Matching model training method, device, electronic device and storage medium

Abstract

The present disclosure relates to a matching model training method, device, electronic device and storage medium, and belongs to the field of Internet technology. The method includes: obtaining multiple sample user attribute information and multiple sample data attribute information, performing fusion processing based on multiple user dimension information items in each sample user attribute information and multiple data dimension information items in each sample data attribute information, obtaining the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, and training the matching model based on each sample user attribute information, each sample data attribute information and the corresponding sample matching degree. The sample matching degree can quantify the correlation between the sample user and the sample multimedia data accurately and objectively. The matching model trained based on the sample user attribute information, the sample data attribute information and the sample matching degree has high accuracy and has the ability to accurately obtain the matching degree between the user and the multimedia data based on the user attribute information and the data attribute information.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technology, and in particular to a matching model training method, device, electronic device and storage medium. Background Art

[0002] With the rapid development of Internet technology, a large number of videos have appeared on the Internet, enriching users' daily lives and being welcomed by the majority of users. Any user can publish videos on the Internet, and can also recommend videos to any user for the user to watch. However, if the user is not interested in the recommended video, it is easy to cause disturbance to the user. Therefore, how to recommend matching videos to users is crucial. Summary of the invention

[0003] The present disclosure provides a matching model training method, device, electronic device and storage medium, which can improve the matching degree between multimedia data and users, enhance the recommendation effect, and are not likely to cause disturbance to users.

[0004] According to a first aspect of an embodiment of the present disclosure, a matching model training method is provided, the method comprising:

[0005] Acquire multiple sample user attribute information, each sample user attribute information includes multiple user dimension information items, and each user dimension information item is used to describe the corresponding sample user from the corresponding user dimension;

[0006] Acquire multiple sample data attribute information, each sample data attribute information includes multiple data dimension information items, and each data dimension information item is used to describe corresponding sample multimedia data from a corresponding data dimension;

[0007] Performing fusion processing on the information items of the multiple user dimensions in each sample user attribute information and the information items of the multiple data dimensions in each sample data attribute information to obtain a sample matching degree corresponding to each sample user attribute information and each sample data attribute information;

[0008] The matching model is trained according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0009] In a possible implementation, the fusing the information items of the multiple user dimensions in each sample user attribute information and the information items of the multiple data dimensions in each sample data attribute information to obtain the sample matching degree corresponding to each sample user attribute information and each sample data attribute information includes:

[0010] Acquire relationship data, the relationship data including weights of the multiple user dimensions and weights of the multiple data dimensions, the relationship data being used to indicate a relationship between a matching degree and information items of the multiple user dimensions, information items of the multiple data dimensions, weights of the multiple user dimensions, and weights of the multiple data dimensions;

[0011] The sample matching degree is acquired according to the information items of the multiple user dimensions in the sample user attribute information, the information items of the multiple data dimensions in the sample data attribute information, and the relationship data.

[0012] In another possible implementation, after training the matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, the method further includes:

[0013] Acquire target user attribute information, wherein the target user attribute information includes information items of the multiple user dimensions;

[0014] Acquire data attribute information of each multimedia data, each data attribute information including information items of the multiple data dimensions;

[0015] According to the data attribute information of each multimedia data and the attribute information of the target user, calling the matching model to obtain the matching degree between each multimedia data and the target user;

[0016] Recommend multimedia data to the target user according to the matching degree of the plurality of multimedia data.

[0017] In another possible implementation, the matching model includes at least one feature extraction layer and at least one fusion layer, and the step of calling the matching model to obtain the degree of matching between the multimedia data and the target user according to the data attribute information of the multimedia data and the target user attribute information includes:

[0018] Calling the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features;

[0019] Calling the at least one feature extraction layer to perform feature extraction on data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data;

[0020] The at least one fusion layer is called to perform fusion processing on the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user.

[0021] In another possible implementation, the target user attribute information includes at least a target user identifier, user static information, user statistical information, and user behavior information; and the calling of the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features includes:

[0022] Calling the at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features;

[0023] The target user identification, the user static information, the user statistical information and the user behavior characteristics are combined to obtain the target user characteristics.

[0024] In another possible implementation, the data attribute information of any multimedia data includes at least data identification, data static information, data statistical information and data content information; the calling of the at least one feature extraction layer to perform feature extraction on the data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data includes:

[0025] Calling the at least one feature extraction layer to perform feature extraction on data content information of the any multimedia data to obtain data content features of the any multimedia data;

[0026] The data identifier, the data static information, the data statistical information and the data content feature are combined to obtain a data feature corresponding to the data attribute information of any multimedia data.

[0027] In another possible implementation, the calling of the at least one fusion layer to fuse the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user includes:

[0028] Acquire at least one of context information or probability information, the context information including at least one of current environment information, historical recommendation data information, or historical viewing data information, and the probability information including a probability of performing at least one preset operation on any multimedia data;

[0029] The at least one fusion layer is called to perform fusion processing on at least one of the context information or the probability information, the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user.

[0030] In another possible implementation, the recommending multimedia data to the target user according to the matching degree of the plurality of multimedia data includes:

[0031] Selecting at least one multimedia data from the plurality of multimedia data in descending order of the matching degrees;

[0032] The selected multimedia data is recommended to the target user.

[0033] In another possible implementation, the recommending multimedia data to the target user according to the matching degree of the plurality of multimedia data includes:

[0034] Selecting at least one multimedia data whose matching degree is greater than a preset threshold from the plurality of multimedia data;

[0035] The selected multimedia data is recommended to the target user.

[0036] According to a second aspect of an embodiment of the present disclosure, a matching model training device is provided, the device comprising:

[0037] A sample user information acquisition unit is configured to acquire a plurality of sample user attribute information, each sample user attribute information includes a plurality of user dimension information items, each user dimension information item is used to describe a corresponding sample user from a corresponding user dimension;

[0038] A sample data information acquisition unit is configured to acquire a plurality of sample data attribute information, each sample data attribute information includes information items of a plurality of data dimensions, and each information item of the data dimension is used to describe the corresponding sample multimedia data from the corresponding data dimension;

[0039] a fusion unit configured to perform fusion processing according to the information items of the multiple user dimensions in each sample user attribute information and the information items of the multiple data dimensions in each sample data attribute information, to obtain a sample matching degree corresponding to each sample user attribute information and each sample data attribute information;

[0040] The training unit is configured to train a matching model according to each sample user attribute information, each sample data attribute information, and a sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0041] In a possible implementation, the fusion unit includes:

[0042] a relationship data acquisition subunit, configured to acquire relationship data, the relationship data including weights of the multiple user dimensions and weights of the multiple data dimensions, the relationship data being used to indicate a relationship between a matching degree and information items of the multiple user dimensions, information items of the multiple data dimensions, weights of the multiple user dimensions, and weights of the multiple data dimensions;

[0043] The sample matching degree acquisition subunit is configured to acquire the sample matching degree according to the information items of the multiple user dimensions in the sample user attribute information, the information items of the multiple data dimensions in the sample data attribute information, and the relationship data.

[0044] In another possible implementation, the device further includes:

[0045] A target user information acquisition unit, configured to acquire target user attribute information, wherein the target user attribute information includes information items of the multiple user dimensions;

[0046] A data acquisition unit, configured to acquire data attribute information of each multimedia data, each data attribute information including information items of the multiple data dimensions;

[0047] A calling unit configured to call the matching model to obtain a matching degree between each multimedia data and the target user according to data attribute information of each multimedia data and attribute information of the target user;

[0048] The recommendation unit is configured to recommend multimedia data to the target user according to the matching degree of the plurality of multimedia data.

[0049] In another possible implementation, the matching model includes at least one feature extraction layer and at least one fusion layer, and the calling unit includes:

[0050] The user feature extraction subunit is configured to execute calling the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features;

[0051] The data feature extraction subunit is configured to execute calling the at least one feature extraction layer to perform feature extraction on the data attribute information of any multimedia data, and obtain data features corresponding to the data attribute information of any multimedia data;

[0052] The fusion subunit is configured to execute calling the at least one fusion layer to perform fusion processing on the target user feature and the data feature to obtain the matching degree between any multimedia data and the target user.

[0053] In another possible implementation, the target user attribute information includes at least a target user identifier, user static information, user statistical information, and user behavior information;

[0054] The user feature extraction subunit is further configured to execute calling the at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features;

[0055] The user feature extraction subunit is further configured to combine the target user identifier, the user static information, the user statistical information and the user behavior feature to obtain the target user feature.

[0056] In another possible implementation, the data attribute information of any multimedia data includes at least data identification, data static information, data statistical information and data content information;

[0057] The data feature extraction subunit is further configured to execute calling the at least one feature extraction layer to perform feature extraction on the data content information of the any multimedia data to obtain the data content feature of the any multimedia data;

[0058] The data feature extraction subunit is further configured to combine the data identifier, the data static information, the data statistical information and the data content feature to obtain data features corresponding to the data attribute information of any multimedia data.

[0059] In another possible implementation, the fusion subunit is further configured to execute at least one of acquiring context information or probability information, the context information including at least one of current environment information, historical recommendation data information, or historical viewing data information, and the probability information including a probability of performing at least one preset operation on any multimedia data;

[0060] The fusion subunit is further configured to execute calling the at least one fusion layer to perform fusion processing on at least one of the context information or the probability information, the target user characteristics and the data characteristics to obtain a matching degree between any multimedia data and the target user.

[0061] In another possible implementation, the recommendation unit includes:

[0062] A selection subunit is configured to select at least one multimedia data from the plurality of multimedia data in the order of the matching degree from large to small;

[0063] The recommendation subunit is configured to recommend the selected multimedia data to the target user.

[0064] In another possible implementation, the recommendation unit includes:

[0065] A selection subunit is configured to select at least one multimedia data having a matching degree greater than a preset threshold from the plurality of multimedia data;

[0066] The recommendation subunit is configured to recommend the selected multimedia data to the target user.

[0067] According to a third aspect of an embodiment of the present disclosure, an electronic device for training a matching model is provided, the electronic device comprising:

[0068] one or more processors;

[0069] a volatile or non-volatile memory for storing the one or more processor-executable instructions;

[0070] Wherein, the one or more processors are configured to execute the matching model training method described in the first aspect.

[0071] According to a fourth aspect of an embodiment of the present disclosure, a non-temporary computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the matching model training method described in the first aspect.

[0072] According to a fifth 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 an electronic device, the electronic device is enabled to execute the matching model training method described in the first aspect.

[0073] The matching model training method, device, electronic device and storage medium provided by the embodiments of the present disclosure, the sample matching degree is obtained after comprehensively considering multiple user dimensions and multiple data dimensions, and the correlation between the sample user and the sample multimedia data can be quantified, and the obtained sample matching degree is accurate and objective. Then, the matching model is trained according to the sample user attribute information, the sample data attribute information and the sample matching degree, so that the matching model has the ability to accurately obtain the matching degree between the user and the multimedia data according to the user attribute information and the data attribute information, and the accuracy of the matching model is effectively improved. The matching model can be called later to recommend multimedia data to the user, which can also improve the recommendation accuracy.

[0074] Furthermore, the present disclosure comprehensively considers information items in multiple user dimensions and information items in multiple data dimensions, and introduces a personalized comprehensive measurement mechanism, which is more targeted and can realize personalized recommendation of multimedia data.

[0075] 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

[0076] 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.

[0077] Figure 1 The figure is a flowchart of a matching model training method according to an exemplary embodiment.

[0078] Figure 2 The figure is a flowchart of another matching model training method according to an exemplary embodiment.

[0079] Figure 3 It is a schematic diagram of the structure of a matching model according to an exemplary embodiment.

[0080] Figure 4 The diagram is a schematic structural diagram of an electronic device according to an exemplary embodiment.

[0081] Figure 5 The figure is a flowchart of a multimedia data recommendation method according to an exemplary embodiment.

[0082] Figure 6 It is a structural diagram of a matching model training device according to an exemplary embodiment.

[0083] Figure 7 It is a schematic structural diagram of another matching model training device according to an exemplary embodiment.

[0084] Figure 8 It is a block diagram of a terminal according to an exemplary embodiment.

[0085] Fig. 9 The figure is a schematic diagram showing the structure of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0086] 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.

[0087] 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.

[0088] The user information involved in the present disclosure (including but not limited to user device information, user personal information, etc.) may be information authorized by the user or fully authorized by all parties.

[0089] Figure 1 is a schematic diagram of a matching model training method provided according to an exemplary embodiment, see Figure 1 , including the following steps:

[0090] 101. Acquire multiple sample user attribute information, each sample user attribute information includes multiple user dimension information items, and each user dimension information item is used to describe a corresponding sample user from a corresponding user dimension.

[0091] 102. Acquire multiple sample data attribute information, each sample data attribute information includes multiple data dimension information items, and each data dimension information item is used to describe corresponding sample multimedia data from a corresponding data dimension.

[0092] 103. Perform fusion processing based on the information items of multiple user dimensions in each sample user attribute information and the information items of multiple data dimensions in each sample data attribute information to obtain the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0093] 104. Train a matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0094] In the method provided by the embodiment of the present disclosure, the sample matching degree is obtained after comprehensively considering multiple user dimensions and multiple data dimensions, and the correlation between the sample user and the sample multimedia data can be quantified. The obtained sample matching degree is accurate and objective. The matching model is trained according to the sample user attribute information, the sample data attribute information and the sample matching degree, so that the matching model has the ability to accurately obtain the matching degree between the user and the multimedia data according to the user attribute information and the data attribute information, which effectively improves the accuracy of the matching model. The matching model can be called later to recommend multimedia data to the user, which can also improve the recommendation accuracy.

[0095] Moreover, the present disclosure comprehensively considers information items in multiple user dimensions and information items in multiple data dimensions, introduces a personalized comprehensive measurement mechanism, is more targeted, realizes personalized recommendation of multimedia data, and achieves a balance between multiple experiences of users, such as consumption experience, social experience, production experience, live broadcast experience, and commercial experience. Among them, personalization is not only reflected in the user dimension, but also in the data dimension.

[0096] In a possible implementation, a fusion process is performed based on information items of multiple user dimensions in each sample user attribute information and information items of multiple data dimensions in each sample data attribute information to obtain a sample matching degree corresponding to each sample user attribute information and each sample data attribute information, including:

[0097] Acquire relationship data, the relationship data including weights of multiple user dimensions and weights of multiple data dimensions, the relationship data being used to indicate a relationship between the matching degree and the information items of the multiple user dimensions, the information items of the multiple data dimensions, the weights of the multiple user dimensions, and the weights of the multiple data dimensions;

[0098] The sample matching degree is obtained according to the information items of multiple user dimensions in the sample user attribute information, the information items of multiple data dimensions in the sample data attribute information, and the relationship data.

[0099] In another possible implementation, after training the matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, the method further includes:

[0100] Acquire target user attribute information, where the target user attribute information includes information items of multiple user dimensions;

[0101] Acquire data attribute information of each multimedia data, each data attribute information including information items of multiple data dimensions;

[0102] According to the data attribute information of each multimedia data and the attribute information of the target user, calling the matching model to obtain the matching degree between each multimedia data and the target user;

[0103] Recommend multimedia data to target users based on the matching degree of multiple multimedia data.

[0104] In another possible implementation, the matching model includes at least one feature extraction layer and at least one fusion layer, and according to data attribute information of each multimedia data and attribute information of the target user, the matching model is called to obtain the matching degree between each multimedia data and the target user, including:

[0105] Calling at least one feature extraction layer to extract features from the target user attribute information to obtain target user features;

[0106] Calling at least one feature extraction layer to perform feature extraction on data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data;

[0107] At least one fusion layer is called to fuse the target user features and the data features to obtain the matching degree between any multimedia data and the target user.

[0108] In another possible implementation, the target user attribute information includes at least a target user identifier, user static information, user statistical information, and user behavior information; calling at least one feature extraction layer to extract features from the target user attribute information to obtain target user features includes:

[0109] Calling at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features;

[0110] The target user identification, user static information, user statistical information and user behavior characteristics are combined to obtain the target user characteristics.

[0111] In another possible implementation, the data attribute information of any multimedia data includes at least data identification, data static information, data statistical information and data content information; calling at least one feature extraction layer to extract features from the data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data, including:

[0112] Calling at least one feature extraction layer to perform feature extraction on data content information of any multimedia data to obtain data content features of any multimedia data;

[0113] The data identification, data static information, data statistical information and data content characteristics are combined to obtain data characteristics corresponding to the data attribute information of any multimedia data.

[0114] In another possible implementation, calling at least one fusion layer to fuse the target user feature and the data feature to obtain the matching degree between any multimedia data and the target user includes:

[0115] Acquire at least one of context information or probability information, the context information including at least one of current environment information, historical recommendation data information, or historical viewing data information, and the probability information including a probability of performing at least one preset operation on any multimedia data;

[0116] At least one fusion layer is called to fuse at least one of the context information or the probability information, the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user.

[0117] In another possible implementation, multimedia data is recommended to a target user according to the matching degree of the plurality of multimedia data, including:

[0118] Selecting at least one multimedia data from the plurality of multimedia data in descending order of matching degree;

[0119] Recommend selected multimedia data to target users.

[0120] In another possible implementation, multimedia data is recommended to a target user according to the matching degree of the plurality of multimedia data, including:

[0121] Selecting at least one multimedia data having a matching degree greater than a preset threshold from the plurality of multimedia data;

[0122] Recommend selected multimedia data to target users.

[0123] Figure 2 is a flowchart of a matching model training method according to an exemplary embodiment, which is applied to an electronic device, see Figure 2 , including the following steps:

[0124] 201. Obtain attribute information of multiple sample users.

[0125] The disclosed embodiment provides a matching model training method, which can be trained according to sample information to obtain a matching model. By calling the matching model to recommend multimedia data to users, the matching degree between the multimedia data recommended to the user and the user can be improved, thereby improving the recommendation accuracy.

[0126] The matching model training method provided in the embodiment of the present disclosure can be applied to electronic devices, which include terminals and servers. The terminals can be mobile phones, computers, tablet computers, etc. The server can be a single server, a server cluster composed of several servers, or a cloud computing service center.

[0127] The electronic device can directly display multimedia data, or it can log in based on the user ID and display multimedia data based on the logged-in user ID. When logging in based on the user ID, the multimedia data currently displayed is the multimedia data viewed by the user corresponding to the user ID. The user ID is used to identify a unique user and can be a nickname, account number, mobile phone number or other identifier that can identify a unique user.

[0128] Regarding the format of multimedia data, the multimedia data may be static images, dynamic images, audio or video, etc. Regarding the content of multimedia data, the multimedia data may be film and television works such as movies and TV series, food videos, beauty videos, travel food, funny videos, etc. In addition, the multimedia data may also be a short video composed of one or more combinations of static images, dynamic images or videos.

[0129] In terms of the source of multimedia data, if the electronic device is a terminal, the multimedia data can be captured by the terminal, downloaded from the Internet by the terminal, or sent to the terminal by other devices, which can be terminals or servers.

[0130] The multimedia data in the server is uploaded by one or more terminals, or provided by the publisher of the multimedia data to the maintenance personnel and stored in the server by the maintenance personnel, or sent to the server by other devices. When the terminal uploads the multimedia data to the server, it can be sent in real time, or once every first preset time, or it can be sent actively when the multimedia data is obtained, or it can be sent when receiving the acquisition request sent by the server, which is not specifically limited here.

[0131] In order to improve the matching degree between the multimedia data recommended to the user and the user, the embodiment of the present disclosure obtains a matching model through training, and then obtains a matching degree used to represent the matching degree between the multimedia data and the user according to the trained matching model, and then recommends multimedia data to the user according to the matching degree. The matching model may adopt a deep learning model or other learning models, which is not specifically limited in the embodiment of the present disclosure.

[0132] Considering that users play multiple roles in the context of widespread popularity of multimedia data, at least as data consumers, users can view multimedia data published by other users; as data producers, users can create multimedia data and upload them to the Internet; as socializers, users can interact with other users and perform related operations on multimedia data uploaded by other users or other users; as live broadcast participants, users can broadcast live based on social platforms and upload multimedia data in the form of live broadcasts. Therefore, user attribute information also includes information items of multiple user dimensions, which describe users from corresponding dimensions.

[0133] In the disclosed embodiment, the electronic device first obtains a plurality of sample user attribute information, each of which includes information items of a plurality of user dimensions, and each of which is used to describe a corresponding sample user from a corresponding user dimension.

[0134] In a possible implementation, the multiple user dimensions include at least two of the following:

[0135] At least one dimension for representing user attributes;

[0136] at least one dimension for representing the type of page;

[0137] At least one dimension for representing the amount of data uploaded;

[0138] At least one dimension used to represent user interaction behavior.

[0139] (1) At least one dimension used to represent user attributes, including at least a user identification dimension, a gender dimension, an age dimension, and a region dimension, and may also include a user tag dimension used to represent user preferences and an activity dimension used to represent the user's activity level. The information item of the activity dimension may be high activity or low activity. High activity indicates that the corresponding user has a high activity level, and low activity indicates that the corresponding user has a low activity level. In addition, the at least one dimension used to represent user attributes may also include other dimensions used to represent user attributes in addition to the above-mentioned multiple dimensions, which are not specifically limited here.

[0140] (2) At least one dimension for representing the page type, including at least the page type dimension to which the page currently displayed on the terminal identified by the logged-in target user belongs, and may also include other dimensions for representing the page type. The information items of the type dimension to which the page belongs may be a follow page of a followed user, a page in the same city indicating that the user is in the same city, a personal page for an individual to publish multimedia data, a details page for a detailed description of the user, and a discovery page, etc.

[0141] The electronic device can perform an add-follow operation on other user identifiers based on the logged-in user identifier. For any user identifier, by performing the add-follow operation, the follower user identifier of the currently logged-in user identifier can be determined. For any two user identifiers, if the two user identifiers are each other's follower user identifiers, it means that the association relationship between the two user identifiers is a double-entendre relationship, and the two are each other's double-entendre user identifiers; if only one of the two user identifiers is the follower user identifier of the other, it means that the association relationship between the two user identifiers is a single-entendre relationship, and the follower is the single-entendre user identifier of the follower.

[0142] (3) At least one dimension used to represent the amount of data uploaded, including at least a quantity dimension used to represent the amount of multimedia data uploaded by the user identifier.

[0143] (4) The electronic device can, based on the logged-in user ID, perform operations such as like, comment, share, favorite, and download on multimedia data uploaded by other user IDs. It can also send private messages to other user IDs, perform operations such as adding followers, etc.

[0144] At least one dimension used to represent user interaction behavior is obtained by counting the historical interaction behaviors of the user identifier, including one or more of the like dimension, comment dimension, share dimension, favorite dimension, upload dimension, download dimension, private message dimension, and follow dimension.

[0145] Among them, the information items of the like dimension include one or more of the number of likes, the time of likes, and the multimedia data liked; the information items of the comment dimension include one or more of the number of comments, the time of comments, the length of time in the comment area, the comment content information, and the multimedia data commented on; the information items of the sharing dimension include one or more of the number of sharing, the time of sharing, the sharing method, the sharing object, and the shared multimedia data; the information items of the collection dimension include one or more of the number of collections, the collection time, and the multimedia data collected; the information items of the upload dimension include one or more of the upload time, the number of uploads, and the uploaded multimedia data; the information items of the download dimension include one or more of the number of downloads, the download time, the storage location, and the multimedia data downloaded; the information items of the private message dimension include one or more of the private message content, the private message time, and the private message object; the information items of the follow dimension include one or more of the number of followed users, the follow time, and the followed user type, and the followed user type can be a single-pun user identifier or a double-pun user identifier.

[0146] At least two of the above-mentioned multiple user dimensions can be combined with each other to jointly constitute multiple user dimensions in the embodiment of the present disclosure.

[0147] For example, the attention dimension and the upload dimension are combined to obtain the number of multimedia data uploaded by user identifiers with pun user identifiers of 0-10, the number of multimedia data uploaded by user identifiers with pun user identifiers of 11-40, the number of multimedia data uploaded by user identifiers with pun user identifiers of 41-90, and the number of multimedia data uploaded by user identifiers with pun user identifiers greater than 90.

[0148] In a possible implementation, the electronic device first obtains historical operation records, and obtains sample user attribute information based on the historical operation records. Since the sample user attribute information is obtained based on the historical operation records, the sample user attribute information has sufficient authenticity and better meets the needs of users. The matching model subsequently trained based on the sample user attribute information is also more accurate.

[0149] 202. Obtain attribute information of multiple sample data.

[0150] The electronic device obtains a plurality of sample data attribute information, each of which includes information items of a plurality of data dimensions, and each of which is used to describe corresponding sample multimedia data from a corresponding data dimension.

[0151] In one possible implementation, the multiple data dimensions include at least two of the following:

[0152] At least one dimension for representing data attributes;

[0153] At least one dimension for representing data interaction behavior;

[0154] At least one dimension used to represent the associated user attributes of the data.

[0155] (1) At least one dimension used to represent data attributes, including at least one or more of the following dimensions: data duration dimension, upload author dimension, upload time dimension, category dimension, and data type dimension, and may also include other dimensions used to represent data attributes. The information items of the category dimension may include food, beauty, travel, education, entertainment, etc., and the information items of the data type dimension may include live broadcast type, recorded broadcast type, promotion type, or non-promotion type, etc.

[0156] (2) At least one dimension used to represent data interaction behavior is obtained by counting behaviors such as historical operations performed on multimedia data based on user identification, including one or more of the like dimension, comment dimension, share dimension, favorite dimension, and download dimension, and may also include other dimensions used to describe data interaction behavior.

[0157] Among them, the information items of the like dimension include one or more of the number of likes, the time of like, and the user ID who performs the like operation; the information items of the comment dimension include one or more of the number of comments, the time of comment, the comment content information, and the user ID who posted the comment; the information items of the sharing dimension include one or more of the number of shares, the time of sharing, the sharing method, the user ID who performs the sharing operation, and the sharing object; the information items of the collection dimension include one or more of the number of collections, the time of collection, and the user ID who performs the collection operation; the information items of the download dimension include one or more of the number of downloads, the download time, and the user ID who performs the download operation.

[0158] (3) at least one dimension for representing the associated user attributes of the data, including at least a user ID dimension, a gender dimension, an age dimension, and a region dimension corresponding to the associated user ID, and may also include a user tag dimension for representing the category to which the associated user ID belongs and an activity dimension for representing the user's activity level, and may also include other dimensions for representing user attributes.

[0159] Among them, the associated user attributes of the data are used to describe the user attributes of the user identifier that has an associated relationship with the multimedia data. The user identifier that has an associated relationship with the multimedia data includes: a user identifier that has performed any operation such as like operation, comment operation, share operation, favorite operation, download operation, etc. on the multimedia data.

[0160] In a possible implementation, the electronic device first obtains historical operation records, and obtains sample data attribute information based on the historical operation records. Since the sample data attribute information is obtained based on the historical operation records, the sample data attribute information has sufficient authenticity and better meets the needs of users. The matching model subsequently trained based on the sample data attribute information is also more accurate.

[0161] 203. Perform fusion processing based on the information items of multiple user dimensions in each sample user attribute information and the information items of multiple data dimensions in each sample data attribute information to obtain the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0162] In one possible implementation, the process of obtaining a sample matching degree includes: the electronic device obtains relationship data, and obtains the sample matching degree based on information items of multiple user dimensions in sample user attribute information, information items of multiple data dimensions in sample data attribute information, and relationship data.

[0163] Among them, the relationship data includes weights of multiple user dimensions and weights of multiple data dimensions, and the relationship data is used to indicate the relationship between the matching degree and the information items of multiple user dimensions, the information items of multiple data dimensions, the weights of multiple user dimensions and the weights of multiple data dimensions.

[0164] The relational data may be expressed in the form of a description statement, in the form of a function, or in other forms.

[0165] In addition, the weights of the multiple user dimensions and the weights of the multiple data dimensions may be pre-set or determined by other means, which is not specifically limited in the embodiments of the present disclosure. The weight of each user dimension or data dimension is positively correlated with the degree of influence of the dimension on the user. The larger the weight, the greater the influence of the dimension on the user, and the smaller the weight, the smaller the influence of the dimension on the user.

[0166] Regarding the specific process of obtaining the matching degree, in one possible implementation, the electronic device first performs a weighted summation of information items of multiple user dimensions in the sample user attribute information according to the weight of each user dimension to obtain a first value, and then performs a weighted summation of multiple data dimensions in any sample data attribute information according to the weight of each data dimension to obtain a second value, and determines the sum of the first value and the second value as the matching degree corresponding to the sample user attribute information and the any sample data attribute information.

[0167] The method of obtaining the matching degree between multimedia data and users based on weights is more in line with user needs. By setting the weights of different user dimensions and the weights of different data dimensions, the influence of different dimensions on the matching degree between multimedia data and users can be targeted to increase or decrease. The sample matching degree obtained based on this method is more accurate. Subsequently, training is performed based on sample user attribute information, sample data attribute information and the corresponding sample matching degree, and the resulting matching model is more accurate.

[0168] In a possible implementation, multiple user dimensions and multiple data dimensions may also be combined with each other, and weighted summation is performed according to the weights of the corresponding dimensions to jointly constitute relational data.

[0169] For example, the relationship data is: Matching degree = low-activity discovery page video duration + 2.91 × low-activity attention page single-pun video duration + 7.10 × low-activity attention page double-pun video duration + 0.17 × low-activity personal page video duration + 4.84 × low-activity same-city page video duration + 2.13 × low-activity other page video duration + 0.99 × 1.5 × low-activity attention page live broadcast duration + 0.58 × 1.5 × low-activity personal page live broadcast duration + 0.42 × 1.5 × low-activity same-city page live broadcast duration +2.16×1.5×Low-activity details page live broadcast duration+0.93×1.5×Low-activity other pages live broadcast duration+33.43×Low-activity number of people watching the discovery page+35.59×Low-activity number of people watching the attention page+25.75×Low-activity number of people watching the same city page+0.64×High-activity discovery page video duration+2.66×High-activity attention page single-pun video duration+13.85×High-activity attention page double-pun video duration+0.14×High-activity personal page video duration+2.67×High Live city page video duration + 0.61 × Gaohuo other page video duration + 0.43 × 1.5 × Gaohuo attention page live broadcast duration + 0.89 × 1.5 × Gaohuo personal page live broadcast duration + 0.16 × 1.5 × Gaohuo city page live broadcast duration + 3.01 × 1.5 × Gaohuo details page live broadcast duration + 0.12 × 1.5 × Gaohuo other page live broadcast duration + 151 × number of short video uploads + 27 × number of uploaded video authors + 43 × number of pun user IDs is 0-1 0 × 68 × 11-40 pun user IDs + 115 × 41-90 pun user IDs + 255 × 90 pun user IDs + 18.3 × traffic + number of followers + 0.67 × low activity comment area duration + 2.23 × high activity comment area duration + 470 × pymk (People You May Know, people you may know) Number of pun user identifiers + 0.07×number of video likes / number of video likes + 0.02×number of live broadcast likes + 1.35×number of video comments / number of comments + 0.34×number of live broadcast comments + 2.94×number of private messages received + 72×number of shares + 55×number of guest watermark downloads + 40×number of host watermark downloads + 0.4×video duration on discovery page + 0.5×video duration on same city page + 2.5×duration of non-e-commerce live broadcasts + 0.35×duration of e-commerce live broadcasts + 0.6×duration of e-commerce short videos on discovery page -0.08×ARRR -0.00005×CPU (central processing unit) utilization ×real-time query rate per second.

[0170] Among them, for example, "low-activity discovery page video duration" refers to the average duration of videos watched by users with low activity in all active dimensions obtained through statistical processing on the discovery page; "low-activity follow page single-pun video duration" refers to the average duration of videos uploaded by single-pun user IDs in the follow page watched by users with low activity in all active dimensions obtained through statistical processing; "low-activity number of people watching the same city page" refers to the number of users watching the same city page with low activity in all active dimensions obtained through statistical processing; "number of short videos uploaded by authors with pun user IDs between 0 and 10" refers to the number of short videos uploaded by authors with pun user IDs between 1 and 10 obtained through statistical processing; "low-activity comment area duration" refers to the time spent in the comment area by users with low activity in all active dimensions obtained through statistical processing. Comments can be entered in the comment area, and other users can also browse The comment content identified by the user; "Number of main watermark downloads" refers to the number of times the video containing the watermark information added by the uploader has been downloaded, which is obtained through statistical processing; "Number of guest watermark downloads" refers to the number of times the video containing the watermark information automatically generated by the system has been downloaded, which is obtained through statistical processing; "RRR" is the user reporting rate, which is determined according to the number of user reports and the actual number of exposures of the corresponding video; "ARRR" is an indicator of model fitting multi-dimensional information, which is used to measure the user's viewing situation. Each video corresponds to an ARRR, which is determined by the product of the low viewing probability of the corresponding video and the number of exposures. Compared with RRR, ARRR is more stable and has a higher coverage rate; "CPU utilization" refers to the CPU utilization required to provide users with real-time video recommendation services; "Real-time query rate per second" refers to the real-time query rate per second of the server when providing video recommendation services to users.

[0171] The above process is only a process of obtaining a corresponding sample matching degree based on a sample user attribute information and a sample data attribute information. In order to improve the accuracy of the matching model, the embodiment of the present disclosure can obtain any sample user attribute information and the sample matching degree corresponding to any sample data attribute information based on any sample user attribute information among the multiple sample user attribute information obtained, any sample data attribute information among the multiple sample data attribute information and relationship data. Each sample user attribute information, each sample data attribute information and the corresponding sample matching degree are a group of sample information, and multiple groups of sample information can be obtained. Subsequently, the matching model can be trained based on the multiple groups of sample information.

[0172] For any group of sample information, the more information items the group of sample information contains, the higher the accuracy of the matching model trained based on the group of sample information. Therefore, in another possible implementation, in order to further improve the accuracy of the matching model, the electronic device in the embodiment of the present disclosure also obtains at least one item of sample context information or sample probability information. In addition to the sample user attribute information, the sample data attribute information and the corresponding sample matching degree, each group of sample information also includes at least one item of sample context information or sample probability information.

[0173] Among them, the context information includes at least one of current environment information, historical recommendation data information or historical viewing data information. The current environment information may include whether it is normal time or holiday, the current user's location, the price of the terminal currently used to view multimedia data, or other information used to describe the current usage environment; the probability information includes the probability of performing at least one preset operation on any multimedia data. The preset operation can be a click operation, a download operation, a forwarding operation, a like operation, and can also include other operations performed on the multimedia data. The embodiments of the present application do not specifically limit this.

[0174] In one possible implementation, the electronic device first obtains historical operation records, and obtains sample user attribute information and sample data attribute information based on the historical operation records. Since the sample matching degree is obtained based on the corresponding sample user attribute information and sample data attribute information, and the sample user attribute information and the sample data attribute information are obtained based on the historical operation records, the sample matching degree is more accurate and more consistent with the actual situation. The matching model subsequently trained based on the sample matching degree is also more accurate.

[0175] 204. Train a matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0176] After obtaining multiple sample user attribute information, multiple sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, the electronic device uses the sample user attribute information and the sample data attribute information as input, and uses the sample user attribute information and the sample matching degree corresponding to the sample data attribute information as output to train the matching model. Each sample user attribute information, each sample data attribute information, and the corresponding sample matching degree are a set of sample information. The more sample information there is, the higher the accuracy of the matching model obtained after training. Therefore, in order to obtain a sufficiently accurate matching model, a sufficient number of sample data can be obtained in steps 201-203.

[0177] In one possible implementation, the electronic device also obtains at least one of sample context information or sample probability information as sample information, and trains a matching model based on each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, including: using at least one of the obtained sample context information or sample probability information, each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information as input to train the matching model.

[0178] According to steps 201-204, a trained matching model can be obtained, and the matching degree can be subsequently obtained based on the matching model.

[0179] 205. Obtain attribute information of the target user.

[0180] After the matching model is trained, the matching model can also be called to recommend multimedia data. The disclosed embodiment takes the target user as an example to illustrate the process of recommending multimedia data for the target user. Therefore, the electronic device obtains the target user attribute information, and the target user attribute information includes multiple user dimension information items, and each user dimension information item is used to describe the target user from the corresponding user dimension.

[0181] The process of obtaining the target user attribute information is similar to the process of obtaining the sample user attribute information in the sample information in the above step 201, and will not be repeated here.

[0182] 206. Obtain data attribute information of each multimedia data.

[0183] In order to recommend multimedia data to a target user, the electronic device obtains a plurality of multimedia data, the plurality of multimedia data are used as candidate multimedia data, for each multimedia data, the degree of matching between the multimedia data and the target user is obtained, and the multimedia data is recommended to the target user according to the matching degree. The data attribute information of each multimedia data includes information items of multiple data dimensions, and each information item of the data dimension is used to describe the corresponding multimedia data from the corresponding data dimension.

[0184] The process of obtaining the data attribute information of each multimedia data is similar to the process of obtaining the sample data attribute information in the above step 202, and will not be repeated here.

[0185] In actual situations, due to the large number of users, each user can upload multimedia data. Therefore, for the target user, the number of multimedia data available for selection is very large. Directly using the huge number of multimedia data as candidate multimedia data will not only cause a large amount of calculation, but also waste system resources, increase consumption time, and reduce recommendation efficiency. Therefore, the process of the electronic device obtaining candidate multimedia data includes at least the following situations:

[0186] (1) A case where multimedia data is filtered once:

[0187] In one possible implementation, the electronic device first performs a screening process on a large amount of multimedia data according to the target user attribute information to obtain multiple multimedia data associated with the target user identifier, and then obtains the data attribute information of each multimedia data in the multiple multimedia data obtained after the screening process.

[0188] For example, after the electronic device obtains 100,000 multimedia data, it filters out, based on the target user attribute information, at least one multimedia data of the same type as the multimedia data browsed by the target user, at least one multimedia data uploaded by the same region as the target user, and at least one multimedia data uploaded by a user followed by the target user from the 100,000 multimedia data, thereby obtaining 10,000 multimedia data and obtaining data attribute information of each multimedia data in the 10,000 multimedia data.

[0189] (2) The case where multimedia data is filtered twice:

[0190] In another possible implementation, the electronic device may also perform two screening processes on the massive multimedia data: the electronic device first performs a first screening process on the massive multimedia data according to the target user attribute information to obtain a first amount of multimedia data, and the first amount of multimedia data is the multimedia data in the massive multimedia data that has an associated relationship with the target user identifier. Then, the electronic device performs a second screening process on the first amount of multimedia data according to the target user attribute information to obtain a second amount of multimedia data. The degree of association between the second amount of multimedia data and the target user identifier is greater than the degree of association between other multimedia data and the target user identifier, and the second amount is less than the first amount.

[0191] For example, after the electronic device obtains 100,000 multimedia data, it performs a first screening process on the 100,000 multimedia data to obtain 10,000 multimedia data, and the 10,000 multimedia data are all associated with the target user identifier. Then, the electronic device performs a second screening process on the 10,000 multimedia data obtained after the first screening process to obtain 1,000 multimedia data, and the degree of association between the 1,000 multimedia data and the target user identifier is greater than the degree of association between the other 9,000 multimedia data and the target user identifier.

[0192] In another possible implementation, the massive multimedia data can be screened more times, which is not specifically limited in the embodiments of the present disclosure, and the degree of association between the multimedia data obtained after each screening process and the target user identifier is greater than the degree of association between the multimedia data obtained after the previous screening process and the target user identifier. The method of screening the multimedia data one by one can greatly reduce the amount of multimedia data to obtain matching degrees, effectively reduce the amount of calculation, shorten the time consumption, and improve the recommendation efficiency.

[0193] 207. According to the data attribute information of each multimedia data and the attribute information of the target user, a matching model is called to obtain the matching degree between each multimedia data and the target user.

[0194] After obtaining the target user attribute information and the data attribute information of each multimedia data, the trained matching model is called respectively, and the target user attribute information and the data attribute information of each multimedia data are used as input to obtain the matching degree between each multimedia data and the target user. The matching degree is used to indicate the matching degree between the multimedia data and the target user. The larger the matching degree, the higher the matching degree between the multimedia data and the target user, that is, the more interested the target user is in the multimedia data. The smaller the matching degree, the lower the matching degree between the multimedia data and the target user, that is, the less interested the target user is in the multimedia data.

[0195] In one possible implementation, see Figure 3The matching model includes at least one feature extraction layer and at least one fusion layer. The matching model is called according to the data attribute information of each multimedia data and the target user attribute information to obtain the matching degree between each multimedia data and the target user. The process includes: the electronic device calls the at least one feature extraction layer to perform feature extraction on the target user attribute information to obtain the target user feature, calls the at least one feature extraction layer to perform feature extraction on the data attribute information of any multimedia data to obtain the data feature corresponding to the data attribute information of any multimedia data, and calls the at least one fusion layer to perform fusion processing on the target user feature and the data feature to obtain the matching degree between the any multimedia data and the target user.

[0196] The at least one feature extraction layer may adopt any one of a convolutional neural network, a recurrent neural network, a deep neural network, etc. The at least one fusion layer may be a fully connected layer or other network layers.

[0197] In another possible implementation, at least one feature extraction layer used for extracting features from target user attribute information is different from at least one feature extraction layer used for extracting features from data attribute information.

[0198] For example, at least one feature extraction layer for extracting features from target user attribute information uses a recurrent neural network, at least one feature extraction layer for extracting features from data attribute information uses a convolutional neural network, and the at least one fusion layer is a fully connected layer. The process of calling a matching model to obtain the degree of matching between each multimedia data and the target user according to the data attribute information of each multimedia data and the target user attribute information includes: calling a recurrent neural network layer to extract features from the target user attribute information to obtain target user features, calling a convolutional neural network layer to extract features from the data attribute information of the multimedia data to obtain corresponding data features. Calling a fully connected layer to fuse the target user features and the data features to obtain the corresponding degree of matching.

[0199] (1) The process of obtaining target user characteristics:

[0200] In one possible implementation, see Figure 3 The target user attribute information includes at least a target user identifier, user static information, user statistical information and user behavior information. The process of calling at least one feature extraction layer to extract features from the target user attribute information to obtain target user features includes: the electronic device calls the at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features, and combines the target user identifier, the user static information, the user statistical information and the user behavior features to obtain target user features.

[0201] Among them, the user static information is used to describe the user's inherent attribute information, such as the user's gender, age, region, etc.; the user statistical information is used to describe the user's statistical attribute information, such as the user's click rate or click volume on any category of multimedia data within a preset time, the ratio of the user's average viewing time of the specified category of multimedia data within the preset time to the average viewing time of other users on the specified category of multimedia data, and other information obtained by counting various information of the user; the user behavior information is used to describe the user's historical behaviors such as operations performed, such as multimedia data viewed by the user, multimedia data liked by the user, multimedia data collected by the user, comments posted by the user, etc.

[0202] (2) The process of obtaining data features:

[0203] In another possible implementation, see Figure 3 The data attribute information of any multimedia data includes at least data identification, data static information, data statistical information and data content information. The process of calling at least one feature extraction layer to perform feature extraction on the data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data includes: for any multimedia data, the electronic device calls the at least one feature extraction layer to perform feature extraction on the data content information of any multimedia data to obtain the data content features of the any multimedia data, and combines the data identification, the data static information, the data statistical information and the data content features to obtain the data features corresponding to the data attribute information of the any multimedia data.

[0204] Among them, the data static information is used to describe the inherent attribute information of multimedia data, such as the duration, upload time, category, and uploading author of the multimedia data; the data statistical information is used to describe the statistical attribute information of the multimedia data, such as the ratio of the number of clicks on the multimedia data to the sum of the number of clicks on multiple multimedia data published by the uploading author of the multimedia data, the ratio of the average viewing time of the multimedia data to the sum of the average viewing time of multiple multimedia data of the same category, and other information obtained by counting multiple pieces of information of the corresponding multimedia data; the data content information is used to describe the content contained in the multimedia data.

[0205] In order to further improve the accuracy of the matching model, in another possible implementation, in addition to the target user attribute information and the data attribute information of the multimedia data, the electronic device also obtains at least one of context information or probability information.

[0206] In this manner, the process of calling at least one fusion layer to fuse the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user includes: the electronic device obtains context information, calls at least one fusion layer to fuse the context information, the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user.

[0207] In another possible implementation, at least one fusion layer is called to fuse the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user, including: the electronic device obtains probability information, and calls at least one fusion layer to fuse the probability information, the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user.

[0208] In another possible implementation, the process of calling at least one fusion layer to fuse the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user includes: the electronic device obtains context information and probability information, and calls at least one fusion layer to fuse the context information, the probability information, the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user.

[0209] It should be noted that, in another possible implementation, the electronic device may also obtain other information, and process the other information, target user attribute information, and data attribute information to obtain a corresponding matching degree, which is not specifically limited in the embodiments of the present disclosure.

[0210] 208. Recommend multimedia data to a target user based on the matching degree of the plurality of multimedia data.

[0211] After obtaining the matching degree of each multimedia data among the multiple multimedia data, the electronic device recommends the multimedia data to the target user according to the matching degree of the multiple multimedia data.

[0212] Among them, the greater the matching degree, the more matching the multimedia data is with the target user, and the more interested the target user is in the multimedia data. Therefore, in one possible implementation, the process of recommending multimedia data to the target user based on the matching degrees of multiple multimedia data includes: the electronic device sorts the multiple multimedia data in order from large to small according to the matching degrees, and then selects at least one multimedia data from the multiple multimedia data in the order of arrangement, and recommends the selected at least one multimedia data to the target user.

[0213] For example, the electronic device obtains the matching degree of 100 multimedia data, sorts the 100 multimedia data in order from large to small matching degree, and selects the first 30 multimedia data in the order of arrangement. The matching degree of these 30 multimedia data is the highest, indicating that the matching degree between the multimedia data and the target user is the highest, and the 30 multimedia data are recommended to the target user.

[0214] In a possible implementation, the terminal logs in based on the target user identifier corresponding to the target user. When the electronic device obtains the matching degrees of multiple multimedia data, it sends the recommended multimedia data to the terminal with the logged in target user identifier according to the multiple matching degrees.

[0215] In another possible implementation, the process of recommending multimedia data to a target user based on the matching degree of multiple multimedia data includes: the electronic device obtains a preset threshold, selects at least one multimedia data having a matching degree greater than the preset threshold from the multiple multimedia data, and recommends the selected at least one multimedia data to the target user.

[0216] For example, the electronic device obtains the matching degree of 100 multimedia data and a preset threshold Y, and selects 30 multimedia data with a matching degree greater than the preset threshold Y from the 100 multimedia data, indicating that the matching degree between the 30 multimedia data and the target user is high enough, and the 30 multimedia data are recommended to the target user.

[0217] The method of selecting in order of matching degree from large to small, or the method of selecting according to matching degree greater than a preset threshold, is closer to the actual situation, and the matching degree between the selected multimedia data and the target user is higher. The selected multimedia data can be subsequently recommended to the target user, which is more in line with the user's needs and has a better recommendation effect.

[0218] For a target user, the electronic device directly recommends the selected multimedia data to the target user. For multiple users, the electronic device recommends the corresponding multimedia data to each user according to the matching degree between each user and the multimedia data, forming a personalized distribution mechanism, which can realize the personalized distribution of multimedia data.

[0219] In the method provided by the embodiment of the present disclosure, the sample matching degree is obtained after comprehensively considering multiple user dimensions and multiple data dimensions, and the correlation between the sample user and the sample multimedia data can be quantified. The obtained sample matching degree is accurate and objective. The matching model is trained according to the sample user attribute information, the sample data attribute information and the sample matching degree, so that the matching model has the ability to accurately obtain the matching degree between the user and the multimedia data according to the user attribute information and the data attribute information, which effectively improves the accuracy of the matching model. The matching model can be called later to recommend multimedia data to the user, which can also improve the recommendation accuracy.

[0220] In addition, the trained matching model is called to recommend multimedia data in combination with user attribute information and data attribute information. Since the matching model is accurate, the recommendation accuracy can be guaranteed, and the recommendation process is simple without the need for complex processing, which effectively reduces the processing volume, reduces the occupied processing resources, and improves the recommendation efficiency.

[0221] Moreover, the present disclosure comprehensively considers information items in multiple user dimensions and information items in multiple data dimensions, introduces a personalized comprehensive measurement mechanism, is more targeted, can realize personalized recommendation of multimedia data, and achieves a balance between multiple experiences such as user consumption experience, social experience, production experience, live broadcast experience, and commercial experience. Among them, personalization is not only reflected in the user dimension, but also in the data dimension.

[0222] It should be noted that the electronic device in the embodiment of the present disclosure may include a terminal, or may include a terminal and a server. When the electronic device includes a terminal, the terminal may obtain multiple sets of sample information, train a matching model based on the multiple sets of sample information, and obtain a trained matching model. When the terminal obtains the target user attribute information and the data attribute information of any multimedia data, the target user attribute information and the data attribute information of any multimedia data are processed by calling the matching model to obtain the matching degree between the any multimedia data and the target user. The terminal may also send the trained matching model to other devices for use by other devices. When the electronic device includes a terminal and a server, the server obtains multiple sets of sample information, trains a matching model based on the multiple sets of sample information, and obtains a trained matching model. The server may send the matching model to the terminal for use by the terminal, or when the terminal obtains the target user attribute information and the data attribute information of any multimedia data, the target user attribute information and the data attribute information sent by the terminal are received, the matching model is called, the matching degree between the any multimedia data and the target user is obtained, and the obtained matching degree is sent to the terminal.

[0223] According to the method provided in the above embodiment, the process of recommending multimedia data is described. The process is applied in an electronic device, the electronic device includes a terminal and a server, the terminal includes a recall module and a rough sorting module, see Figure 3 , Figure 4 and Figure 5 ,include:

[0224] 1. The server obtains sample context information, sample probability information, sample user attribute information and sample data attribute information based on the historical operation records sent by the terminal.

[0225] Among them, the sample user attribute information includes user identification, user static information, user behavior information and user statistical information; the sample data attribute information includes data identification, data static information, data statistical information and data content information.

[0226] 2. The server trains the matching model according to the acquired sample context information, sample probability information, sample user attribute information and sample data attribute information to obtain a trained matching model.

[0227] Among them, the matching model includes at least one recurrent neural network layer, at least one convolutional neural network layer and at least one fully connected layer.

[0228] 3. The server sends the trained matching model to the terminal, and the matching model constitutes the recommendation module in the terminal.

[0229] 4. The terminal obtains target context information, target probability information, target user attribute information and 100,000 multimedia data, and performs a first screening on the 100,000 multimedia data by calling the recall module to obtain 1,000 multimedia data.

[0230] Among them, the target user attribute information includes the target user identification, user static information, user behavior information and user statistical information; the data attribute information of any multimedia data includes data identification, data static information, data statistical information and data content information.

[0231] 5. The terminal calls the rough sorting module to perform a second screening on the 1000 multimedia data to obtain 100 multimedia data.

[0232] 6. The terminal calls at least one recurrent neural network layer in the matching model to extract features of the user behavior information in the target user attribute information to obtain user behavior features, and combines the target user identifier, user static information, user behavior feature information and user statistical information to obtain target user features.

[0233] 7. For any multimedia data, the terminal calls at least one convolutional neural network layer in the matching model to perform feature extraction on the data content information in the data attribute information of the multimedia data to obtain data content features, and combines the data identifier, data static information, data statistical information and data content features to obtain the data features of any multimedia data.

[0234] 8. At least one fully connected layer in the matching model is called to fuse the target context information, the target probability information, the target user characteristics and the data characteristics of any multimedia data to obtain the corresponding matching degree.

[0235] 9. Repeat steps 7 and 8 to obtain the matching degree corresponding to each of the 100 multimedia data.

[0236] 10. The terminal calls the recommendation module to sort the 100 multimedia data in descending order of matching degree.

[0237] 11. The terminal selects first 10 multimedia data from the 100 multimedia data according to the arrangement order of the 100 multimedia data.

[0238] 12. The terminal recommends the 10 multimedia data to the target user.

[0239] Figure 6 is a block diagram of a matching model training device according to an exemplary embodiment. Figure 6 , the device comprises:

[0240] The sample user information acquisition unit 601 is configured to acquire multiple sample user attribute information, each sample user attribute information includes multiple user dimension information items, and each user dimension information item is used to describe the corresponding sample user from the corresponding user dimension;

[0241] The sample data information acquisition unit 602 is configured to acquire a plurality of sample data attribute information, each of which has information items of a plurality of data dimensions, and each of which is used to describe corresponding sample multimedia data from a corresponding data dimension;

[0242] A fusion unit 603 is configured to perform fusion processing according to information items of multiple user dimensions in each sample user attribute information and information items of multiple data dimensions in each sample data attribute information, so as to obtain a sample matching degree corresponding to each sample user attribute information and each sample data attribute information;

[0243] The training unit 604 is configured to train the matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

[0244] In one possible implementation, see Figure 7 , the fusion unit 603 includes:

[0245] The relationship data acquisition subunit 6031 is configured to acquire relationship data, wherein the relationship data includes weights of multiple user dimensions and weights of multiple data dimensions, and the relationship data is used to indicate the relationship between the matching degree and the information items of multiple user dimensions, the information items of multiple data dimensions, the weights of multiple user dimensions, and the weights of multiple data dimensions;

[0246] The sample matching degree acquisition subunit 6032 is configured to acquire the sample matching degree according to the information items of multiple user dimensions in the sample user attribute information, the information items of multiple data dimensions in the sample data attribute information, and the relationship data.

[0247] In another possible implementation, see Figure 7 , the device further comprises:

[0248] A target user information acquisition unit 605 is configured to acquire target user attribute information, where the target user attribute information includes information items of multiple user dimensions;

[0249] The data acquisition unit 606 is configured to acquire data attribute information of each multimedia data, each data attribute information including information items of multiple data dimensions;

[0250] The calling unit 607 is configured to call the matching model to obtain the matching degree between each multimedia data and the target user according to the data attribute information of each multimedia data and the target user attribute information;

[0251] The recommendation unit 608 is configured to recommend multimedia data to a target user according to the matching degree of the plurality of multimedia data.

[0252] In another possible implementation, see Figure 7 The matching model includes at least one feature extraction layer and at least one fusion layer. The calling unit 607 includes:

[0253] The user feature extraction subunit 6071 is configured to execute calling at least one feature extraction layer to extract the target user attribute information to obtain the target user feature;

[0254] The data feature extraction subunit 6072 is configured to execute calling at least one feature extraction layer to extract features from data attribute information of any multimedia data, and obtain data features corresponding to the data attribute information of any multimedia data;

[0255] The fusion subunit 6073 is configured to execute calling at least one fusion layer to perform fusion processing on the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user.

[0256] In another possible implementation, see Figure 7 , the target user attribute information at least includes the target user identification, user static information, user statistical information and user behavior information;

[0257] The user feature extraction subunit 6071 is further configured to execute calling at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features;

[0258] The user feature extraction subunit 6071 is further configured to combine the target user identification, user static information, user statistical information and user behavior characteristics to obtain target user features.

[0259] In another possible implementation, see Figure 7 , the data attribute information of any multimedia data at least includes data identification, data static information, data statistical information and data content information;

[0260] The data feature extraction subunit 6072 is further configured to execute calling at least one feature extraction layer to extract features from data content information of any multimedia data to obtain data content features of any multimedia data;

[0261] The data feature extraction subunit 6072 is further configured to combine the data identification, data static information, data statistical information and data content features to obtain data features corresponding to the data attribute information of any multimedia data.

[0262] In another possible implementation, see Figure 7 The fusion subunit 6073 is further configured to obtain at least one of context information or probability information, the context information includes at least one of current environment information, historical recommendation data information or historical viewing data information, and the probability information includes a probability of performing at least one preset operation on any multimedia data;

[0263] The fusion subunit 6073 is further configured to execute calling at least one fusion layer to perform fusion processing on at least one of the context information or the probability information, the target user characteristics and the data characteristics to obtain the matching degree between any multimedia data and the target user.

[0264] In another possible implementation, see Figure 7 , the recommendation unit 608 includes:

[0265] The selection subunit 6081 is configured to select at least one multimedia data from the plurality of multimedia data in descending order of matching degree;

[0266] The recommendation subunit 6082 is configured to perform recommendation of selected multimedia data for target users.

[0267] In another possible implementation, see Figure 7 , recommended units include:

[0268] The selection subunit 6081 is configured to select at least one multimedia data having a matching degree greater than a preset threshold from the plurality of multimedia data;

[0269] The recommendation subunit 6082 is configured to perform recommendation of selected multimedia data for target users.

[0270] 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.

[0271] Figure 8 800 is a block diagram of a terminal 800 according to an exemplary embodiment. The terminal 800 is used to execute the steps executed by the terminal in the above matching model training method, and can 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 (Moving Picture Experts Group Audio Layer IV), a laptop or a desktop computer. The terminal 800 may also be called a user device, a portable terminal, a laptop terminal, a desktop terminal, and other names.

[0272] Typically, the terminal 800 includes: one or more processors 801 and one or more memories 802 .

[0273] The processor 801 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 801 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 801 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 801 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 801 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0274] The memory 802 may include one or more computer-readable storage media, which may be non-transitory. The memory 802 may also include volatile memory or 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 802 is used to store at least one instruction, which is used to be possessed by the processor 801 to implement the matching model training method provided in the method embodiment of the present application.

[0275] In some embodiments, the terminal 800 may further optionally include: a peripheral device interface 803 and at least one peripheral device. The processor 801, the memory 802 and the peripheral device interface 803 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 803 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 804, a touch display screen 805, a camera 806, an audio circuit 807, a positioning component 808 and a power supply 809.

[0276] The peripheral device interface 803 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 801 and the memory 802. In some embodiments, the processor 801, the memory 802, and the peripheral device interface 803 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 801, the memory 802, and the peripheral device interface 803 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0277] The radio frequency circuit 804 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 804 communicates with communication networks and other communication devices through electromagnetic signals. The radio frequency circuit 804 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency circuit 804 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, etc. The radio frequency circuit 804 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: a metropolitan area network, various generations of mobile communication networks (2G, 3G, 4G and 13G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 804 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0278] The display screen 805 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos and any combination thereof. When the display screen 805 is a touch display screen, the display screen 805 also has the ability to collect touch signals on the surface or above the surface of the display screen 805. The touch signal can be input to the processor 801 as a control signal for processing. At this time, the display screen 805 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 805 can be one, and the front panel of the terminal 800 is set; in other embodiments, the display screen 805 can be at least two, which are respectively set on different surfaces of the terminal 800 or are folded; in some other embodiments, the display screen 805 can be a flexible display screen, which is set on the curved surface or folded surface of the terminal 800. Even, the display screen 805 can also be set to a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 805 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode, organic light-emitting diode).

[0279] The camera assembly 806 is used to capture images or videos. Optionally, the camera assembly 806 includes a front camera and a rear camera. Typically, 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 panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 806 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. A 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.

[0280] The audio circuit 807 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 801 for processing, or input them into the radio frequency circuit 804 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 800. 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 801 or the radio frequency circuit 804 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 807 may also include a headphone jack.

[0281] The positioning component 808 is used to locate the current geographical location of the terminal 800 to implement navigation or LBS (Location Based Service). The positioning component 808 can be a positioning component based on the US GPS (Global Positioning System), China's Beidou system, Russia's Grenas system or the European Union's Galileo system.

[0282] The power supply 809 is used to power various components in the terminal 800. The power supply 809 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 809 includes a rechargeable battery, the rechargeable battery can support wired charging or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0283] In some embodiments, the terminal 800 further includes one or more sensors 180 , including but not limited to: an acceleration sensor 811 , a gyroscope sensor 812 , a pressure sensor 813 , a fingerprint sensor 814 , an optical sensor 815 , and a proximity sensor 816 .

[0284] The acceleration sensor 811 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal 800. For example, the acceleration sensor 811 can be used to detect the components of gravity acceleration on the three coordinate axes. The processor 801 can control the touch display screen 805 to display the user interface in a horizontal view or a vertical view according to the gravity acceleration signal collected by the acceleration sensor 811. The acceleration sensor 811 can also be used for collecting game or user motion data.

[0285] The gyroscope sensor 812 can detect the body direction and rotation angle of the terminal 800, and the gyroscope sensor 812 can cooperate with the acceleration sensor 811 to collect the user's 3D actions on the terminal 800. The processor 801 can implement the following functions based on the data collected by the gyroscope sensor 812: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0286] The pressure sensor 813 can be set on the side frame of the terminal 800 and / or the lower layer of the touch display screen 805. When the pressure sensor 813 is set on the side frame of the terminal 800, it can detect the user's holding signal of the terminal 800, and the processor 801 performs left and right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor 813. When the pressure sensor 813 is set on the lower layer of the touch display screen 805, the processor 801 controls the operability controls on the UI interface according to the user's pressure operation on the touch display screen 805. The operability controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0287] The fingerprint sensor 814 is used to collect the user's fingerprint, and the processor 801 identifies the user's identity based on the fingerprint collected by the fingerprint sensor 814, or the fingerprint sensor 814 identifies the user's identity based on the collected fingerprint. When the user's identity is identified as a trusted identity, the processor 801 authorizes the user to perform relevant sensitive operations, including unlocking the screen, viewing encrypted information, downloading software, paying, and changing settings. The fingerprint sensor 814 can be set on the front, back, or side of the terminal 800. When a physical button or a manufacturer logo is set on the terminal 800, the fingerprint sensor 814 can be integrated with the physical button or the manufacturer logo.

[0288] The optical sensor 815 is used to collect the ambient light intensity. In one embodiment, the processor 801 can control the display brightness of the touch display screen 805 according to the ambient light intensity collected by the optical sensor 815. Specifically, when the ambient light intensity is high, the display brightness of the touch display screen 805 is increased; when the ambient light intensity is low, the display brightness of the touch display screen 805 is reduced. In another embodiment, the processor 801 can also dynamically adjust the shooting parameters of the camera assembly 806 according to the ambient light intensity collected by the optical sensor 815.

[0289] The proximity sensor 816, also called a distance sensor, is usually arranged on the front panel of the terminal 800. The proximity sensor 816 is used to collect the distance between the user and the front of the terminal 800. In one embodiment, when the proximity sensor 816 detects that the distance between the user and the front of the terminal 800 is gradually decreasing, the processor 801 controls the touch display screen 805 to switch from the screen-on state to the screen-off state; when the proximity sensor 816 detects that the distance between the user and the front of the terminal 800 is gradually increasing, the processor 801 controls the touch display screen 805 to switch from the screen-off state to the screen-on state.

[0290] Those skilled in the art will understand that Figure 8 The structure shown in the figure does not constitute a limitation on the terminal 800, and the terminal 800 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0291] Figure 8 This is a schematic diagram of the structure of a server according to an exemplary embodiment. The server 800 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 801 and one or more memories 802, wherein the memory 802 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 801 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.

[0292] Server 800 can be used to execute the steps performed by the server in the above-mentioned matching model training method.

[0293] In an exemplary embodiment, a non-temporary computer-readable storage medium is also provided. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the steps performed by the server or terminal in the above-mentioned matching model training method.

[0294] In an exemplary embodiment, a computer program product is also provided. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute the steps executed by the server or terminal in the above-mentioned matching model training method.

[0295] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses or adaptations of the present disclosure, which 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 specification and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0296] 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 matching model training method, characterized in that: The method comprises: Acquire multiple sample user attribute information, each sample user attribute information includes multiple user dimension information items, and each user dimension information item is used to describe the corresponding sample user from the corresponding user dimension; Acquire multiple sample data attribute information, each sample data attribute information includes multiple data dimension information items, and each data dimension information item is used to describe corresponding sample multimedia data from a corresponding data dimension; Obtaining weights of the multiple user dimensions in each sample user attribute information and weights of the multiple data dimensions in each sample data attribute information; the weight corresponding to each dimension indicates the influence of the dimension on determining the matching degree between the multimedia data and the user; For a sample user attribute information and a sample data attribute information, weighted summing the information items of multiple user dimensions in the sample user attribute information according to the weight of each user dimension is performed to obtain a first value; weighted summing the multiple data dimensions in the sample data attribute information according to the weight of each data dimension is performed to obtain a second value; and the sum of the first value and the second value is determined as the sample matching degree corresponding to the sample user attribute information and the sample data attribute information; The matching model is trained according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

2. The method according to claim 1, characterized in that The method further comprises: Acquire relationship data, the relationship data including weights of the multiple user dimensions and weights of the multiple data dimensions, the relationship data being used to indicate a relationship between a matching degree and information items of the multiple user dimensions, information items of the multiple data dimensions, weights of the multiple user dimensions, and weights of the multiple data dimensions; The sample matching degree is acquired according to the information items of the multiple user dimensions in the sample user attribute information, the information items of the multiple data dimensions in the sample data attribute information, and the relationship data.

3. The method according to claim 1, characterized in that After training the matching model according to each sample user attribute information, each sample data attribute information, and the sample matching degree corresponding to each sample user attribute information and each sample data attribute information, the method further includes: Acquire target user attribute information, wherein the target user attribute information includes information items of the multiple user dimensions; Acquire data attribute information of each multimedia data, each data attribute information including information items of the multiple data dimensions; According to the data attribute information of each multimedia data and the attribute information of the target user, calling the matching model to obtain the matching degree between each multimedia data and the target user; Recommend multimedia data to the target user according to the matching degree of the plurality of multimedia data.

4. The method according to claim 3, characterized in that The matching model includes at least one feature extraction layer and at least one fusion layer, and the matching model is called to obtain the matching degree between the multimedia data and the target user according to the data attribute information of the multimedia data and the target user attribute information, including: Calling the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features; Calling the at least one feature extraction layer to perform feature extraction on data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data; The at least one fusion layer is called to perform fusion processing on the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user.

5. The method according to claim 4, characterized in that The target user attribute information at least includes a target user identifier, user static information, user statistical information, and user behavior information; the calling of the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features includes: Calling the at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features; The target user identification, the user static information, the user statistical information and the user behavior characteristics are combined to obtain the target user characteristics.

6. The method according to claim 4, characterized in that The data attribute information of any multimedia data at least includes data identification, data static information, data statistical information and data content information; the calling of the at least one feature extraction layer to extract features from the data attribute information of any multimedia data to obtain data features corresponding to the data attribute information of any multimedia data includes: Calling the at least one feature extraction layer to perform feature extraction on data content information of the any multimedia data to obtain data content features of the any multimedia data; The data identifier, the data static information, the data statistical information and the data content feature are combined to obtain a data feature corresponding to the data attribute information of any multimedia data.

7. The method according to claim 4, characterized in that The calling of the at least one fusion layer to fuse the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user includes: Acquire at least one of context information or probability information, the context information including at least one of current environment information, historical recommendation data information, or historical viewing data information, and the probability information including a probability of performing at least one preset operation on any multimedia data; The at least one fusion layer is called to perform fusion processing on at least one of the context information or the probability information, the target user feature and the data feature to obtain a matching degree between any multimedia data and the target user.

8. The method according to claim 3, characterized in that The recommending multimedia data to the target user according to the matching degree of the plurality of multimedia data includes: Selecting at least one multimedia data from the plurality of multimedia data in descending order of the matching degrees; The selected multimedia data is recommended to the target user.

9. The method according to claim 3, characterized in that: The recommending multimedia data to the target user according to the matching degree of the plurality of multimedia data includes: Selecting at least one multimedia data whose matching degree is greater than a preset threshold from the plurality of multimedia data; The selected multimedia data is recommended to the target user.

10. A matching model training device, characterized in that: The device comprises: A sample user information acquisition unit is configured to acquire a plurality of sample user attribute information, each sample user attribute information includes a plurality of user dimension information items, each user dimension information item is used to describe a corresponding sample user from a corresponding user dimension; A sample data information acquisition unit is configured to acquire a plurality of sample data attribute information, each sample data attribute information includes information items of a plurality of data dimensions, and each information item of the data dimension is used to describe the corresponding sample multimedia data from the corresponding data dimension; The fusion unit is configured to execute the steps of obtaining the weights of the multiple user dimensions in each sample user attribute information and the weights of the multiple data dimensions in each sample data attribute information; the weight corresponding to each dimension indicates the influence of the dimension on determining the matching degree between multimedia data and the user; for a sample user attribute information and a sample data attribute information, performing weighted summation on the information items of the multiple user dimensions in the sample user attribute information according to the weight of each user dimension to obtain a first value; performing weighted summation on the multiple data dimensions in the sample data attribute information according to the weight of each data dimension to obtain a second value; and determining the sum of the first value and the second value as the sample matching degree corresponding to the sample user attribute information and the sample data attribute information; The training unit is configured to train a matching model according to each sample user attribute information, each sample data attribute information, and a sample matching degree corresponding to each sample user attribute information and each sample data attribute information.

11. The device according to claim 10, characterized in that The fusion unit comprises: a relationship data acquisition subunit, configured to acquire relationship data, the relationship data including weights of the multiple user dimensions and weights of the multiple data dimensions, the relationship data being used to indicate a relationship between a matching degree and information items of the multiple user dimensions, information items of the multiple data dimensions, weights of the multiple user dimensions, and weights of the multiple data dimensions; The sample matching degree acquisition subunit is configured to acquire the sample matching degree according to the information items of the multiple user dimensions in the sample user attribute information, the information items of the multiple data dimensions in the sample data attribute information, and the relationship data.

12. The device according to claim 10, characterized in that The device also includes: A target user information acquisition unit, configured to acquire target user attribute information, wherein the target user attribute information includes information items of the multiple user dimensions; A data acquisition unit, configured to acquire data attribute information of each multimedia data, each data attribute information including information items of the multiple data dimensions; A calling unit configured to call the matching model to obtain a matching degree between each multimedia data and the target user according to the data attribute information of each multimedia data and the target user attribute information; The recommendation unit is configured to recommend multimedia data to the target user according to the matching degree of the plurality of multimedia data.

13. The device according to claim 12, characterized in that The matching model includes at least one feature extraction layer and at least one fusion layer, and the calling unit includes: The user feature extraction subunit is configured to execute calling the at least one feature extraction layer to extract features from the target user attribute information to obtain target user features; The data feature extraction subunit is configured to execute calling the at least one feature extraction layer to perform feature extraction on the data attribute information of any multimedia data, and obtain data features corresponding to the data attribute information of any multimedia data; The fusion subunit is configured to execute calling the at least one fusion layer to perform fusion processing on the target user feature and the data feature to obtain the matching degree between any multimedia data and the target user.

14. The device according to claim 13, characterized in that The target user attribute information at least includes the target user identification, user static information, user statistical information and user behavior information; The user feature extraction subunit is further configured to execute calling the at least one feature extraction layer to extract features from the user behavior information to obtain user behavior features; The user feature extraction subunit is further configured to combine the target user identifier, the user static information, the user statistical information and the user behavior feature to obtain the target user feature.

15. The device according to claim 13, characterized in that The data attribute information of any multimedia data at least includes data identification, data static information, data statistical information and data content information; The data feature extraction subunit is further configured to execute calling the at least one feature extraction layer to perform feature extraction on the data content information of the any multimedia data to obtain the data content feature of the any multimedia data; The data feature extraction subunit is further configured to combine the data identifier, the data static information, the data statistical information and the data content feature to obtain data features corresponding to the data attribute information of any multimedia data.

16. The device according to claim 13, characterized in that The fusion subunit is further configured to execute at least one of acquiring context information or probability information, the context information including at least one of current environment information, historical recommendation data information, or historical viewing data information, and the probability information including a probability of performing at least one preset operation on any multimedia data; The fusion subunit is further configured to execute calling the at least one fusion layer to perform fusion processing on at least one of the context information or the probability information, the target user characteristics and the data characteristics to obtain a matching degree between any multimedia data and the target user.

17. The device according to claim 12, characterized in that The recommendation unit comprises: A selection subunit is configured to select at least one multimedia data from the plurality of multimedia data in the order of the matching degree from large to small; The recommendation subunit is configured to recommend the selected multimedia data to the target user.

18. The device according to claim 12, characterized in that The recommendation unit comprises: A selection subunit is configured to select at least one multimedia data having a matching degree greater than a preset threshold from the plurality of multimedia data; The recommendation subunit is configured to recommend the selected multimedia data to the target user.

19. An electronic device, characterized in that: The electronic device comprises: one or more processors; a volatile or non-volatile memory for storing the one or more processor-executable instructions; Wherein, the one or more processors are configured to execute the matching model training method as described in any one of claims 1 to claim 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 an electronic device, the electronic device is enabled to execute the matching model training 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 an electronic device, the electronic device is enabled to execute the matching model training method as described in any one of claims 1 to 9.

Images