Data recommendation method, device, server and storage medium

Abstract

The present disclosure is about a data recommendation method, device, server and storage medium, and belongs to the field of computer technology. The method includes: obtaining target data features corresponding to the target multimedia data to be recommended, calling a feature conversion model, performing feature conversion processing on the target data features, obtaining the predicted operation features after the conversion, determining the matching degree between each user and the target multimedia data according to the user features and predicted operation features corresponding to multiple users, selecting at least one target user from multiple users according to the matching degrees corresponding to the multiple users, and recommending the target multimedia data to the at least one target user. The predicted operation features of the multimedia data are used to represent the operations that the user may perform on the multimedia data, determine the users matching the multimedia data, and subsequently recommend the multimedia data to the matching users, thereby realizing the accurate recommendation of the multimedia data, ensuring the accuracy of the data recommendation, and realizing the cold start of the multimedia data.

Description

Technical Field

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

[0002] With the development of computer technology and Internet technology, multimedia data on the Internet is becoming more and more abundant. Since there are too many multimedia data on the Internet, users cannot view them one by one. Therefore, multimedia data of interest is usually recommended to users.

[0003] In the related art, multimedia data whose data features match the user features are usually determined based on the data features of multiple multimedia data to be recommended and the user features of the target user, so as to recommend the multimedia data to the user. Since the above method only considers the data features of the multimedia data, it is impossible to achieve a cold start of the multimedia data. Summary of the invention

[0004] The present disclosure provides a data recommendation method, device, server and storage medium, which improve the accuracy of data recommendation.

[0005] According to one aspect of an embodiment of the present disclosure, a data recommendation method is provided, the method comprising:

[0006] Acquire target data features corresponding to target multimedia data to be recommended;

[0007] Calling a feature conversion model to perform feature conversion processing on the target data feature to obtain a transformed predicted operation feature, where the predicted operation feature is used to represent a predicted operation to be performed on the target multimedia data;

[0008] Determining a matching degree between each user and the target multimedia data according to user characteristics corresponding to a plurality of users and the predicted operation characteristics;

[0009] At least one target user is selected from the multiple users according to the matching degrees corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

[0010] In some embodiments, the method further comprises:

[0011] Calling the feature conversion model to perform feature conversion processing on the first sample data feature corresponding to the first sample multimedia data to obtain a converted first sample operation feature;

[0012] calling a feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, wherein the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature;

[0013] The feature conversion model is trained according to the first sample operation feature and the discriminant identifier.

[0014] In some embodiments, the training of the feature conversion model according to the first sample operation feature and the discriminant identifier includes:

[0015] Determining a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier;

[0016] The feature conversion model is trained according to the first loss value.

[0017] In some embodiments, after training the feature conversion model according to the first sample operation feature and the discriminant identifier, the method further includes:

[0018] Calling the feature conversion model to perform feature conversion processing on the second sample data feature corresponding to the second sample multimedia data to obtain a converted second sample operation feature;

[0019] Calling the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determining discrimination identifiers of the second sample data feature and the second sample operation feature, wherein the discrimination identifier is used to indicate whether the corresponding feature belongs to a data feature or an operation feature;

[0020] The feature discrimination model is trained according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

[0021] In some embodiments, the training of the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature includes:

[0022] Determining a second loss value of the feature discrimination model according to the second sample data feature, the second sample operation feature, and a discrimination identifier of the second sample data feature and the second sample operation feature;

[0023] The feature discrimination model is trained according to the second loss value.

[0024] In some embodiments, after training the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature, the method further includes:

[0025] Calling the feature conversion model to perform feature conversion processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature;

[0026] calling the feature discrimination model to perform discrimination processing on the third sample operation feature and determine a discrimination identifier of the third sample operation feature, wherein the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature;

[0027] The feature conversion model is trained according to the third sample operation feature and the discriminant identifier.

[0028] In some embodiments, the step of obtaining target data features corresponding to the target multimedia data to be recommended includes:

[0029] A feature extraction model is called to perform feature extraction on the target multimedia data to obtain target data features corresponding to the target multimedia data.

[0030] In some embodiments, the feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model, and calling the feature extraction model to extract features of the target multimedia data to obtain target data features corresponding to the target multimedia data includes:

[0031] Calling the text feature extraction sub-model to perform feature extraction on the target multimedia data to obtain text features of the target multimedia data;

[0032] Calling the image feature extraction sub-model to perform feature extraction on the target multimedia data to obtain image features of the target multimedia data;

[0033] The text feature and the image feature are concatenated to obtain the target data feature of the target multimedia data.

[0034] In some embodiments, determining the matching degree between each user and the target multimedia data according to the user characteristics corresponding to the plurality of users and the predicted operation characteristics includes:

[0035] The matching degree between each user and the target multimedia data is determined according to the user characteristics corresponding to the multiple users, the target data characteristics and the predicted operation characteristics.

[0036] In some embodiments, determining the matching degree between each user and the target multimedia data according to the user characteristics corresponding to the plurality of users, the target data characteristics and the predicted operation characteristics includes:

[0037] For the user features of any user, the user features, the target data features and the predicted operation features are matched to obtain the matching degree between the user and the target multimedia data.

[0038] In some embodiments, selecting at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommending the target multimedia data to the at least one target user includes:

[0039] Determining recommendation parameters corresponding to the multiple users according to the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data;

[0040] At least one target user is selected from the multiple users according to the recommendation parameters corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

[0041] According to another aspect of the embodiments of the present disclosure, a data recommendation device is provided, the device comprising:

[0042] A feature acquisition unit, configured to acquire target data features corresponding to the target multimedia data to be recommended;

[0043] a transformation processing unit configured to perform feature transformation processing on the target data feature to obtain a transformed predicted operation feature, wherein the predicted operation feature is used to represent a predicted operation to be performed on the target multimedia data;

[0044] a determination unit configured to determine a matching degree between each user and the target multimedia data according to user characteristics corresponding to a plurality of users and the predicted operation characteristics;

[0045] The recommendation unit is configured to select at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommend the target multimedia data to the at least one target user.

[0046] In some embodiments, the apparatus further comprises:

[0047] The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the first sample data feature corresponding to the first sample multimedia data to obtain a transformed first sample operation feature;

[0048] a discrimination processing unit configured to call a feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, wherein the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature;

[0049] The first training unit is configured to train the feature conversion model according to the first sample operation feature and the discriminant identifier.

[0050] In some embodiments, the first training unit includes:

[0051] A first determining subunit is configured to determine a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier;

[0052] The first training subunit is configured to train the feature conversion model according to the first loss value.

[0053] In some embodiments, the apparatus further comprises:

[0054] The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the second sample data feature corresponding to the second sample multimedia data to obtain a transformed second sample operation feature;

[0055] The discrimination processing unit is further configured to call the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determine discrimination identifiers of the second sample data feature and the second sample operation feature, wherein the discrimination identifier is used to indicate whether the corresponding feature belongs to a data feature or an operation feature;

[0056] The second training unit is configured to train the feature discrimination model according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

[0057] In some embodiments, the second training unit includes:

[0058] A second determining subunit is configured to determine a second loss value of the feature discriminant model according to the second sample data feature, the second sample operation feature, and a discriminant identifier of the second sample data feature and the second sample operation feature;

[0059] The second training subunit is configured to train the feature discrimination model according to the second loss value.

[0060] In some embodiments, the apparatus further comprises:

[0061] The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature;

[0062] The discrimination processing unit is further configured to call the feature discrimination model, perform discrimination processing on the third sample operation feature, and determine a discrimination identifier of the third sample operation feature, wherein the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature;

[0063] The third training unit is configured to train the feature conversion model according to the third sample operation feature and the discriminant identifier.

[0064] In some embodiments, the feature acquisition unit includes:

[0065] The feature extraction subunit is configured to call a feature extraction model to perform feature extraction on the target multimedia data to obtain target data features corresponding to the target multimedia data.

[0066] In some embodiments, the feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model, and the feature extraction sub-unit is configured to call the text feature extraction sub-model to perform feature extraction on the target multimedia data to obtain text features of the target multimedia data; call the image feature extraction sub-model to perform feature extraction on the target multimedia data to obtain image features of the target multimedia data; and concatenate the text features with the image features to obtain target data features of the target multimedia data.

[0067] In some embodiments, the determining unit includes:

[0068] The third determining subunit is configured to determine the matching degree between each user and the target multimedia data according to user characteristics corresponding to a plurality of users, the target data characteristics and the predicted operation characteristics.

[0069] In some embodiments, the third determination subunit is configured to match the user characteristics, the target data characteristics and the predicted operation characteristics of any user to obtain a matching degree between the user and the target multimedia data.

[0070] In some embodiments, the recommendation unit is configured to determine recommendation parameters corresponding to the multiple users based on the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data; select at least one target user from the multiple users based on the recommendation parameters corresponding to the multiple users, and recommend the target multimedia data to the at least one target user.

[0071] According to another aspect of an embodiment of the present disclosure, a server is provided, the server comprising:

[0072] one or more processors;

[0073] volatile or non-volatile memory for storing the one or more processor-executable instructions;

[0074] The one or more processors are configured to execute the data recommendation method described in the first aspect.

[0075] According to another aspect of the embodiments of the present disclosure, a non-temporary computer-readable storage medium is provided. When the instructions in the storage medium are executed by the processor of the server, the server can execute the data recommendation method described in the above aspects.

[0076] According to another aspect of an embodiment of the present disclosure, a computer program product is provided. When instructions in the computer program product are executed by a processor of a server, the server is enabled to execute the data recommendation method described in the above aspects.

[0077] The data recommendation method, device, server and storage medium provided by the embodiments of the present disclosure use the predicted operation characteristics of multimedia data to represent the operations that a user may perform on the multimedia data, thereby determining the users who may perform operations on the multimedia data based on the predicted operation characteristics, i.e., the users who match the multimedia data, and subsequently recommending the multimedia data to the matched users, thereby achieving accurate recommendation of the multimedia data. Regardless of whether the operations performed on the multimedia data can be obtained, the accuracy of the data recommendation is guaranteed, and even if the operations performed on the multimedia data have not been obtained, the multimedia data can be recommended, i.e., a cold start of the multimedia data is achieved.

[0078] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0080] Figure 1 It is a schematic diagram showing an implementation environment according to an exemplary embodiment.

[0081] Figure 2 The figure is a flow chart of a data recommendation method according to an exemplary embodiment.

[0082] Figure 3 The figure is a flow chart of a data recommendation method according to an exemplary embodiment.

[0083] Figure 4 The figure is a flow chart of obtaining data features according to an exemplary embodiment.

[0084] Figure 5 The present invention is a flow chart showing a method of obtaining an operation feature corresponding to a data feature according to an exemplary embodiment.

[0085] Figure 6 The figure is a flow chart of obtaining a matching degree according to an exemplary embodiment.

[0086] Figure 7 The figure is a flowchart of a method for recommending data to a user according to an exemplary embodiment.

[0087] Figure 8 The figure is a flow chart of a data recommendation method according to an exemplary embodiment.

[0088] Fig. 9 A method for training a feature conversion model is shown according to an exemplary embodiment.

[0089] Fig.10 The present invention is a flowchart showing a method of obtaining a discrimination identifier of an input feature according to an exemplary embodiment.

[0090] Fig.11 It is a block diagram of a data recommendation device according to an exemplary embodiment.

[0091] Fig.12 It is a block diagram of a data recommendation device according to an exemplary embodiment.

[0092] Fig.13 It is a block diagram of a terminal according to an exemplary embodiment.

[0093] Fig.14 It is a block diagram of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0094] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0095] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned figure descriptions are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0096] The terms "at least one", "a plurality", "each", and "any" used in the present disclosure include one, two, or more than two, a plurality includes two or more than two, each refers to each of the corresponding plurality, and any refers to any one of the plurality. For example, the plurality of users includes three users, and each refers to each of the three users, and any refers to any one of the three users, which may be the first user, the second user, or the third user.

[0097] It should be noted that the user information involved in this disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0098] Figure 1 It is a schematic diagram of an implementation environment provided according to an exemplary embodiment. The implementation environment includes: a terminal 101 and a server 102. The terminal 101 is connected to the server 102 through a network and can interact with the server 102 through the network connection.

[0099] In some embodiments, the terminal is a terminal of various types such as a mobile phone, a tablet computer or a computer. In some embodiments, the server is a server, or a server cluster composed of several servers, or a cloud computing service center.

[0100] The server 102 has a data processing function and a data recommendation function, can process multimedia data, and can recommend multimedia data to users. The terminal 101 has a multimedia data display function, can display multimedia data recommended by the server 102 for users to view.

[0101] In some embodiments, the terminal 101 installs an application provided by the server 102, for example, the application is a shopping application, a video application, etc. The terminal 101 can run the application and display the multimedia data in the application to the user. The server 102 can recommend multimedia data to the user so that it can be displayed in the application for the user to view.

[0102] The method provided by the embodiments of the present disclosure can be applied in a variety of scenarios.

[0103] For example, in the advertising recommendation scenario.

[0104] When a new advertisement is delivered, since the new advertisement has not been delivered before or has been delivered for a short time, it is impossible to obtain the operations performed by the user on the advertisement, for example, it is impossible to obtain the number of times the user has viewed multimedia data, or it is impossible to obtain the click rate of the user on the multimedia data, etc. In this case, the method provided by the embodiment of the present disclosure is used to recommend the advertisement to matching users through the obtained predicted operation features, thereby improving the accuracy of the advertisement recommendation and achieving a cold start of the advertisement.

[0105] Figure 2 is a flow chart of a data recommendation method according to an exemplary embodiment. Figure 2 The method is applied in a server and comprises the following steps:

[0106] In step 201, the server obtains target data features corresponding to target multimedia data to be recommended.

[0107] In step 202, the server calls the feature conversion model to perform feature conversion processing on the target data features to obtain the transformed predicted operation features, where the predicted operation features are used to represent the predicted operations to be performed on the target multimedia data.

[0108] In step 203, the server uses user characteristics and predicted operation characteristics corresponding to multiple users to determine the matching degree between each user and the target multimedia data.

[0109] In step 204, the server selects at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommends the target multimedia data to the at least one target user.

[0110] The method provided by the embodiments of the present disclosure uses the predicted operation features of multimedia data to represent the operations that a user may perform on the multimedia data, thereby determining the users who may perform operations on the multimedia data, i.e., the users who match the multimedia data, based on the predicted operation features. The multimedia data is then recommended to the matched users, thereby achieving accurate recommendation of the multimedia data. Regardless of whether the operations performed on the multimedia data can be obtained, the accuracy of the data recommendation is guaranteed. Even if the operations performed on the multimedia data have not been obtained, the multimedia data can be recommended, thereby achieving a cold start of the multimedia data.

[0111] In some embodiments, the method further comprises:

[0112] Calling the feature conversion model to perform feature conversion processing on the first sample data feature corresponding to the first sample multimedia data to obtain a converted first sample operation feature;

[0113] Calling a feature discrimination model to discriminate the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, where the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature;

[0114] The feature conversion model is trained according to the first sample operation characteristics and the discriminant identifier.

[0115] In some embodiments, training a feature conversion model according to the first sample operation feature and the discriminant identifier includes:

[0116] Determining a first loss value of a feature conversion model according to the first sample operation feature and the discriminant identifier;

[0117] According to the first loss value, the feature conversion model is trained.

[0118] In some embodiments, after training the feature conversion model according to the first sample operation feature and the discriminant identifier, the method further includes:

[0119] Calling the feature conversion model to perform feature conversion processing on the second sample data feature corresponding to the second sample multimedia data to obtain a converted second sample operation feature;

[0120] Calling a feature discrimination model to discriminate the second sample data feature and the second sample operation feature respectively, and determining discrimination identifiers of the second sample data feature and the second sample operation feature, where the discrimination identifier is used to indicate whether the corresponding feature is a data feature or an operation feature;

[0121] The feature discrimination model is trained according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature.

[0122] In some embodiments, training a feature discrimination model according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature includes:

[0123] Determine a second loss value of the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature;

[0124] According to the second loss value, the feature discrimination model is trained.

[0125] In some embodiments, after training the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature, the method further includes:

[0126] Calling the feature conversion model to perform feature conversion processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature;

[0127] Calling a feature discrimination model to perform discrimination processing on the third sample operation feature and determine a discrimination identifier of the third sample operation feature, where the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature;

[0128] The feature conversion model is trained according to the third sample operation characteristics and the discriminant identifier.

[0129] In some embodiments, obtaining target data features corresponding to target multimedia data to be recommended includes:

[0130] The feature extraction model is called to extract features of the target multimedia data to obtain target data features corresponding to the target multimedia data.

[0131] In some embodiments, the feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model, and calling the feature extraction model to extract features of the target multimedia data to obtain target data features corresponding to the target multimedia data includes:

[0132] Calling the text feature extraction sub-model to extract features of the target multimedia data to obtain text features of the target multimedia data;

[0133] Calling the image feature extraction sub-model to extract features of the target multimedia data to obtain image features of the target multimedia data;

[0134] The text features and image features are concatenated to obtain the target data features of the target multimedia data.

[0135] In some embodiments, determining the matching degree between each user and the target multimedia data according to the user characteristics corresponding to the plurality of users and the predicted operation characteristics includes:

[0136] The matching degree between each user and the target multimedia data is determined according to user characteristics, target data characteristics and predicted operation characteristics corresponding to the multiple users.

[0137] In some embodiments, determining the matching degree between each user and the target multimedia data according to the user characteristics, target data characteristics and predicted operation characteristics corresponding to the plurality of users includes:

[0138] For the user characteristics of any user, the user characteristics, target data characteristics and predicted operation characteristics are processed to obtain the matching degree between the user and the target multimedia data.

[0139] In some embodiments, selecting at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommending the target multimedia data to the at least one target user includes:

[0140] Determining recommendation parameters corresponding to the multiple users according to the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data;

[0141] At least one target user is selected from the multiple users according to the recommendation parameters corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

[0142] Figure 3 is a flow chart of a data recommendation method according to an exemplary embodiment. Figure 3 The method is applied in a server and comprises the following steps:

[0143] In step 301, the server calls a feature extraction model to extract features from target multimedia data to obtain target data features corresponding to the target multimedia data.

[0144] In the embodiments of the present disclosure, the target multimedia data is multimedia data that has been released for a short time or has not been released yet among the multimedia data to be recommended. In some embodiments, the target multimedia data is an advertisement, a video, a novel, etc. Since the target multimedia data has been released for a short time, few operations performed by the user on the target multimedia data are obtained, and the target multimedia data cannot be accurately recommended based on the operations performed by the user on the target multimedia data. Alternatively, since the target multimedia data has not been released yet, the operations performed by the user on the target multimedia data cannot be obtained, such as the number of times the user has viewed the multimedia data, or the click rate of the user on the multimedia data. Therefore, by predicting the operations performed by the user on the target multimedia data, the operation characteristics corresponding to the target multimedia data are obtained, so that the target multimedia data can be accurately recommended based on the operation characteristics in the future, thereby achieving a cold start of the target multimedia data.

[0145] The feature extraction model is used to extract data features of multimedia data, and the corresponding multimedia data is represented by the data features. In some embodiments, the target data features include data feature vectors or data feature matrices. The target data features corresponding to the target multimedia data are extracted by the feature extraction model to ensure the accuracy of the target data features, and the target data features are subsequently used to obtain the predicted operation features of the target multimedia data.

[0146] Since the server stores multiple multimedia data, the process of selecting the target multimedia from the stored multiple multimedia data includes the following two methods:

[0147] The first method is to store a plurality of multimedia data and the delivery time of each multimedia data in a database, and then, according to the delivery time of the plurality of multimedia data, use the multimedia data whose delivery time is shorter than the reference time as the target multimedia data.

[0148] The delivery time indicates the time for delivering the multimedia data. The reference time is any time, such as 10 hours, 2 days, etc. After any multimedia data is delivered, the user can view the multimedia data.

[0149] Because when multimedia data is just delivered, the operations performed by users on the multimedia data are limited during the delivery process, it is impossible to accurately obtain the operations performed by users on the multimedia data, so that it is impossible to achieve accurate recommendation of multimedia data based on the obtained operations. Therefore, multimedia data with a delivery duration shorter than the reference duration is used as target multimedia data.

[0150] The second method is to store a plurality of multimedia data and the operation times corresponding to each multimedia data in a database, and then, according to the operation times corresponding to the plurality of multimedia data, use the multimedia data whose operation times are less than the reference times as the target multimedia data.

[0151] After any multimedia data is delivered, if any user performs an operation on the multimedia data, the number of operations corresponding to the multimedia data increases by 1. In some embodiments, the number of operations is the number of clicks.

[0152] When the number of operations performed by the user on the multimedia data is low, it is impossible to accurately recommend the multimedia data based on the operations performed by the user on the multimedia data. Therefore, multimedia data with a delivery duration shorter than the reference duration is used as the target multimedia data.

[0153] In some embodiments, the feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model, and step 301 includes the following steps 3011-3013:

[0154] 3011. Call the text feature extraction sub-model to extract features of the target multimedia data to obtain text features of the target multimedia data.

[0155] In the embodiment of the present disclosure, the target multimedia data includes text and images. For example, the multimedia data is a video, and the video includes text and images. Features of the target multimedia data are extracted from the text dimension and the image dimension respectively, and the obtained text features and image features are spliced ​​to enrich the data contained in the target data features corresponding to the target multimedia data, thereby improving the accuracy of the target data features. The process of obtaining the target data features through the text feature extraction submodel and the image feature extraction submodel, such as Figure 4 shown.

[0156] The text feature extraction submodel is used to extract text features from multimedia data. In some embodiments, the text feature extraction submodel is BERT (Bidirectional Encoder Representations from Transformers), or other models. The text feature is used to represent the text contained in the target multimedia data. In some embodiments, the text feature includes a text feature vector or a text feature matrix.

[0157] Since the text feature extraction sub-model is a model for extracting text features in multimedia data, the accuracy of the text features is ensured by extracting text features in target multimedia data using the text feature extraction sub-model.

[0158] 3012. Call the image feature extraction sub-model to extract features of the target multimedia data to obtain image features of the target multimedia data.

[0159] The image feature extraction submodel is used to extract image features from the target multimedia data. In some embodiments, the image feature extraction submodel is VGG-16 (Visual Geometry Group Network), or other models. The image features are used to represent the images contained in the target multimedia data. In some embodiments, the image features include image feature vectors or image feature matrices.

[0160] Since the image feature extraction sub-model is a model for extracting image features in multimedia data, the image features in the target multimedia data are extracted using the image feature extraction sub-model, thereby ensuring the accuracy of the image features.

[0161] 3013. Concatenate the text features with the image features to obtain target data features of the target multimedia data.

[0162] Since text features and image features describe the target multimedia data from different dimensions, by splicing text features and image features together as the target data features of the target multimedia data, the data contained in the target data features is enriched, thereby improving the accuracy of the target data features.

[0163] In some embodiments, the text feature is a text feature vector, and the image feature is an image feature vector, then step 3013 includes: concatenating the text feature vector and the image feature vector to obtain a target data feature vector of the target multimedia data.

[0164] In some embodiments, the sum of the number of feature dimensions included in the text feature vector and the number of feature dimensions included in the image feature vector is equal to the number of feature dimensions included in the target data feature vector. For example, the text feature vector includes 4 feature dimensions, and the text feature vector is (1, 1, 1, 1), and the image feature vector includes 3 feature dimensions, and the image feature vector is (2, 2, 2), then the obtained target data feature vector includes 7 feature dimensions, and the target data feature vector is (1, 1, 1, 1, 2, 2, 2), or the target data feature vector is (2, 2, 2, 1, 1, 1, 1).

[0165] It should be noted that the embodiment of the present disclosure obtains the target data features corresponding to the target multimedia data through a feature extraction model, while in another embodiment, the server can adopt other methods to obtain the target data features corresponding to the target multimedia data to be recommended.

[0166] In step 302, the server calls the feature conversion model to perform feature conversion processing on the target data features to obtain the transformed predicted operation features.

[0167] The feature conversion model is used to obtain the predicted operation features of the multimedia data, and the predicted operation features are used to represent the predicted operations performed on the target multimedia data. For example, the predicted operation features represent the number of times the user has viewed the multimedia data, or the operation features represent the click rate of the user on the multimedia data. In some embodiments, the form of the predicted operation features includes but is not limited to an operation feature vector or an operation feature matrix.

[0168] Through the feature conversion model, the target data features used to represent the target multimedia data are transformed to obtain the transformed prediction features to represent the predicted operations performed by the user on the target multimedia data, so that the users matching the target multimedia data can be determined based on the predicted operations performed by the user on the target multimedia data, thereby ensuring the accurate recommendation of the multimedia data and realizing the cold start of the multimedia data.

[0169] In some embodiments, the target data feature is a data feature vector, the predicted operation feature is an operation feature vector, and the number of feature dimensions of the data feature vector is equal to the number of feature dimensions of the operation feature vector.

[0170] In some embodiments, the target data feature is a data feature vector, and the feature conversion model includes at least two hidden layers, then step 302 includes: based on the first hidden layer of the at least two hidden layers, performing feature conversion processing on the data feature, inputting the feature output by the first hidden layer to the next hidden layer, based on the next hidden layer, performing feature conversion processing on the feature output by the previous hidden layer, and using the feature output by the most hidden layer of the at least two hidden layers as the prediction operation feature.

[0171] Through multiple hidden layers in the feature conversion model, the target data features are transformed to predict the operation performed by the user on the target multimedia data, and the predicted operation features corresponding to the target data features are obtained, that is, the predicted operation features corresponding to the target multimedia data are obtained. Figure 5 As shown, the predicted operation features corresponding to the target data features are obtained through the two hidden layers in the feature conversion model.

[0172] In some embodiments, each hidden layer corresponds to a feature transformation matrix, and different hidden layers correspond to different feature transformation matrices. The target data features are transformed by the feature transformation matrices corresponding to the multiple hidden layers to obtain the predicted operation features of the target multimedia data.

[0173] It should be noted that the embodiment of the present disclosure is described by taking the example of obtaining the predicted operation features corresponding to the target multimedia data through the feature conversion model. In another embodiment, there is no need to execute step 302, and other methods can be adopted to perform feature transformation processing on the target data features to obtain the converted predicted operation features.

[0174] In step 303, the server determines the matching degree between each user and the target multimedia data according to the user characteristics, target data characteristics and predicted operation characteristics corresponding to the multiple users.

[0175] The matching degree is used to indicate the user's preference for the multimedia data. The greater the matching degree, the higher the user's preference for the multimedia data. The smaller the matching degree, the lower the user's preference for the multimedia data. User features are used to describe the user. In some embodiments, the user features include user feature vectors or user feature matrices. In some embodiments, the user features of the user are obtained by extracting features from the user information. The user information includes gender, age, place of residence, occupation, preferences, etc.

[0176] Through user characteristics, target data characteristics and predicted operation characteristics, it is possible to determine the degree of preference of each user for the target multimedia data, that is, determine the matching degree between each user and the target multimedia data, so that the target multimedia data can be recommended to matching users based on the matching degree between each user and the target multimedia data.

[0177] In some embodiments, step 303 includes: for user features of any user, matching the user features, target data features and predicted operation features to obtain a matching degree between the user and the target multimedia data.

[0178] Since the target data features and the predicted operation features describe the target multimedia data from different dimensions, and the user features describe the user, a matching process is performed through the user features, the target data features and the predicted operation features to determine the matching degree between the user features, the target data features and the predicted operation features, that is, to obtain the matching degree between the user and the target multimedia data.

[0179] In some embodiments, a feature matching model is called to match the user features, target data features, and predicted operation features to obtain the matching degree between the user and the target multimedia data. The feature matching model is used to obtain the matching degree between the user and the multimedia data. After obtaining the user features, the target data features corresponding to the multimedia data, and the predicted operation features, the matching degree between the user and the multimedia data can be obtained through the feature matching model. Figure 6 As shown, user features, target data features and predicted operation features are input into the feature matching model, the features are matched through the hidden layer in the feature matching model, and the matching degree between the user and the target multimedia data is output through the output layer.

[0180] In some embodiments, the user feature is a user feature vector, the target data feature is a data feature vector, the predicted operation feature is an operation feature vector, and the user feature vector, the data feature vector, and the operation feature vector all include feature values ​​of the same number of dimensions. The feature values ​​of the same dimensions of the user feature vector, the data feature vector, and the operation feature vector are statistically processed to obtain a fused feature vector, and the fused feature values ​​of multiple feature dimensions of the fused feature vector are statistically processed to obtain a matching degree between the user and the target multimedia data.

[0181] In some embodiments, when statistics are taken on the feature values ​​of the same dimension of the user feature vector, the data feature vector, and the operation feature vector, the sum of the feature values ​​of the same dimension is taken as the fused feature value of the corresponding dimension; alternatively, the feature values ​​of the same dimension are weighted averaged to obtain the fused feature value of the corresponding dimension.

[0182] In some embodiments, when processing the fused feature values ​​of multiple feature dimensions of the fused feature vector, the fused feature values ​​of multiple feature dimensions are weighted averaged to obtain the matching degree; or, the sum of the fused feature values ​​of multiple feature dimensions is taken as the matching degree.

[0183] In some embodiments, the target multimedia data corresponds to a specified user characteristic, and the specified user characteristic indicates that the target multimedia data is to be recommended to a user who meets the specified user characteristic. Then, before step 303, the method includes: the server selects a reference user whose user characteristics match the specified user characteristics based on user characteristics corresponding to multiple reference users and the specified user characteristics, as the user to whom the target multimedia data is to be recommended.

[0184] For example, if the user characteristic is specified as: female, then female users are selected from multiple reference users as users to be recommended for target multimedia data; or, if the user characteristic is specified as: living in XX city, then users living in XX city are selected from multiple reference users as users to be recommended for target multimedia data.

[0185] In step 304, the server selects at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommends the target multimedia data to the at least one target user.

[0186] According to the matching degree between each user and the multimedia data, a user matching the target multimedia data is selected from multiple users as a target user, and the target multimedia data is recommended to the target user.

[0187] In some embodiments, among the matching degrees corresponding to the multiple users, the matching degree corresponding to the target user is greater than the matching degrees corresponding to other users among the multiple users.

[0188] In some embodiments, step 304 includes: determining recommendation parameters corresponding to multiple users based on the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data, selecting at least one target user from the multiple users based on the recommendation parameters corresponding to the multiple users, and recommending the target multimedia data to the at least one target user.

[0189] The number of resources is the number of resources that can be obtained by recommending the target multimedia data. The recommendation parameter is used to indicate the possibility of recommendation to the user. The higher the recommendation parameter, the greater the possibility of recommending the target multimedia resource to the corresponding user. The lower the recommendation parameter, the smaller the possibility of recommending the target multimedia resource to the corresponding user. By selecting the target user to be recommended from multiple users according to the recommendation parameter, the maximum number of resources can be obtained while ensuring that the target multimedia data is accurately recommended.

[0190] In some embodiments, for any user, the product of the user matching degree and the number of resources corresponding to the target multimedia data is used as the recommendation parameter corresponding to the user.

[0191] In some embodiments, the recommendation parameters corresponding to the multiple users are arranged in descending order to select at least one target user, wherein the recommendation parameter of the target user is greater than the recommendation parameters of other users among the multiple users.

[0192] It should be noted that the disclosed embodiment recommends target multimedia data to target users based on data features, operation features and user features. In another embodiment, step 304 does not need to be executed, and the target multimedia data is recommended directly based on the operation features.

[0193] When recommending multimedia data, if the multimedia data has been released for a long time, the user's operation on the multimedia data can be obtained, such as the number of times the user views the multimedia data, or the click rate of the multimedia data, etc. When the multimedia data is recommended to other users, the matching user can be determined based on the obtained user's operation on the multimedia data, so that the multimedia data can be recommended to the determined user. If the multimedia data is multimedia data that has not been released yet, or multimedia data that has been released but in the release time period, the user's operation on the multimedia data cannot be obtained at this time, or the user's operation on the multimedia data is obtained. Few operations performed by the user on the multimedia data cannot be accurately recommended, resulting in the inability to accurately recommend the multimedia data based on the user's operation on the multimedia data, so that the multimedia data cannot be recommended to the user, and further lack of user operations on the multimedia data forms a vicious circle, so that the multimedia data cannot be accurately recommended, that is, the cold start of the multimedia data cannot be achieved. Therefore, in order to achieve the cold start of multimedia data, through the method provided by the present invention, when the operations performed on the multimedia data cannot be obtained, the user's operations on the multimedia data are predicted, and then the matching users are determined based on the predicted user operations on the multimedia data, and the multimedia data are recommended to the matching users, thereby achieving accurate recommendation of multimedia data, ensuring the accuracy of data recommendation, and achieving the cold start of multimedia data.

[0194] The method provided by the embodiments of the present disclosure uses the predicted operation features of multimedia data to represent the operations that a user may perform on the multimedia data, thereby determining the users who may perform operations on the multimedia data, i.e., the users who match the multimedia data, based on the predicted operation features. The multimedia data is then recommended to the matched users, thereby achieving accurate recommendation of the multimedia data. Regardless of whether the operations performed on the multimedia data can be obtained, the accuracy of the data recommendation is guaranteed. Even if the operations performed on the multimedia data have not been obtained, the multimedia data can be recommended, thereby achieving a cold start of the multimedia data.

[0195] Furthermore, multimedia data is recommended according to the matching degree between the user and the multimedia data, so as to improve the accuracy of data recommendation.

[0196] Based on the above embodiments, the present disclosure provides a method for recommending data to a user. Figure 7 As shown, applied in a server, the method includes:

[0197] In step 701, the server selects multimedia data whose designated user characteristics match the target user characteristics from multiple multimedia data according to the target user characteristics corresponding to the target user and the designated user characteristics corresponding to multiple multimedia data as the first reference multimedia data to be recommended.

[0198] This step is similar to the solution in the above step 303 in which the server selects a user to whom target multimedia data is to be recommended from multiple reference users, and will not be described in detail here.

[0199] In step 702, the server selects a plurality of second reference multimedia data according to the number of resources corresponding to the selected plurality of first reference multimedia data.

[0200] The number of resources corresponding to the second reference multimedia data is greater than the number of resources corresponding to other reference multimedia data in the plurality of first reference multimedia data.

[0201] In the process of selecting the second reference multimedia resource by the server, the lightweight model can be called to select the second reference multimedia resource data with a large resource quantity from the resource quantities corresponding to the plurality of first reference multimedia data.

[0202] In step 703, the server calls the feature extraction model to perform feature extraction on each second reference multimedia data respectively to obtain the target data feature corresponding to each second reference multimedia data.

[0203] This step is similar to the above step 301 and will not be described again here.

[0204] In step 704, the server calls the feature conversion model to perform feature conversion processing on the target data feature corresponding to each second reference multimedia data to obtain the predicted operation feature corresponding to each second reference multimedia data.

[0205] This step is similar to the above step 302 and will not be described again here.

[0206] In step 705, for any second reference multimedia data, the server determines the matching degree between the target user and the second reference multimedia data according to the target user characteristics, the target data characteristics corresponding to the second reference multimedia data, and the predicted operation characteristics.

[0207] This step is similar to the above step 303 and will not be described again here.

[0208] In step 706, the server determines the recommended parameters of each second reference multimedia resource according to the matching degree and the resource quantity corresponding to each second reference multimedia data.

[0209] This step is similar to the solution of obtaining recommendation parameters corresponding to multiple users in the above step 304, and will not be repeated here.

[0210] In step 707, the server selects at least one target multimedia resource from the plurality of second reference multimedia resources according to the recommendation parameters of the plurality of second reference multimedia resources and recommends it to the target user.

[0211] The recommendation parameter corresponding to the target multimedia resource is greater than the recommendation parameters of other multimedia resources in the plurality of second reference multimedia resources. By selecting the target multimedia resource to be recommended from the plurality of second reference multimedia resources according to the recommendation parameter, the matching target multimedia resource can be recommended to the user while ensuring that a maximum number of resources are obtained.

[0212] In the disclosed embodiment, after multimedia data is accurately recommended to the user, the corresponding number of resources can be obtained. For example, after an advertisement is accurately recommended to the user, a certain amount of revenue can be obtained. Therefore, when recommending multimedia data to a target user, if the operation to be performed on the recommended multimedia data cannot be obtained, in order to achieve a cold start of the multimedia data and obtain a large number of resources, multimedia data with a large number of corresponding resources is selected, and the operation performed by the user on each multimedia resource data is predicted, so that the multimedia resources matching the target user can be determined and the matching multimedia resources can be recommended to the target user, which not only ensures the accuracy of multimedia data recommendation, but also can obtain a large number of resources and achieve a cold start of multimedia resources.

[0213] The method provided by the embodiments of the present disclosure uses the predicted operation features of multimedia data to represent the operations that a user may perform on the multimedia data, thereby determining the users who may perform operations on the multimedia data, i.e., the users who match the multimedia data, based on the predicted operation features. The multimedia data is then recommended to the matched users, thereby achieving accurate recommendation of the multimedia data. Regardless of whether the operations performed on the multimedia data can be obtained, the accuracy of the data recommendation is guaranteed. Even if the operations performed on the multimedia data have not been obtained, the multimedia data can be recommended, thereby achieving a cold start of the multimedia data.

[0214] Based on the steps in the above embodiments, the present disclosure provides a data recommendation process, such as Figure 8 As shown, the recommended process includes:

[0215] 1. Based on the user characteristics of the target user, multiple first reference multimedia data matching the target user are directed.

[0216] 2. Determine a plurality of second reference multimedia data with a large number of resources by recalling the plurality of first reference multimedia data.

[0217] 3. Recall each second reference multimedia data through the neural network model to determine the recommended parameters of each second reference multimedia data.

[0218] 4. According to the recommendation parameters of the plurality of second reference multimedia data, bid on the plurality of second reference multimedia data, select the target multimedia data, and recommend the target multimedia data to the target user.

[0219] exist Figure 3 and Figure 7 Based on the illustrated embodiment, before calling the feature conversion model, the feature conversion model needs to be trained. The training process is detailed in the following embodiment.

[0220] Fig. 9 This is a training method for a feature conversion model provided by an embodiment of the present disclosure, referring to Fig. 9 , applied in a server, the method comprises:

[0221] In step 901, the server calls a feature conversion model to perform feature conversion processing on first sample data features corresponding to first sample multimedia data to obtain converted first sample operation features.

[0222] In the disclosed embodiment, the feature conversion model and the feature discrimination model are alternately and iteratively trained by using the feature conversion model and the feature discrimination model for adversarial training to obtain an accurate feature conversion model. For example, the feature conversion model is first trained, and then the feature discrimination model is trained based on the trained feature conversion model, and then the feature conversion model is trained again by the trained feature discrimination model, and the above process is repeated to train the feature conversion model and the feature discrimination model respectively to improve the accuracy of the feature conversion model.

[0223] The first sample multimedia data is any multimedia data. Acquiring the first sample data feature corresponding to the first sample multimedia data is similar to the above step 301; the step 901 is similar to the above step 302, and will not be described in detail here.

[0224] In step 902, the server calls the feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature.

[0225] Among them, the feature discrimination model is used to discriminate whether the input feature is a data feature or an operation feature. The discrimination mark is used to indicate whether the first sample operation feature belongs to a data feature or an operation feature. In some embodiments, the discrimination mark is a probability, indicating the probability that the input feature is a data feature, or indicating the probability that the input feature is an operation feature. For example, the probability range is (0,1). If the probability is 0, it indicates that the discriminated feature is an operation feature. If the probability is 1, it indicates that the discriminated feature is a data feature. Or, if the probability is 0, it indicates that the discriminated feature is a data feature. If the probability is 1, it indicates that the discriminated feature is an operation feature. The process of obtaining the discrimination mark of the input feature through the feature discrimination model, such as Fig.10 shown.

[0226] The first sample operation feature is discriminated by the feature discrimination model to determine whether the first sample operation feature generated by the feature conversion model belongs to a data feature or an operation feature, thereby determining whether the first sample operation feature generated by the feature conversion model is accurate according to the discrimination identifier.

[0227] In step 903, the server trains the feature conversion model according to the first sample operation feature and the discriminant identifier.

[0228] Through the sample operation feature and the discriminant mark, it is determined whether the first sample operation feature generated by the feature conversion model is accurate, so that the feature conversion model can be adjusted according to the discriminant mark to improve the accuracy of the feature conversion model.

[0229] In some embodiments, step 903 includes: determining a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier, and training the feature conversion model according to the first loss value.

[0230] The first loss value is used to represent the inaccuracy of the current feature conversion model. The feature conversion model is trained by the first loss value to reduce the loss value of the feature conversion model, thereby improving the accuracy of the feature conversion model.

[0231] In some embodiments, the first sample data feature, the first sample operation feature and the first loss value satisfy the following relationship:

[0232]

[0233] in, represents the first loss value; G(z) represents the first sample operation feature; D(G(z)) represents the identification mark of the first sample operation feature; z~P z (z) represents the distribution of the first sample data feature z of multiple first sample multimedia data; E( ) represents the mean of the loss values ​​of multiple sample multimedia data.

[0234] It should be noted that in the embodiment of the present disclosure, when training the feature conversion model, only one training is used as an example for explanation, while in another embodiment, the feature conversion model is trained multiple times using multiple first sample multimedia data.

[0235] In some embodiments, the feature conversion model is iteratively trained using multiple first sample multimedia data, and in response to a loss value obtained in a current training round being less than a reference loss value, the training of the feature conversion model is stopped; or, in response to the training round of the feature conversion model reaching a reference round, the training of the feature conversion model is stopped.

[0236] The feature conversion model is trained once using a first sample multimedia data, which represents a training round. The reference loss value is any value, such as 0.3 or 0.4. The reference round is any round, for example, the reference round is 10 times or 15 times.

[0237] After the feature conversion model is trained multiple times using multiple first sample multimedia data, when the condition for stopping training of the feature conversion model is reached, it means that based on the current feature discrimination model, it is no longer possible to identify whether the sample operation features generated by the feature conversion model belong to operation features or data features. Therefore, the feature discrimination model needs to be trained subsequently so that the feature conversion model can continue to be trained through the trained feature discrimination model.

[0238] In step 904, the server calls the feature conversion model to perform feature conversion processing on the second sample data features corresponding to the second sample multimedia data to obtain converted second sample operation features.

[0239] The second sample multimedia data is different from the first sample multimedia data. Step 904 is similar to step 901, and will not be described in detail herein.

[0240] In step 905, the server calls the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determines the discrimination identifiers of the second sample data feature and the second sample operation feature.

[0241] The discrimination flag is used to indicate whether the corresponding feature belongs to a data feature or an operation feature.

[0242] Since the feature discrimination model is used to discriminate whether the input features belong to data features or operation features, in order to ensure the accuracy of the feature discrimination model, when training the discrimination model, the discrimination model is required to discriminate between the sample data features and the sample operation features, so that the discriminant model after subsequent training can distinguish between data features and operation features.

[0243] The process of calling the feature discrimination model to discriminate the second sample data feature and the second sample operation feature is similar to the above step 902 and will not be repeated here.

[0244] In step 906, the server trains a feature discrimination model according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

[0245] Through the discrimination identifiers of the second sample data feature, the second sample operation feature, the second sample data feature and the second sample operation feature, it is determined whether the discrimination feature of the feature discrimination model is accurate, so that the feature discrimination model can be adjusted to enhance the discrimination performance of the feature discrimination model and improve the accuracy of the feature discrimination model.

[0246] In some embodiments, step 906 includes: taking the second sample data feature as a positive sample, taking the second sample operation feature, and training the feature discrimination model according to the second sample data feature, the second sample operation feature, the second sample data feature and the discrimination identifier of the second sample operation feature.

[0247] In some embodiments, step 906 includes: determining a second loss value of the feature discrimination model based on the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature, and training the feature discrimination model based on the second loss value.

[0248] The second loss value is used to represent the inaccuracy of the current feature discrimination model. The feature discrimination model is trained by the second loss value to reduce the loss value of the feature discrimination model, so as to enhance the discrimination performance of the feature discrimination model and thus improve the accuracy of the feature discrimination model.

[0249] In some embodiments, the second sample data feature, the second sample operation feature and the second loss value satisfy the following relationship:

[0250]

[0251] in, Represents the second loss value; x~P data (z) represents the distribution of the plurality of second sample multimedia data x and the second sample data features of the second sample multimedia data; z~P z (z) represents the distribution of the second sample data feature z of multiple second sample multimedia data; D(x) is the discriminant identifier of the second sample data feature; G(z) represents the second sample operation feature; D(G(z)) represents the discriminant identifier of the second sample operation feature; E( ) represents the mean of the loss values ​​of multiple sample multimedia data.

[0252] It should be noted that in the embodiment of the present disclosure, when training the feature discrimination model, only one training is used as an example for explanation, while in another embodiment, the feature discrimination model is trained multiple times using multiple second sample multimedia data.

[0253] In some embodiments, the feature discrimination model is iteratively trained using multiple second sample multimedia data, and in response to a loss value obtained in a current training round being less than a reference loss value, the training of the feature discrimination model is stopped; or, in response to the training round of the feature discrimination model reaching a reference round, the training of the feature discrimination model is stopped.

[0254] Training the feature discrimination model once with a second sample multimedia data represents a training round. After the feature discrimination model is trained multiple times with multiple second sample multimedia data, when the stop training condition of the feature discrimination model is reached, it means that based on the current feature conversion model, the feature discrimination model can accurately distinguish the sample data features from the sample operation features generated by the current feature conversion model. Therefore, the feature conversion model can be trained further with the currently obtained feature discrimination model.

[0255] In step 907, the server calls the feature conversion model to perform feature conversion processing on the third sample data features corresponding to the third sample multimedia data to obtain the converted third sample operation features.

[0256] The third sample multimedia data is different from the first sample multimedia data and the second sample multimedia data.

[0257] This step 907 is similar to the above step 901 and will not be described again here.

[0258] In step 908, the server calls the feature discrimination model, performs discrimination processing on the third sample operation feature, and determines a discrimination identifier of the third sample operation feature.

[0259] The identification mark is used to indicate whether the third sample operation feature is a data feature or an operation feature. This step 908 is similar to the above step 902 and will not be described in detail here.

[0260] In step 909, the server trains the feature conversion model according to the third sample operation feature and the discriminant identifier.

[0261] This step 909 is similar to the above step 903 and will not be described again here.

[0262] It should be noted that the embodiment of the present disclosure is only illustrated by taking the example of first training the feature conversion model, then training the feature discrimination model, and finally training the feature conversion model. In another embodiment, after step 909, the feature discrimination model is trained again based on the trained feature conversion model, and then the feature conversion model is trained again based on the trained feature discrimination model. The above process is repeated to perform alternating iterative training on the feature discrimination model and the feature conversion model. In response to the feature conversion model and the feature discrimination model reaching a equilibrium state, the alternating iterative training of the feature conversion model and the feature discrimination model is stopped.

[0263] In some embodiments, after the feature conversion model and the feature discrimination model have undergone multiple rounds of alternating iterative training, when the feature conversion model is trained again, the loss value of the feature conversion model converges, and when the feature discrimination model is trained again, the loss value of the feature discrimination model converges, which indicates that the current feature conversion model and feature discrimination model have reached a equilibrium state, and therefore, the alternating iterative training of the feature conversion model and the feature discrimination model is stopped.

[0264] In the process of training the feature conversion model, the feature conversion model acts as a Generator and the feature discrimination model acts as a Discriminator. The feature conversion model and the feature discrimination model form a GAN (Generative Adversarial Network). By training the generative adversarial network, an accurate feature conversion model is obtained, so that the obtained feature conversion model can accurately predict the operation of multimedia data. Subsequently, the predicted operation of multimedia data can be used to ensure accurate recommendation of multimedia data, so as to realize the cold start of multimedia data.

[0265] The method provided by the embodiment of the present disclosure adopts the adversarial training method of the feature conversion model and the feature discrimination model to alternately iteratively train the feature conversion model and the feature discrimination model so that the feature conversion model and the feature discrimination model reach a balanced state, thereby obtaining an accurate feature conversion model.

[0266] Fig.11 is a block diagram of a data recommendation device according to an exemplary embodiment. Fig.11 , the device comprises:

[0267] The feature acquisition unit 1101 is configured to acquire target data features corresponding to the target multimedia data to be recommended;

[0268] The transformation processing unit 1102 is configured to call the feature conversion model to perform feature transformation processing on the target data feature to obtain a transformed predicted operation feature, where the predicted operation feature is used to represent the predicted operation to be performed on the target multimedia data;

[0269] The determining unit 1103 is configured to determine a matching degree between each user and the target multimedia data according to user characteristics corresponding to the plurality of users and the predicted operation characteristics;

[0270] The recommendation unit 1104 is configured to select at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommend the target multimedia data to the at least one target user.

[0271] In some embodiments, reference Fig.12 , the device further comprises:

[0272] The transformation processing unit 1102 is configured to call the feature conversion model to perform feature transformation processing on the first sample data feature corresponding to the first sample multimedia data to obtain a transformed first sample operation feature;

[0273] The discrimination processing unit 1105 is configured to call the feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, where the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature;

[0274] The first training unit 1106 is configured to train the feature conversion model according to the first sample operation feature and the discriminant identifier.

[0275] In some embodiments, reference Fig.12 , the first training unit 1106 includes:

[0276] The first determining subunit 1161 is configured to determine a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier;

[0277] The first training subunit 1162 is configured to train the feature conversion model according to the first loss value.

[0278] In some embodiments, reference Fig.12 , the device further comprises:

[0279] The transformation processing unit 1102 is configured to call the feature conversion model to perform feature transformation processing on the second sample data feature corresponding to the second sample multimedia data to obtain a transformed second sample operation feature;

[0280] The discrimination processing unit 1105 is further configured to call the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determine the discrimination identifiers of the second sample data feature and the second sample operation feature, where the discrimination identifier is used to indicate whether the corresponding feature belongs to a data feature or an operation feature;

[0281] The second training unit 1107 is configured to train the feature discrimination model according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

[0282] In some embodiments, reference Fig.12 , the second training unit 1107 includes:

[0283] The second determining subunit 1171 is configured to determine a second loss value of the feature discriminant model according to the second sample data feature, the second sample operation feature, and the discriminant identifier of the second sample data feature and the second sample operation feature;

[0284] The second training subunit 1172 is configured to train the feature discrimination model according to the second loss value.

[0285] In some embodiments, reference Fig.12 , the device further comprises:

[0286] The transformation processing unit 1102 is configured to call the feature conversion model to perform feature transformation processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature;

[0287] The discrimination processing unit 1105 is further configured to call the feature discrimination model, perform discrimination processing on the third sample operation feature, and determine a discrimination identifier of the third sample operation feature, where the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature;

[0288] The third training unit 1108 is configured to train the feature conversion model according to the third sample operation feature and the discriminant identifier.

[0289] In some embodiments, reference Fig.12 , the feature acquisition unit 1101 includes:

[0290] The feature extraction subunit 1111 is configured to call the feature extraction model, perform feature extraction on the target multimedia data, and obtain target data features corresponding to the target multimedia data.

[0291] In some embodiments, the feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model. The feature extraction sub-unit 1111 is configured to call the text feature extraction sub-model to perform feature extraction on the target multimedia data to obtain text features of the target multimedia data; call the image feature extraction sub-model to perform feature extraction on the target multimedia data to obtain image features of the target multimedia data; and concatenate the text features with the image features to obtain target data features of the target multimedia data.

[0292] In some embodiments, reference Fig.12 , the determining unit 1103 includes:

[0293] The third determining subunit 1131 is configured to determine the matching degree between each user and the target multimedia data according to the user characteristics, target data characteristics and predicted operation characteristics corresponding to the multiple users.

[0294] In some embodiments, the third determining subunit 1131 is configured to match the user characteristics, target data characteristics and predicted operation characteristics of any user to obtain a matching degree between the user and the target multimedia data.

[0295] In some embodiments, the recommendation unit 1104 is configured to determine recommendation parameters corresponding to multiple users based on the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data; select at least one target user from the multiple users based on the recommendation parameters corresponding to the multiple users, and recommend the target multimedia data to the at least one target user.

[0296] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0297] Fig.13 1 is a block diagram of a terminal 1300 according to an exemplary embodiment. The terminal 1300 may be a portable mobile terminal, such as a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer or a desktop computer. The terminal 1300 may also be referred to as a user device, a portable terminal, a laptop terminal, a desktop terminal or other names.

[0298] The terminal 1300 includes a processor 1301 and a memory 1302 .

[0299] The processor 1301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1301 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1301 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1301 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1301 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0300] The memory 1302 may include one or more computer-readable storage media, which may be non-transitory. The memory 1302 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1302 is used to store at least one program code, which is used to be executed by the processor 1301 to implement the data recommendation method provided by the method embodiment of the present disclosure.

[0301] In some embodiments, the terminal 1300 may also optionally include: a peripheral device interface 1303 and at least one peripheral device. The processor 1301, the memory 1302 and the peripheral device interface 1303 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 1303 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1304, a display screen 1305, a camera assembly 1306, an audio circuit 1307, a positioning assembly 1308 and a power supply 1309.

[0302] The peripheral device interface 1303 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1301 and the memory 1302. In some embodiments, the processor 1301, the memory 1302, and the peripheral device interface 1303 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1301, the memory 1302, and the peripheral device interface 1303 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0303] The radio frequency circuit 1304 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1304 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1304 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. In some embodiments, the radio frequency circuit 1304 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency circuit 1304 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1304 may also include circuits related to NFC (Near Field Communication), which is not limited in the present disclosure.

[0304] The display screen 1305 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1305 is a touch display screen, the display screen 1305 also has the ability to collect touch signals on the surface or above the surface of the display screen 1305. The touch signal can be input to the processor 1301 as a control signal for processing. At this time, the display screen 1305 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1305 can be one, set on the front panel of the terminal 1300; in other embodiments, the display screen 1305 can be at least two, respectively set on different surfaces of the terminal 1300 or in a folding design; in other embodiments, the display screen 1305 can be a flexible display screen, set on a curved surface or a folding surface of the terminal 1300. Even, the display screen 1305 can also be set to a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 1305 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0305] The camera assembly 1306 is used to capture images or videos. In some embodiments, the camera assembly 1306 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1306 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. The dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0306] The audio circuit 1307 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 1301 for processing, or input them into the radio frequency circuit 1304 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1300. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 1301 or the radio frequency circuit 1304 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 1307 may also include a headphone jack.

[0307] The positioning component 1308 is used to locate the current geographical location of the terminal 1300 to implement navigation or LBS (Location Based Service). The positioning component 1308 can be a positioning component based on the US GPS (Global Positioning System), China's Beidou system or Russia's Galileo system.

[0308] The power supply 1309 is used to power various components in the terminal 1300. The power supply 1309 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 1309 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0309] In some embodiments, the terminal 1300 further includes one or more sensors 1310 , including but not limited to: an acceleration sensor 1311 , a gyroscope sensor 1312 , a pressure sensor 1313 , a fingerprint sensor 1314 , an optical sensor 1315 , and a proximity sensor 1316 .

[0310] The acceleration sensor 1311 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal 1300. For example, the acceleration sensor 1311 can be used to detect the components of gravity acceleration on the three coordinate axes. The processor 1301 can control the display screen 1305 to display the user interface in a horizontal view or a vertical view according to the gravity acceleration signal collected by the acceleration sensor 1311. The acceleration sensor 1311 can also be used to collect motion data of games or users.

[0311] The gyro sensor 1312 can detect the body direction and rotation angle of the terminal 1300, and the gyro sensor 1312 can cooperate with the acceleration sensor 1311 to collect the user's 3D actions on the terminal 1300. The processor 1301 can implement the following functions based on the data collected by the gyro sensor 1312: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0312] The pressure sensor 1313 can be set on the side frame of the terminal 1300 and / or the lower layer of the display screen 1305. When the pressure sensor 1313 is set on the side frame of the terminal 1300, it can detect the user's holding signal of the terminal 1300, and the processor 1301 performs left and right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor 1313. When the pressure sensor 1313 is set on the lower layer of the display screen 1305, the processor 1301 controls the operability controls on the UI interface according to the user's pressure operation on the display screen 1305. The operability controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0313] The fingerprint sensor 1314 is used to collect the user's fingerprint, and the processor 1301 identifies the user's identity based on the fingerprint collected by the fingerprint sensor 1314, or the fingerprint sensor 1314 identifies the user's identity based on the collected fingerprint. When the user's identity is identified as a trusted identity, the processor 1301 authorizes the user to perform relevant sensitive operations, which include unlocking the screen, viewing encrypted information, downloading software, paying, and changing settings. The fingerprint sensor 1314 can be set on the front, back, or side of the terminal 1300. When a physical button or a manufacturer logo is set on the terminal 1300, the fingerprint sensor 1314 can be integrated with the physical button or the manufacturer logo.

[0314] The optical sensor 1315 is used to collect the ambient light intensity. In one embodiment, the processor 1301 can control the display brightness of the display screen 1305 according to the ambient light intensity collected by the optical sensor 1315. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1305 is increased; when the ambient light intensity is low, the display brightness of the display screen 1305 is reduced. In another embodiment, the processor 1301 can also dynamically adjust the shooting parameters of the camera assembly 1306 according to the ambient light intensity collected by the optical sensor 1315.

[0315] The proximity sensor 1316, also called a distance sensor, is disposed on the front panel of the terminal 1300. The proximity sensor 1316 is used to collect the distance between the user and the front of the terminal 1300. In one embodiment, when the proximity sensor 1316 detects that the distance between the user and the front of the terminal 1300 is gradually decreasing, the processor 1301 controls the display screen 1305 to switch from the screen-on state to the screen-off state; when the proximity sensor 1316 detects that the distance between the user and the front of the terminal 1300 is gradually increasing, the processor 1301 controls the display screen 1305 to switch from the screen-off state to the screen-on state.

[0316] Those skilled in the art will understand that Fig.13 The structure shown in the figure does not constitute a limitation on the terminal 1300, and the terminal 1300 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0317] Fig.14 This is a schematic diagram of the structure of a server according to an exemplary embodiment. The server 1400 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 1401 and one or more memories 1402, wherein the memory 1402 stores at least one program code, and the at least one program code is loaded and executed by the processor 1401 to implement the methods provided by the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for implementing device functions, which will not be described in detail here.

[0318] The server 1400 may be used to execute the steps performed by the server in the above data recommendation method.

[0319] In an exemplary embodiment, a non-transitory computer-readable storage medium is also provided, and when the program code in the storage medium is executed by the processor of the server, the server is enabled to perform the steps performed by the server in the above-mentioned data recommendation method. In some embodiments, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0320] In an exemplary embodiment, a server is further provided, wherein the server comprises:

[0321] one or more processors;

[0322] volatile or non-volatile memory for storing one or more processor executable instructions;

[0323] Among them, one or more processors are configured to execute the steps performed by the server in the above data recommendation method.

[0324] In an exemplary embodiment, a non-transitory computer-readable storage medium is also provided. When the instructions in the storage medium are executed by a processor of a server, the server is enabled to execute the steps executed by the server in the above-mentioned data recommendation method.

[0325] In an exemplary embodiment, a computer program product is also provided. When instructions in the computer program product are executed by a processor of a server, the server is enabled to execute the steps executed by the server in the above-mentioned data recommendation method.

[0326] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0327] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A data recommendation method, characterized in that: The method comprises: Obtaining target data features corresponding to target multimedia data to be recommended, the target data features being obtained by combining text features and image features, the text features being used to represent texts contained in the target multimedia data, and the image features being used to represent images contained in the target multimedia data; the delivery duration of the target multimedia data being less than a reference duration, or the number of operations corresponding to the target multimedia data being less than a reference number; Calling a feature conversion model to perform feature conversion processing on the target data feature to obtain a transformed predicted operation feature, where the predicted operation feature is used to represent a predicted operation to be performed on the target multimedia data; For the user features of any user, the sum of the feature values ​​of the same dimension of the user feature vector, data feature vector and operation feature vector of the user is taken as the fused feature value of the corresponding dimension; or the feature values ​​of the same dimension are weighted averaged to obtain the fused feature value of the corresponding dimension, wherein the user feature vector represents the user feature, the data feature vector represents the target multimedia feature, and the operation feature vector represents the predicted operation feature. The user feature is obtained by extracting features from the user information and is used to describe the user; Performing weighted average processing on the fused feature values ​​of multiple feature dimensions to obtain the matching degree between the user and the target multimedia data; or taking the sum of the fused feature values ​​of multiple feature dimensions as the matching degree between the user and the target multimedia data, the matching degree is used to indicate the user's preference for the multimedia data, and the matching degree is positively correlated with the preference degree; At least one target user is selected from the multiple users according to the matching degrees corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

2. The method according to claim 1, characterized in that The method further comprises: Calling the feature conversion model to perform feature conversion processing on the first sample data feature corresponding to the first sample multimedia data to obtain a converted first sample operation feature; calling a feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, wherein the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature; The feature conversion model is trained according to the first sample operation feature and the discriminant identifier.

3. The method according to claim 2, characterized in that The step of training the feature conversion model according to the first sample operation feature and the discriminant identifier includes: Determining a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier; The feature conversion model is trained according to the first loss value.

4. The method according to claim 2, characterized in that: After training the feature conversion model according to the first sample operation feature and the discrimination identifier, the method further includes: Calling the feature conversion model to perform feature conversion processing on the second sample data feature corresponding to the second sample multimedia data to obtain a converted second sample operation feature; Calling the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determining discrimination identifiers of the second sample data feature and the second sample operation feature, wherein the discrimination identifier is used to indicate whether the corresponding feature belongs to a data feature or an operation feature; The feature discrimination model is trained according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

5. The method according to claim 4, characterized in that The step of training the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature comprises: Determining a second loss value of the feature discrimination model according to the second sample data feature, the second sample operation feature, and a discrimination identifier of the second sample data feature and the second sample operation feature; The feature discrimination model is trained according to the second loss value.

6. The method according to claim 4, characterized in that After training the feature discrimination model according to the second sample data feature, the second sample operation feature, the discrimination identifier of the second sample data feature and the second sample operation feature, the method further includes: Calling the feature conversion model to perform feature conversion processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature; calling the feature discrimination model to perform discrimination processing on the third sample operation feature and determine a discrimination identifier of the third sample operation feature, wherein the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature; The feature conversion model is trained according to the third sample operation feature and the discriminant identifier.

7. The method according to claim 1, characterized in that The step of obtaining target data features corresponding to the target multimedia data to be recommended includes: A feature extraction model is called to perform feature extraction on the target multimedia data to obtain target data features corresponding to the target multimedia data.

8. The method according to claim 7, characterized in that The feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model. The feature extraction model is called to extract features of the target multimedia data to obtain target data features corresponding to the target multimedia data, including: Calling the text feature extraction sub-model to perform feature extraction on the target multimedia data to obtain text features of the target multimedia data; Calling the image feature extraction sub-model to perform feature extraction on the target multimedia data to obtain image features of the target multimedia data; The text feature and the image feature are concatenated to obtain the target data feature of the target multimedia data.

9. The method according to claim 1, characterized in that: The selecting at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommending the target multimedia data to the at least one target user includes: Determining recommendation parameters corresponding to the multiple users according to the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data; At least one target user is selected from the multiple users according to the recommendation parameters corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

10. A data recommendation device, characterized in that: The device comprises: a feature acquisition unit configured to acquire a target data feature corresponding to the target multimedia data to be recommended, wherein the target data feature is obtained by combining a text feature and an image feature, wherein the text feature is used to represent the text contained in the target multimedia data, and the image feature is used to represent the image contained in the target multimedia data; the delivery duration of the target multimedia data is less than a reference duration, or the number of operations corresponding to the target multimedia data is less than a reference number; a transformation processing unit configured to call a feature conversion model to perform feature transformation processing on the target data feature to obtain a transformed predicted operation feature, wherein the predicted operation feature is used to represent a predicted operation to be performed on the target multimedia data; The determination unit is configured to, for any user feature, take the sum of the feature values ​​of the same dimension of the user feature vector, data feature vector and operation feature vector of the user as the fused feature value of the corresponding dimension; or perform weighted average processing on the feature values ​​of the same dimension to obtain the fused feature value of the corresponding dimension, wherein the user feature vector represents the user feature, the data feature vector represents the target multimedia feature, and the operation feature vector represents the predicted operation feature, and the user feature is obtained by extracting features from user information and is used to describe the user; perform weighted average processing on the fused feature values ​​of multiple feature dimensions to obtain the matching degree between the user and the target multimedia data; or take the sum of the fused feature values ​​of multiple feature dimensions as the matching degree between the user and the target multimedia data, wherein the matching degree is used to represent the user's preference for the multimedia data, and the matching degree is positively correlated with the preference degree; Determining a matching degree between each user and the target multimedia data according to user characteristics corresponding to a plurality of users and the predicted operation characteristics; The recommendation unit is configured to select at least one target user from the multiple users according to the matching degrees corresponding to the multiple users, and recommend the target multimedia data to the at least one target user.

11. The device according to claim 10, characterized in that The device also includes: The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the first sample data feature corresponding to the first sample multimedia data to obtain a transformed first sample operation feature; a discrimination processing unit configured to call a feature discrimination model to perform discrimination processing on the first sample operation feature to obtain a discrimination identifier of the first sample operation feature, wherein the discrimination identifier is used to indicate whether the first sample operation feature is a data feature or an operation feature; The first training unit is configured to train the feature conversion model according to the first sample operation feature and the discriminant identifier.

12. The device according to claim 11, characterized in that The first training unit comprises: A first determining subunit is configured to determine a first loss value of the feature conversion model according to the first sample operation feature and the discriminant identifier; The first training subunit is configured to train the feature conversion model according to the first loss value.

13. The device according to claim 11, characterized in that The device also includes: The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the second sample data feature corresponding to the second sample multimedia data to obtain a transformed second sample operation feature; The discrimination processing unit is further configured to call the feature discrimination model to perform discrimination processing on the second sample data feature and the second sample operation feature respectively, and determine discrimination identifiers of the second sample data feature and the second sample operation feature, wherein the discrimination identifier is used to indicate whether the corresponding feature belongs to a data feature or an operation feature; The second training unit is configured to train the feature discrimination model according to the second sample data feature, the second sample operation feature, and the discrimination identifier of the second sample data feature and the second sample operation feature.

14. The device according to claim 13, characterized in that The second training unit comprises: A second determining subunit is configured to determine a second loss value of the feature discriminant model according to the second sample data feature, the second sample operation feature, and a discriminant identifier of the second sample data feature and the second sample operation feature; The second training subunit is configured to train the feature discrimination model according to the second loss value.

15. The device according to claim 13, characterized in that The device also includes: The transformation processing unit is configured to call the feature conversion model to perform feature transformation processing on the third sample data feature corresponding to the third sample multimedia data to obtain a transformed third sample operation feature; The discrimination processing unit is further configured to call the feature discrimination model, perform discrimination processing on the third sample operation feature, and determine a discrimination identifier of the third sample operation feature, wherein the discrimination identifier is used to indicate whether the third sample operation feature is a data feature or an operation feature; The third training unit is configured to train the feature conversion model according to the third sample operation feature and the discriminant identifier.

16. The device according to claim 10, characterized in that The feature acquisition unit comprises: The feature extraction subunit is configured to call a feature extraction model to perform feature extraction on the target multimedia data to obtain target data features corresponding to the target multimedia data.

17. The device according to claim 16, characterized in that The feature extraction model includes a text feature extraction sub-model and an image feature extraction sub-model. The feature extraction sub-unit is configured to call the text feature extraction sub-model to perform feature extraction on the target multimedia data to obtain text features of the target multimedia data; call the image feature extraction sub-model to perform feature extraction on the target multimedia data to obtain image features of the target multimedia data; and concatenate the text features with the image features to obtain target data features of the target multimedia data.

18. The device according to claim 10, characterized in that The recommendation unit is configured to determine the recommendation parameters corresponding to the multiple users according to the matching degrees corresponding to the multiple users and the number of resources corresponding to the target multimedia data; At least one target user is selected from the multiple users according to the recommendation parameters corresponding to the multiple users, and the target multimedia data is recommended to the at least one target user.

19. A server, characterized in that: The server comprises: one or more processors; volatile or non-volatile memory for storing the one or more processor-executable instructions; The one or more processors are configured to execute the data recommendation method according to any one of claims 1 to 9.

20. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of the server, the server is enabled to execute the data recommendation method as described in any one of claims 1 to 9.

21. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of a server, the server is enabled to execute the data recommendation method according to any one of claims 1 to 9.

Images