sample of the virtual world, and the output is the corresponding sample clothing features; the input of the decoder is the sample clothing features, and the output is the reconstructed clothing data sample, that is, the sample reconstruction data;

[0029] After obtaining the sample reconstruction data, the parameters of the virtual data reconstruction model to be trained are adjusted based on the reconstruction loss of the sample reconstruction data and the clothing data sample until the model converges; the encoder in the virtual data reconstruction model obtained through training is used as a virtual encoding network.

[0030] It should also be noted that the clothing data samples include clothing matching data of users collected in the virtual world within a preset time or a preset number of times; for example, clothing matching data of nearly 10 times of the virtual images of one or more users in the virtual world are collected as clothing data samples; wherein the clothing matching data includes tops data, bottoms data, shoes data, decoration data, etc.

[0031] In addition, the above-mentioned process of training the virtual data reconstruction model can also be replaced by inputting the clothing data samples in the virtual world into the virtual data reconstruction model to be trained for data reconstruction; calculating the reconstruction loss based on the clothing data samples and the output sample reconstruction data; adjusting the parameters of the virtual data reconstruction model to be trained according to the reconstruction loss; and forming a new implementation method with the other steps provided in this embodiment. Optionally, the virtual data reconstruction model to be trained includes an encoding network and a decoding network. Optionally, after adjusting the parameters until the model converges, the encoding network in the virtual data reconstruction model obtained by model training is used as the virtual encoding network.

[0032] Step S104, mapping the virtual clothing features from the virtual world to the real world to obtain the user's real clothing features.

[0033] Corresponding to the virtual clothing features, the real clothing features include features corresponding to clothing data representing the user in the real world.

[0034] On the basis of obtaining the above-mentioned virtual clothing features, since it is necessary to recommend real-world clothing to the user, there are domain differences between the virtual clothing features in the virtual world and the clothing features in the real world. Therefore, it is impossible to directly use the virtual clothing features to recommend real-world clothing to the user. Therefore, in this embodiment, when the virtual clothing features are obtained, the virtual clothing features are mapped from the virtual world to the real world to obtain the user's real clothing features.

[0035] In the specific implementation, in order to improve the efficiency of feature mapping, ensure the effectiveness of feature mapping, and improve the accuracy of clothing recommendation for users in the real world, in an optional implementation provided by this embodiment, the feature mapping of virtual clothing features from the virtual world to the real world is achieved in the following manner:

[0036] The virtual clothing features are input into a cross-domain mapping model for cross-domain mapping to obtain the real clothing features.

[0037] In practical applications, the cross-domain mapping model can be pre-trained, for example, the cross-domain mapping model is trained on a cloud server. Specifically, in the training process of the cross-domain mapping model, in an optional implementation manner provided in this embodiment, the cross-domain mapping model is trained in the following manner:

[0038] Input the virtual feature samples with real feature annotation into the cross-domain mapping model to be trained for cross-domain mapping to obtain the mapped real features;

[0039] Calculating the regression loss of the mapped real features and the real features labeled by the real features;

[0040] Parameters of the cross-domain mapping model to be trained are adjusted based on the regression loss.

[0041] Among them, the cross-domain mapping model can adopt a three-layer neural network architecture, with virtual clothing features as input and real clothing features as output; during the training process, the input is virtual feature samples with real feature annotations, and the output is the mapped real features. The loss function is the regression loss, and the parameters of the cross-domain mapping model to be trained are adjusted based on the regression loss until the model converges to obtain the cross-domain mapping model.

[0042] Specifically, in the process of training the cross-domain mapping model, virtual feature samples with real feature annotation are needed; the virtual feature samples include virtual clothing features obtained by inputting clothing data samples in the virtual world into a virtual coding network for feature encoding; in order to train the cross-domain mapping model, supervised training is performed so that the cross-domain mapping model can map the virtual clothing features in the virtual world to the real clothing features in the real world; therefore, the virtual feature samples are annotated with real features, and then the cross-domain mapping model is trained based on the virtual feature samples annotated with real features.

[0043] The real feature of the real feature annotation is obtained by feature encoding the clothing data of the real clothing of the real image of the user in the real world. Optionally, the real feature is determined in the following manner: feature encoding the clothing data of the real clothing of the real image of the user in the real world to obtain the real feature;

[0044] The step of encoding the clothing data of the real clothing of the real image of the user in the real world to obtain the real features includes:

[0045] The clothing data of the real clothing is input into a real coding network for feature coding processing to obtain the real features.

[0046] In practical applications, the real coding network can be pre-trained, for example, the real coding network is trained on a cloud server. Specifically, in the training process of the real coding network, in an optional implementation manner provided in this embodiment, the real coding network is trained in the following manner:

[0047] Inputting clothing data samples in the real world into the encoding network in the real data reconstruction model to be trained for feature encoding, and obtaining sample clothing features of the clothing data samples;

[0048] Inputting the sample clothing features into a decoding network in the real data reconstruction model to perform data decoding, and obtaining sample reconstruction data of the sample clothing features;

[0049] The training losses of the sample reconstruction data and the clothing data samples are calculated, and parameters of the real data reconstruction model are adjusted based on the training losses.

[0050] In order to ensure the effectiveness and accuracy of feature encoding by the real encoding network, a real data reconstruction model including the encoding network is trained, and the encoding network in the real data reconstruction model obtained through training is used as the real encoding network to perform feature encoding on clothing data in the real world.

[0051] The real data reconstruction model to be trained can adopt semantic segmentation, such as UNET architecture; the real data reconstruction model to be trained includes an encoding network (encoder) and a decoding network (decoder); the input of the encoder is a clothing data sample in the real world, and the output is the corresponding sample clothing features; the input of the decoder is the sample clothing features, and the output is the reconstructed clothing data sample, that is, the sample reconstruction data;

[0052] After obtaining the sample reconstruction data, the parameters of the real data reconstruction model to be trained are adjusted based on the reconstruction loss of the sample reconstruction data and the clothing data sample until the model converges; the encoder in the real data reconstruction model obtained through training is used as the real encoding network.

[0053] It should also be noted that the real-world clothing data samples include clothing matching data corresponding to the clothing purchase records of users within a preset time period on the online platform or clothing matching data of a preset number of times; for example, clothing data corresponding to nearly 20 clothing purchase records of one or more users on the shopping platform are collected; among which, the clothing matching data includes tops data, bottoms data, shoes data, decoration data, etc.

[0054] In addition, the above-mentioned process of training the real data reconstruction model can also be replaced by inputting clothing data samples in the real world into the real data reconstruction model to be trained for data reconstruction; calculating the reconstruction loss based on the clothing data samples and the output sample reconstruction data; adjusting the parameters of the real data reconstruction model to be trained according to the reconstruction loss; and forming a new implementation method with the other steps provided in this embodiment. Optionally, the real data reconstruction model to be trained includes an encoding network and a decoding network. Optionally, after adjusting the parameters until the model converges, the encoding network in the real data reconstruction model obtained by model training is used as the real virtual encoding network.

[0055] Since there are certain differences between the virtual clothing features in the virtual world and the real clothing features in the real world, after obtaining the real feature and virtual feature samples, in an optional implementation manner provided by this embodiment, the following operations are further performed:

[0056] Calculate the correlation between real features and virtual feature samples;

[0057] The real features and virtual feature samples whose correlation is higher than a preset threshold are taken as a group of data for real feature labeling.

[0058] That is, by calculating the correlation between the real feature and the virtual feature sample, for each virtual feature sample, the real feature with the highest correlation and the virtual feature sample are used as a set of data, and the virtual feature sample is labeled with the real feature based on the real feature. The calculation of the correlation between the real feature and the virtual feature sample includes calculating the similarity between the real feature and the virtual feature sample.

[0059] In the specific implementation process, the cross-domain mapping model is trained in the following way:

[0060] (1) The virtual data reconstruction model to be trained is trained based on clothing data samples in the virtual world to obtain a virtual data reconstruction model.

[0061] (2) The real data reconstruction model to be trained is trained based on clothing data samples in the real world to obtain a real data reconstruction model.

[0062] (3) The clothing data samples in the virtual world are input into the virtual coding network in the virtual data reconstruction model for feature encoding to obtain virtual feature samples.

[0063] (4) The clothing data samples in the real world are input into the virtual coding network in the real data reconstruction network for feature encoding to obtain real features.

[0064] (5) Calculate the similarity between each real feature and each virtual feature sample, and obtain at least one sample group.

[0065] Each sample group includes a virtual feature sample and a real feature; a real feature here includes a feature corresponding to a set of clothing data. The similarity between the virtual feature sample and the real feature in each sample group is higher than the similarity between the virtual feature sample and other real features, and the similarity between the real feature and other virtual feature samples.

[0066] (6) Based on the real features in the sample group, the virtual feature samples in the sample group are labeled to obtain virtual feature samples that are labeled with real features.

[0067] (7) The cross-domain mapping model to be trained is trained based on the virtual feature samples with real feature annotations to obtain a cross-domain mapping model.

[0068] It should be noted that the above processes (1) and (2) can be implemented preferentially.

[0069] In addition, the above step S104 can also be replaced by inputting the virtual clothing features into the cross-domain mapping model for feature mapping to obtain the real clothing features of the user in the real world; and together with the other processing steps provided in this embodiment, a new implementation method is formed. Alternatively, the above steps S102 to S104 can also be replaced by performing feature mapping from the virtual world to the real world on the virtual clothing features of the virtual image of the user in the virtual world to obtain the real clothing features of the user; and together with the other processing steps provided in this embodiment, a new implementation method is formed.

[0070] Step S106, determining clothing recommendation features for the user based on the real clothing features and the user connection network.

[0071] The user connection relationship network includes a network with connection relationships between users and related users that is created based on real features.

[0072] In order to improve the effectiveness of the determined clothing recommendation features of the user, in an optional implementation manner provided in this embodiment, the clothing recommendation features of the user are determined in the following manner:

[0073] The real clothing features are input into a graph model for feature determination to obtain the clothing recommendation features; the graph model includes the user connection relationship network.

[0074] Specifically, the real clothing features are input into a graph model including the user connection relationship network to perform feature determination, thereby obtaining the clothing recommendation features.

[0075] The graph model is determined in the following manner:

[0076] Determine the relevant real feature samples of each real feature sample, establish a connection relationship between each real feature sample and the corresponding relevant real feature sample, and obtain the user connection relationship network;

[0077] Performing dimensionality reduction processing on two real feature samples having a connection relationship in the user connection relationship network;

[0078] The dimension reduction results whose similarity between the two real feature samples is higher than the similarity of other dimension reduction results are used as the clothing recommendation features of each of the two real feature samples.

[0079] Specifically, the determination of the graph model is actually the determination of the user connection network and the clothing recommendation features of each user in the user connection network; therefore, in the process of determining the graph model, the related users of each user are determined, the connection relationship between each user and the related users is established, and the real feature samples of the users with the connection relationship and the related users are subjected to dimensionality reduction processing, and the target dimensionality reduction results with a higher similarity between the two users than other dimensionality reduction results are obtained as the clothing recommendation features of the two users. Among them, the process of processing the users is actually the process of processing the real feature samples of the users. Among them, in the process of determining the relevant real feature samples of each real feature sample, the users who have the same clothing data as the user to which the real feature sample belongs are regarded as the relevant users of the user; further, the real feature samples of the related users are regarded as the relevant real feature samples of the real feature samples of the user.

[0080] For example, for any user A and user B related to user A, a connection relationship between user A and related user B is established, and the real feature samples of user A and related user B are subjected to dimensionality reduction processing, and the dimensionality reduction processing results (same dimension) of user A and related user B with a higher similarity than the dimensionality reduction processing results of other dimensions are obtained, and the dimensionality reduction processing result of user A in the obtained dimensionality reduction processing results is used as the clothing recommendation feature of user A; the dimensionality reduction processing result of related user B in the obtained dimensionality reduction processing results is used as the clothing recommendation feature of related user B. The determined clothing recommendation features of each user are stored in correspondence with the node where the user is located.

[0081] In a specific implementation process, in an optional implementation mode provided by this embodiment, in the process of inputting the real clothing features into the graph model for feature determination, the graph model uses the following method to determine the features:

[0082] If the real clothing feature corresponds to an existing node in the user connection relationship network, then reading and outputting the clothing recommendation feature corresponding to the existing node;

[0083] If the real clothing feature does not correspond to an existing node in the user feature network, the user connection relationship network is updated and trained based on the real clothing feature, and the clothing recommendation feature corresponding to the existing node corresponding to the real service feature in the trained user connection relationship network is read and output.

[0084] Specifically, if the real clothing feature corresponds to an existing node in the user connection network, the clothing recommendation feature corresponding to the existing node is read and output; if the real clothing feature does not correspond to an existing node in the user connection relationship network, the user connection relationship network is updated based on the real clothing feature to obtain an updated user connection relationship network, and the clothing recommendation feature corresponding to the existing node corresponding to the real clothing feature in the updated user connection relationship network is read and output.

[0085] In the above process of determining the user's clothing recommendation features based on real clothing features and the user connection relationship network, the graph model may not be introduced. Optionally, the user connection relationship network is determined in the following manner: determining the relevant real feature samples of each real feature sample, establishing a connection relationship between each real feature sample and the corresponding relevant real feature sample, and obtaining the user connection relationship network; performing dimensionality reduction processing on two real feature samples with a connection relationship in the user connection relationship network; taking the dimensionality reduction result with the highest similarity between the two real feature samples as the clothing recommendation features of each of the two real feature samples and storing them in correspondence with the corresponding real feature samples.

[0086] In addition, the above step S106 can also be replaced by determining the clothing recommendation feature of the user based on the real clothing feature; and forming a new implementation method with other processing steps provided in this embodiment. Alternatively, the real clothing feature is input into the user connection relationship diagram for feature determination to obtain the clothing recommendation feature of the user.

[0087] In a specific implementation, in order to ensure the efficiency of determining the clothing recommendation features and to ensure the realization of the graph model, in an optional implementation provided by this embodiment, before inputting the real clothing features into the graph model for feature determination, the following operations are further performed:

[0088] If the number of features of the real clothing features is at least two, calculating average data of the real clothing features;

[0089] Accordingly, the step of inputting the real clothing features into the graph model to determine the features and obtaining the clothing recommendation features includes:

[0090] The average data is input into a graph model to perform feature determination, thereby obtaining the service recommendation feature.

[0091] Specifically, when the obtained real clothing features include at least two features, an average value of the at least two features is calculated, and the average feature is input into the graph model for feature determination to obtain the service recommendation feature.

[0092] Step S108: performing clothing recommendation processing for the user according to the clothing recommendation features.

[0093] In a specific implementation, after determining the clothing recommendation features of the user, clothing recommendation processing is performed to the user based on the clothing recommendation features.

[0094] In order to improve the effectiveness of clothing recommendations and enhance the user's perception of the recommended clothing, in an optional implementation provided by this embodiment, in the process of making clothing recommendations to the user according to clothing recommendation features, the following operations are performed:

[0095] Calculating the correlation between the clothing recommendation feature and each clothing feature in the clothing database;

[0096] Target clothing with correlation higher than a recommendation threshold is screened out as recommended clothing for the user and recommended to the user.

[0097] Specifically, target clothing corresponding to clothing features in the clothing library whose similarity with clothing recommendation features is higher than a preset threshold is recommended to the user, that is, the recommended clothing or recommended clothing is sent to the user's access device.

[0098] In addition, the above steps S106 to S108 may also be replaced by performing clothing recommendation processing for the user based on the real clothing features.

[0099] In summary, the cross-domain recommendation processing method provided in this embodiment first inputs the clothing data of the virtual clothing of the virtual image of the user in the virtual world into the virtual encoder for feature encoding to obtain virtual clothing features; then inputs the virtual clothing features into the cross-domain mapping model for feature mapping to obtain the real clothing features of the user in the real world; then calculates the average features of the real clothing features, inputs the average features into the user connection relationship network for feature determination, obtains the user's clothing recommendation features, and finally performs clothing recommendation processing to the user based on the clothing recommendation features. In this way, the clothing data in the virtual world is used as the source domain, and the real clothing features in the real world are used as the target domain to achieve feature mapping from the virtual world to the real world. Finally, the real clothing features after feature mapping are used to perform clothing recommendation processing on the user in the real world, thereby improving the effectiveness of clothing recommendation processing for the user, and making the clothing recommended to the user more in line with the user's preferences.

[0100] The following takes the application of a cross-domain recommendation processing method provided by this embodiment in a cross-domain recommendation scenario as an example to further illustrate the cross-domain recommendation processing method provided by this embodiment. Figure 2 , a cross-domain recommendation processing method applied to a cross-domain recommendation scenario specifically includes the following steps.

[0101] Step S202 , collecting clothing data of virtual clothing of the user's virtual image in the virtual world.

[0102] Step S204, inputting the clothing data into a virtual coding network for feature coding to obtain virtual clothing features.

[0103] Step S206: input the virtual clothing features into the cross-domain mapping model for feature mapping to obtain the real clothing features of the user in the real world.

[0104] Step S208: input the real clothing features into the user connection relationship network for feature determination.

[0105] Step S210: If the real clothing feature corresponds to an existing node in the user connection relationship network, the clothing recommendation feature corresponding to the corresponding existing node is read.

[0106] Step S212: if the real clothing feature does not correspond to an existing node in the user connection relationship network, the user connection relationship network is updated to create a network node corresponding to the real clothing feature.

[0107] Step S214, determining clothing recommendation features of the network node.

[0108] Step S216, recommending clothing to the user based on the clothing recommendation features.

[0109] An embodiment of a cross-domain recommendation processing device provided in this specification is as follows:

[0110] In the above-mentioned embodiment, a cross-domain recommendation processing method is provided, and correspondingly, a cross-domain recommendation processing device is also provided, which will be described below with reference to the accompanying drawings.

[0111] Reference Figure 3 , which shows a schematic diagram of a cross-domain recommendation processing device provided by this embodiment.

[0112] Since the device embodiment corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the corresponding description of the method embodiment provided above. The device embodiment described below is only illustrative.

[0113] This embodiment provides a cross-domain recommendation processing device, including:

[0114] The feature encoding module 302 is configured to perform feature encoding on clothing data of virtual clothing of the user's avatar in the virtual world to obtain virtual clothing features;

[0115] A feature mapping module 304 is configured to perform feature mapping from the virtual world to the real world for the virtual clothing feature, so as to obtain the real clothing feature of the user;

[0116] A clothing recommendation feature determination module 306 is configured to determine clothing recommendation features of the user based on the real clothing features and the user connection relationship network;

[0117] The clothing recommendation processing module 308 is configured to perform clothing recommendation processing for the user according to the clothing recommendation features.

[0118] An embodiment of a cross-domain recommendation processing device provided in this specification is as follows:

[0119] Corresponding to the cross-domain recommendation processing method described above, based on the same technical concept, one or more embodiments of this specification also provide a cross-domain recommendation processing device, which is used to execute the cross-domain recommendation processing method provided above. Figure 4 A schematic diagram of the structure of a cross-domain recommendation processing device provided in one or more embodiments of this specification.

[0120] This embodiment provides a cross-domain recommendation processing device, including:

[0121] like Figure 4 As shown, the cross-domain recommendation processing device may have relatively large differences due to different configurations or performances, and may include one or more processors 401 and memory 402, and the memory 402 may store one or more storage applications or data. Among them, the memory 402 may be a short-term storage or a persistent storage. The application stored in the memory 402 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the cross-domain recommendation processing device. Furthermore, the processor 401 may be configured to communicate with the memory 402 and execute a series of computer executable instructions in the memory 402 on the cross-domain recommendation processing device. The cross-domain recommendation processing device may also include one or more power supplies 403, one or more wired or wireless network interfaces 404, one or more input / output interfaces 405, one or more keyboards 406, etc.

[0122] In a specific embodiment, the cross-domain recommendation processing device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the cross-domain recommendation processing device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following:

[0123] Characteristic encoding is performed on clothing data of virtual clothing of a virtual image of a user in the virtual world to obtain characteristics of the virtual clothing;

[0124] Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user;

[0125] Determining clothing recommendation features for the user based on the real clothing features and the user connection relationship network;

[0126] Perform clothing recommendation processing for the user according to the clothing recommendation features.

[0127] An embodiment of a storage medium provided in this specification is as follows:

[0128] Corresponding to the cross-domain recommendation processing method described above, based on the same technical concept, one or more embodiments of this specification also provide a storage medium.

[0129] The storage medium provided in this embodiment is used to store computer executable instructions, and the computer executable instructions implement the following process when executed by a processor:

[0130] Characteristic encoding is performed on clothing data of virtual clothing of a virtual image of a user in the virtual world to obtain characteristics of the virtual clothing;

[0131] Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user;

[0132] Determining clothing recommendation features for the user based on the real clothing features and the user connection relationship network;

[0133] Perform clothing recommendation processing for the user according to the clothing recommendation features.

[0134] It should be noted that the embodiment of the storage medium in this specification and the embodiment of the cross-domain recommendation processing method in this specification are based on the same inventive concept, so the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0135] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0136] In the 1930s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0137] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0138] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0139] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0140] It should be understood by those skilled in the art that one or more embodiments of this specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take 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.

[0141] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0142] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0144] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0145] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0146] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0147] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0148] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0149] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0150] The above description is only an embodiment of this document and is not intended to limit this document. For those skilled in the art, this document may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this document should be included in the scope of the claims of this document.

Claims

1. A cross-domain recommendation processing method, comprising: Characteristic encoding is performed on clothing data of virtual clothing of a virtual image of a user in the virtual world to obtain characteristics of the virtual clothing; Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user; Determining clothing recommendation features for the user based on the real clothing features and the user connection relationship network; The user connection relationship network includes a real feature sample of a user and a connection relationship between real feature samples of related users, wherein the real feature sample is determined based on feature encoding of clothing data of real clothing of the real image of the user in the real world; the clothing recommendation feature of each of two real feature samples having a connection relationship in the user connection relationship network is determined based on a dimensionality reduction result of the dimensionality reduction results of the two real feature samples whose similarity is higher than the similarity of other dimensionality reduction results; Perform clothing recommendation processing for the user according to the clothing recommendation features.

2. According to the cross-domain recommendation processing method of claim 1, the step of encoding the clothing data of the virtual clothing of the virtual image of the user in the virtual world to obtain the virtual clothing features comprises: The clothing data is input into a virtual coding network for feature coding to obtain the virtual clothing features.

3. According to the cross-domain recommendation processing method of claim 2, the virtual coding network is trained in the following manner: Inputting clothing data samples in the virtual world into the encoding network in the virtual data reconstruction model to be trained for feature encoding, and obtaining sample clothing features of the clothing data samples; Inputting the sample clothing features into a decoding network in the virtual data reconstruction model to perform data decoding, and obtaining sample reconstruction data of the sample clothing features; The training losses of the sample reconstruction data and the clothing data samples are calculated, and parameters of the virtual data reconstruction model are adjusted based on the training losses, and the encoding network in the trained virtual data reconstruction model is used as the virtual encoding network.

4. The cross-domain recommendation processing method according to claim 1, wherein the step of mapping the virtual clothing features from the virtual world to the real world to obtain the user's real clothing features comprises: The virtual clothing features are input into a cross-domain mapping model for cross-domain mapping to obtain the real clothing features.

5. According to the cross-domain recommendation processing method of claim 4, the cross-domain mapping model is trained in the following manner: Input the virtual feature samples with real feature annotation into the cross-domain mapping model to be trained for cross-domain mapping to obtain the mapped real features; Calculating the regression loss of the mapped real features and the real features labeled by the real features; Parameters of the cross-domain mapping model to be trained are adjusted based on the regression loss.

6. According to the cross-domain recommendation processing method of claim 5, the real features are determined in the following manner: Performing feature encoding on clothing data of real clothing of a real image of a user in the real world to obtain the real features; in, The step of encoding the clothing data of the real clothing of the real image of the user in the real world to obtain the real features includes: The clothing data of the real clothing is input into a real coding network for feature coding to obtain the real features.

7. According to the cross-domain recommendation processing method of claim 5, the real feature annotation comprises: Calculate the correlation between real features and virtual feature samples; The real features and virtual feature samples whose correlation is higher than a preset threshold are taken as a group of data for real feature labeling.

8. According to the cross-domain recommendation processing method of claim 6, the real encoding network is trained in the following manner: Inputting clothing data samples in the real world into the encoding network in the real data reconstruction model to be trained for feature encoding, and obtaining sample clothing features of the clothing data samples; Inputting the sample clothing features into a decoding network in the real data reconstruction model to perform data decoding, and obtaining sample reconstruction data of the sample clothing features; The training losses of the sample reconstruction data and the clothing data samples are calculated, and parameters of the real data reconstruction model are adjusted based on the training losses, and the encoding network in the trained real data reconstruction model is used as the real encoding network.

9. The cross-domain recommendation processing method according to claim 1, wherein determining the clothing recommendation features of the user based on the real clothing features and the user connection relationship network comprises: Inputting the real clothing feature into the graph model for feature determination to obtain the clothing recommendation feature; wherein, if the real clothing feature corresponds to an existing node in the user connection relationship network, reading the clothing recommendation feature corresponding to the existing node and outputting it; The graph model includes the user connection relationship network.

10. According to the cross-domain recommendation processing method of claim 9, the graph model is determined in the following manner: Determine the relevant real feature samples of each real feature sample, establish a connection relationship between each real feature sample and the corresponding relevant real feature sample, and obtain the user connection relationship network; Performing dimensionality reduction processing on two real feature samples having a connection relationship in the user connection relationship network; The dimension reduction results whose similarity between the two real feature samples is higher than the similarity of other dimension reduction results are used as the clothing recommendation features of each of the two real feature samples.

11. The cross-domain recommendation processing method according to claim 10, wherein the feature determination further comprises: If the real clothing feature does not correspond to an existing node in the user connection relationship network, the user connection relationship network is updated and trained based on the real clothing feature, and the clothing recommendation features corresponding to the existing nodes corresponding to the real clothing features in the trained user connection relationship network are read and output.

12. The cross-domain recommendation processing method according to claim 9, before inputting the real clothing features into the graph model for feature determination and obtaining the clothing recommendation features, further comprising: If the number of features of the real clothing features is at least two, calculating an average feature of the real clothing features; Accordingly, the step of inputting the real clothing features into the graph model to determine the features and obtaining the clothing recommendation features includes: The average feature is input into a graph model to perform feature determination, thereby obtaining the clothing recommendation feature.

13. The cross-domain recommendation processing method according to claim 1, wherein the step of performing clothing recommendation processing for the user based on the clothing recommendation feature comprises: Calculating the correlation between the clothing recommendation feature and each clothing feature in the clothing database; Target clothing with correlation higher than a recommendation threshold is screened out as recommended clothing for the user and recommended to the user.

14. A cross-domain recommendation processing device, comprising: A feature encoding module is configured to perform feature encoding on clothing data of virtual clothing of a user's virtual image in the virtual world to obtain virtual clothing features; A feature mapping module is configured to perform feature mapping from the virtual world to the real world for the virtual clothing features, so as to obtain the real clothing features of the user; A clothing recommendation feature determination module, configured to determine the clothing recommendation feature of the user based on the real clothing feature and the user connection relationship network; The user connection relationship network includes a real feature sample of a user and a connection relationship between real feature samples of related users, wherein the real feature sample is determined based on feature encoding of clothing data of real clothing of the real image of the user in the real world; the clothing recommendation feature of each of two real feature samples having a connection relationship in the user connection relationship network is determined based on a dimensionality reduction result of the dimensionality reduction results of the two real feature samples whose similarity is higher than the similarity of other dimensionality reduction results; The clothing recommendation processing module is configured to perform clothing recommendation processing for the user according to the clothing recommendation features.

15. A cross-domain recommendation processing device, comprising: processor; and a memory configured to store computer executable instructions that, when executed, cause the processor to: Characteristic encoding is performed on clothing data of virtual clothing of a virtual image of a user in the virtual world to obtain characteristics of the virtual clothing; Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user; Determining clothing recommendation features for the user based on the real clothing features and the user connection relationship network; The user connection relationship network includes a real feature sample of a user and a connection relationship between real feature samples of related users, wherein the real feature sample is determined based on feature encoding of clothing data of real clothing of the real image of the user in the real world; the clothing recommendation feature of each of two real feature samples having a connection relationship in the user connection relationship network is determined based on a dimensionality reduction result of the dimensionality reduction results of the two real feature samples whose similarity is higher than the similarity of other dimensionality reduction results; Perform clothing recommendation processing for the user according to the clothing recommendation features.

16. A storage medium for storing computer executable instructions, wherein the computer executable instructions, when executed by a processor, implement the following process: Characteristic encoding is performed on clothing data of virtual clothing of a virtual image of a user in the virtual world to obtain characteristics of the virtual clothing; Performing feature mapping from the virtual world to the real world for the virtual clothing features to obtain the real clothing features of the user; Determining clothing recommendation features for the user based on the real clothing features and the user connection relationship network; The user connection relationship network includes a real feature sample of a user and a connection relationship between real feature samples of related users, wherein the real feature sample is determined based on feature encoding of clothing data of real clothing of the real image of the user in the real world; the clothing recommendation feature of each of two real feature samples having a connection relationship in the user connection relationship network is determined based on a dimensionality reduction result of the dimensionality reduction results of the two real feature samples whose similarity is higher than the similarity of other dimensionality reduction results; Perform clothing recommendation processing for the user according to the clothing recommendation features.

Images