Server, media asset tag recommendation method, and storage medium

Abstract

The present disclosure relates to a server, a media asset tag recommendation method and a storage medium. The server comprises a controller configured to: determine a plurality of to-be-recommended media asset tags corresponding to a user according to a user portrait, the recommended media asset tags being used to represent user preferences; input the to-be-recommended media asset tags into a target recommended media asset tag model to obtain a predicted probability corresponding to each of the to-be-recommended media asset tags output by the target recommended media asset tag model, wherein the target recommended media asset tag model is obtained by training a plurality of training samples; determine at least one target recommended media asset tag from the plurality of to-be-recommended media asset tags corresponding to the user according to a plurality of predicted probabilities, and send the at least one target recommended media asset tag to a terminal device to enable the terminal device to display the at least one target recommended media asset tag. The present disclosure can recommend target media asset tags to users according to user portraits, thereby improving user satisfaction.

Description

Technical Field

[0001] This disclosure relates to the field of recommendation algorithm technology, and in particular to a server, a media asset tag recommendation method, and a storage medium. Background Technology

[0002] In existing technology, users can filter and view media assets using tags provided by a media asset application; for example, refer to... Figure 1 As shown, Figure 1 This is a schematic diagram of a media asset recommendation interface for a smart TV. Users can select multiple tags in the tag display area 21 according to their preferences. After receiving at least one selected tag from the user, the system filters multiple media assets that match the user's selected tags from the media asset database and displays these multiple media assets on the display interface. For example, after filtering, multiple media assets matching the user's tags are displayed. However, because the size of the tag display area 21 is limited, only a limited number of tags can be listed for the user to choose from. Therefore, the definition of the range of tags available for the user to choose from is usually quite broad.

[0003] For example, if a user wants to watch a movie about "Muay Thai" but there is no "Muay Thai" tag in the tag display area, the user can only narrow down the search by using the "action" tag. After the system provides the user with multiple candidate media, the user can then select from these candidates until they find the target media. If there are too many candidate media, the user is very likely to lose patience and give up watching.

[0004] Therefore, when obtaining target media assets through tags, the above method requires users to spend a lot of time selecting target media assets, and cannot efficiently provide users with the media assets they need through tag filtering, thus reducing the user experience. Summary of the Invention

[0005] To solve, or at least partially solve, the aforementioned technical problems, this disclosure provides a server, a media asset tag recommendation method, and a storage medium.

[0006] In a first aspect, this disclosure provides a server, including:

[0007] The controller is configured as follows:

[0008] Based on the user profile, multiple media asset tags to be recommended for each user are determined, and the recommended media asset tags are used to represent user preferences.

[0009] The media asset tags to be recommended are input into the target media asset tag model, and the predicted probabilities corresponding to each media asset tag to be recommended are obtained from the output of the target media asset tag model. The target media asset tag model is trained based on multiple training samples, which include: recommended media asset tags corresponding to sample user profiles and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag and the adoption result.

[0010] Based on the multiple predicted probabilities, at least one target recommended media asset tag is determined from the multiple media asset tags to be recommended corresponding to the user, and sent to the terminal device so that the terminal device displays at least one of the target recommended media asset tags.

[0011] As an optional implementation of this disclosure, the controller is specifically configured as follows:

[0012] The predicted probabilities corresponding to each of the media asset tags to be recommended are sorted, and based on the sorting results, at least one target recommended media asset tag is determined from the plurality of media asset tags to be recommended.

[0013] As an optional implementation of this disclosure, the controller is further configured to:

[0014] Obtain user profiles for multiple users, as well as user behavior data for multiple users within a preset period;

[0015] The multiple training samples are constructed based on the multiple recommended media asset tags, the click counts corresponding to the multiple recommended media asset tags, and the adoption results;

[0016] Multiple training samples are input into the initial recommended media asset tag model, and the initial recommended media asset tag model is trained using the multiple training samples to obtain the target recommended media asset tag model.

[0017] As an optional implementation of this disclosure, the server stores multiple media resources to be recommended, and the controller is further configured to:

[0018] Based on the number of clicks and the number of tags corresponding to each media asset to be recommended within a preset period, the popularity parameters corresponding to each media asset to be recommended are obtained. The popularity parameters are used to characterize the recommendation popularity of the media asset to be recommended.

[0019] Based on the popularity parameters and the preset threshold, at least one first initial media asset to be recommended is determined from the plurality of media assets to be recommended, and the number of clicks and the number of clicks on the media asset tags corresponding to each first initial media asset to be recommended are increased until the popularity parameters are equal to the preset popularity parameters.

[0020] Based on the at least one first initial media asset to be recommended, at least one tag corresponding to the first initial media asset to be recommended, the number of clicks and the number of clicks on the media asset tag, at least one second initial media asset to be recommended, at least one tag corresponding to the second initial media asset to be recommended, the number of clicks and the number of clicks on the media asset tag, a target set of media assets to be recommended is constructed.

[0021] Wherein, the at least one second initial media asset to be recommended is the other media assets to be recommended among the plurality of media assets to be recommended, excluding at least one first initial media asset to be recommended.

[0022] As an optional implementation of this disclosure, the controller is specifically configured as follows:

[0023] Obtain the click count of each of the multiple media assets to be recommended;

[0024] Obtain the number of tags corresponding to the multiple media assets to be recommended, and multiply them by the corresponding number of clicks on the media asset to obtain the number of clicks on the media asset tags;

[0025] For each media asset to be recommended, the popularity parameters corresponding to each media asset to be recommended are obtained based on the number of clicks on the media asset, the number of clicks on the media asset tags, the number of media asset tags, and the number of media assets to be recommended.

[0026] As an optional implementation of this disclosure, the controller is further configured to:

[0027] After obtaining the target recommended media asset tag, based on the target recommended media asset tag and the tags corresponding to each target recommended media asset in the target recommended media asset set, at least one target recommended media asset corresponding to the target recommended media asset tag is determined in the target recommended media asset set and sent to the terminal device so that the terminal device displays the at least one target recommended media asset.

[0028] As an optional implementation of this disclosure, the controller is further configured to:

[0029] Obtain the evaluation parameters corresponding to the target recommended media asset tags;

[0030] Based on the evaluation parameters and the target recommended media asset tags, update multiple training samples;

[0031] The target recommendation media asset label model is periodically updated using the updated training samples.

[0032] Secondly, a media asset tag recommendation method is provided, the method comprising:

[0033] Based on the user profile, multiple media asset tags to be recommended for each user are determined, and the recommended media asset tags are used to represent user preferences.

[0034] The media asset tags to be recommended are input into the target media asset tag model, and the predicted probabilities corresponding to each media asset tag to be recommended are obtained from the output of the target media asset tag model. The target media asset tag model is trained based on multiple training samples, which include: recommended media asset tags corresponding to sample user profiles and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag and the adoption result.

[0035] Based on the multiple predicted probabilities, at least one target recommended media asset tag is determined from the multiple media asset tags to be recommended corresponding to the user, and sent to the terminal device so that the terminal device displays at least one of the target recommended media asset tags.

[0036] As an optional implementation of this disclosure, after inputting the media asset tags to be recommended into the target media asset tag model and obtaining the predicted probabilities corresponding to each of the media asset tags to be recommended output by the target media asset tag model, the method further includes:

[0037] The predicted probabilities corresponding to each of the media asset tags to be recommended are sorted, and based on the sorting results, at least one target recommended media asset tag is determined from the plurality of media asset tags to be recommended.

[0038] Thirdly, a computer-readable storage medium is provided, comprising: storing a computer program on the computer-readable storage medium, wherein when the computer program is executed by a processor, it implements the media asset tag recommendation method as shown in the second aspect.

[0039] The technical solution provided in this disclosure has the following advantages compared with the prior art:

[0040] In the technical solution provided in this embodiment, the server controller determines multiple media asset tags to be recommended for a user based on the user profile. These media asset tags are used to characterize user preferences. The media asset tags to be recommended are input into a target media asset tag model to obtain the predicted probabilities corresponding to each media asset tag output by the target media asset tag model. The target media asset tag model is trained based on multiple training samples, including: recommended media asset tags corresponding to the sample user profile and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag and the adoption result. Based on the multiple predicted probabilities, at least one target recommended media asset tag is determined from the multiple media asset tags to be recommended for the user and sent to the terminal device so that the terminal device displays at least one target recommended media asset tag. In the above technical solution, multiple media asset tags to be recommended can be obtained by combining user profiles. Then, the predicted probability corresponding to each media asset tag to be recommended can be calculated by using a target recommended media asset tag model. Finally, at least one target recommended media asset tag is determined and sent to the terminal device. This allows the user to be provided with at least one media asset tag that matches the user's preferences, eliminating the need for the user to select tags from multiple fixed tags for media asset filtering. This application can recommend tags that match the user's preferences based on user preferences. Based on this, the accuracy of media asset filtering by media asset tags can be improved, thus ensuring that more accurate candidate media assets are provided to the user when filtering media assets by media asset recommendation tags, thereby improving the user experience. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0042] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of a media asset recommendation interface for a smart TV.

[0044] Figure 2 This is a schematic diagram illustrating an application scenario of a media asset tag recommendation method provided in an embodiment of this disclosure;

[0045] Figure 3 This is a hardware configuration block diagram of a server according to one or more embodiments of the present disclosure;

[0046] Figure 4 This is a framework diagram of a media asset recommendation system based on one or more embodiments of this disclosure;

[0047] Figure 5 A schematic flowchart illustrating a media asset tag recommendation method provided in this embodiment of the disclosure;

[0048] Figure 6 A schematic diagram of the interface of a display device for another media asset tag recommendation method provided in an embodiment of this disclosure;

[0049] Figure 7 A schematic diagram of a media asset recommendation interface provided in an embodiment of this disclosure;

[0050] Figure 8 This is a flowchart illustrating another media asset tag recommendation method provided in this embodiment of the disclosure;

[0051] Figure 9 This is a flowchart illustrating another media asset tag recommendation method provided in this embodiment of the disclosure;

[0052] Figure 10 This is a flowchart illustrating another media asset tag recommendation method provided in this embodiment. Detailed Implementation

[0053] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0054] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0055] The terms "first," "second," "third," etc., used in this disclosure, in the specification, claims, and accompanying drawings are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0056] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0057] In existing technology, users can filter and view media assets using tags provided by a media asset application; for example, refer to... Figure 1 As shown, Figure 1 This is a schematic diagram of a media asset recommendation interface for a smart TV. Users can select multiple tags in the tag display area 21 according to their preferences. After receiving at least one selected tag from the user, the system filters multiple media assets that match the user's selected tags from the media asset database and displays these multiple media assets on the display interface. For example, after filtering, multiple media assets matching the user's tags are displayed. However, because the size of the tag display area 21 is limited, only a limited number of tags can be listed for the user to choose from. Therefore, the definition of the range of tags available for the user to choose from is usually quite broad.

[0058] For example, if a user wants to watch a movie about "Muay Thai" but there is no "Muay Thai" tag in the tag display area, the user can only narrow down the search by using the "action" tag. After the system provides the user with multiple candidate media, the user can then select from these candidates until they find the target media. If there are too many candidate media, the user is very likely to lose patience and give up watching.

[0059] Therefore, when obtaining target media assets through tags, the above method requires users to spend a lot of time selecting target media assets, and cannot efficiently provide users with the media assets they need through tag filtering, thus reducing the user experience.

[0060] To address the aforementioned issues, this disclosure proposes a server, a media asset tag recommendation method, and a storage medium. The server controller determines multiple media asset tags to be recommended for a user based on a user profile. These recommended media asset tags characterize user preferences. The tags to be recommended are input into a target recommended media asset tag model to obtain the predicted probabilities corresponding to each of the tags. The target recommended media asset tag model is trained using multiple training samples, including: recommended media asset tags corresponding to sample user profiles and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag and the adoption result. Based on the multiple predicted probabilities, at least one target recommended media asset tag is determined from the multiple tags to be recommended for the user and sent to a terminal device, so that the terminal device displays at least one target recommended media asset tag. In the above technical solution, multiple media asset tags to be recommended can be obtained by combining user profiles. Then, the predicted probability corresponding to each media asset tag to be recommended can be calculated by using a target recommended media asset tag model. Finally, at least one target recommended media asset tag is determined and sent to the terminal device. This allows the user to be provided with at least one media asset tag that matches the user's preferences, eliminating the need for the user to select tags from multiple fixed tags for media asset filtering. This application can recommend tags that match the user's preferences based on user preferences. Based on this, the accuracy of media asset filtering by media asset tags can be improved, thus ensuring that more accurate candidate media assets are provided to the user when filtering media assets by media asset recommendation tags, thereby improving the user experience.

[0061] For example, such as Figure 2 As shown, Figure 2 This is a schematic diagram illustrating an application scenario of a media asset tag recommendation method provided in this embodiment of the disclosure. Figure 2 In this embodiment, server 100 uses a target recommended media asset tagging model to obtain the predicted probabilities corresponding to each media asset in the media asset dataset to be recommended. Based on multiple predicted probabilities, server 100 determines at least one target recommended media asset from among the media assets to be recommended. Server 100 then sends the at least one target recommended media asset to terminal device 200, so that terminal device 200 displays the at least one target recommended media asset. In this embodiment, terminal device 200 can be a smart device with smart display capabilities, such as a smart TV, smart tablet, or smartphone. This disclosure does not limit the specific type of terminal device.

[0062] In some embodiments, the terminal device 200 also communicates with the server 100. The terminal device 200 may communicate via a local area network (LAN), wireless local area network (WLAN), and other networks. The server 100 may provide various content and interactive features to the terminal device 200. The server may be a server that provides various services, such as a server that supports media data collected by the terminal device. The server may analyze and process the received media data and feed back the processing results (e.g., endpoint information) to the terminal device. The server may be a server cluster or multiple server clusters, and may include one or more types of servers.

[0063] The media asset tag recommendation method provided in this disclosure can be implemented based on a server, or a functional module or functional entity within the server.

[0064] For example, Figure 3 This is a hardware configuration block diagram of a server according to one or more embodiments of this disclosure. Figure 3 As shown, the server includes at least one of the following: a tuner / demodulator 310, a communicator 320, a controller 330, a display interface 340, a memory, and a power supply. The controller 330 includes a central processing unit, a video processor, an audio processor, a graphics processor, RAM, ROM, and a first to nth interface for input / output. The display interface 340 may be a VGA (Video Graphics Array). Generally, server graphics card requirements are not high, so server graphics card interfaces are typically VGA interfaces, generally used for installing the server operating system or for daily debugging. The tuner / demodulator 310 receives broadcast television signals via wired or wireless means, and demodulates audio and video signals, such as EPG audio and video data signals, from multiple wireless or wired broadcast television signals. The communicator 320 is a component used to communicate with external devices according to various communication protocol types. For example, the communicator may include at least one of the following: a Wi-Fi module, a Bluetooth module, a wired Ethernet module, other network communication protocol chips or near-field communication protocol chips, and an infrared receiver. The server can establish the transmission and reception of control signals and data signals with the terminal device 300 or the local control device through the communicator 320. The controller 330 and the tuner / demodulator 310 can be located in different separate devices, that is, the tuner / demodulator 310 can also be located in an external device of the main device where the controller 330 is located, such as an external set-top box.

[0065] In some embodiments, the controller 330 controls the operation of the server and responds to user operations through various software control programs stored in memory. The controller 330 controls the overall operation of the server. Users can input commands through a graphical user interface (GUI) displayed on the display interface 340, and the user input interface receives user input commands through the graphical user interface (GUI).

[0066] Figure 4 A framework diagram of a media asset recommendation system based on one or more embodiments of this disclosure is provided, such as... Figure 4 As shown, the system may include an acquisition module 41, a prediction probability acquisition module 42, and a target recommended media asset tag determination module 43. The acquisition module 41 is used to determine multiple media asset tags to be recommended for a user based on the user profile, where the recommended media asset tags represent user preferences. The prediction probability acquisition module 42 is used to input the media asset tags to be recommended into a target recommended media asset tag model, and obtain the prediction probabilities corresponding to each of the media asset tags to be recommended output by the target recommended media asset tag model. The target recommended media asset tag model is trained based on multiple training samples, including: recommended media asset tags corresponding to sample user profiles and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each of the recommended media asset tags and the adoption results. The target recommended media asset tag determination module 43 is used to determine at least one target recommended media asset tag from the multiple media asset tags to be recommended for the user based on the multiple prediction probabilities, and send it to the terminal device so that the terminal device displays at least one target recommended media asset tag.

[0067] In the above technical solution, the target recommended media asset tag model is trained based on multiple training samples. These training samples include: recommended media asset tags corresponding to the sample user profiles and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks and adoption results for each recommended media asset tag. This allows for the calculation of multiple media asset tags to be recommended by combining the user profile and its corresponding user behavior data, obtaining the corresponding prediction probabilities. Based on this, at least one media asset recommendation tag that matches the user's preferences can be fed back to the user through the prediction probability. Thus, when filtering media assets through media asset recommendation tags, more accurate candidate media assets are provided to the user, improving the user experience.

[0068] To illustrate this solution in more detail, the following will use examples to illustrate it. Figure 5To clarify, although the steps in flowchart 5 are shown sequentially as indicated by the arrows, these steps are not necessarily executed in that exact order. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps; they can be performed in other orders. Furthermore, Figure 5 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps. The method should be designed to implement the media asset tag recommendation method provided in the embodiments of this disclosure.

[0069] Figure 5 This is a flowchart illustrating a media asset tag recommendation method provided in an embodiment of this disclosure, such as... Figure 5 As shown, the media asset tag recommendation method specifically includes the following steps:

[0070] S501. Based on the user profile, determine multiple media asset tags to be recommended for each user.

[0071] The recommended media asset tags are used to characterize user preferences.

[0072] In some embodiments, users can select and watch media assets on various display devices through various media asset applications on the market. For example, users can log in to their personal accounts to watch TV series, movies, etc., through media asset applications on a television. Therefore, in this embodiment, user profiles are created based on the user's historical viewing records to obtain user profiles corresponding to different users. The essence of the user profile is a description of the user's needs. Specifically, the user profile is a user characteristic that reflects the user's preferences. When recommending media assets to users, media asset tags can be recommended in conjunction with the user profile. Tags are abstractions and descriptions of user attributes, interests, behaviors, and other characteristics, reflecting the specific characteristics of the current user. This allows for the recommendation of tags that are closer to the user's needs and satisfy the user as much as possible when recommending media asset tags in the future, in conjunction with the user profile.

[0073] For example, if user A's media viewing history is mostly suspense and horror movies, then user A's user profile can be obtained through the viewing history and user A's account information: user A is a young male user who prefers suspense, horror, and thriller genres; based on user A's user profile, it can be determined that multiple media asset tags corresponding to user A can include media asset tags of the same type such as suspense, horror, and thriller.

[0074] It should be noted that the multiple media asset tags to be recommended also include multiple media asset tags obtained based on user behavior data. Specifically, the system establishes a log to store user behavior data, i.e., user viewing records. By reading this log, the user's viewing history within a preset time period can be obtained, and the corresponding media asset tags can be further extracted from the viewing history. For example, by analyzing the viewing history, if the user has recently watched mostly science fiction films, it can be determined that the user recently enjoys watching science fiction films, and science fiction can be included as one of the multiple media asset tags to be recommended. This makes the multiple tags among the multiple media asset tags to be recommended more closely aligned with the user's current preferences, thereby recommending media assets that better suit the user's needs.

[0075] S502. Input the media asset tags to be recommended into the target media asset tag model, and obtain the predicted probabilities corresponding to each media asset tag to be recommended output by the target media asset tag model.

[0076] The target recommended media asset tag model is trained based on multiple training samples. The training samples include: recommended media asset tags corresponding to the sample user profiles, and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag, and the adoption results.

[0077] In some embodiments, the target recommended media asset tag model can be obtained by training any neural network model; it is used to combine user profiles, use model inference to give the predicted probability of each corresponding tag to be recommended, and select target recommended media asset tags to feed back to the user based on the magnitude of the predicted probability.

[0078] Specifically, when inputting the media asset tags to be recommended into the target recommended media asset tag model, the media asset tags obtained from the user profile corresponding to the user and the media asset tags obtained based on user behavior data are input into the target recommended media asset tag model in the form of a feature matrix. Taking user 1 as an example, the media asset tags corresponding to user 1's user profile will be converted into vector form to obtain the feature matrix A about the user profile. i Then, initialize a 1-column, m-dimensional vector B containing all zeros. i,j Iterate through all media assets clicked by the user within a preset period and obtain the corresponding media asset tag sequence; if the current media asset tag is p, and tag p has not been adopted by the user, then add media asset tag p to vector B. i,j The number at position B is set to -1. If the media asset tag p is accepted by the user, then the media asset tag p is placed in vector B. i,j The number at position B is set to 1. That is, vector B can be used. i,jThis indicates whether user 1's media asset tag was adopted in round j. Similarly, initialize a 1-column, m-dimensional vector C containing all zeros. i,j Iterate through all media assets clicked by the user within a preset period, obtain the corresponding media asset tag sequence, and if the media asset tag p has been recommended twice, then add the media asset tag p to the vector C. i,j The number at position 2 represents the number of times user 1 is recommended by media asset tag p in the j-th round.

[0079] The above three feature matrices, namely the feature matrix A of multiple media asset tags to be recommended, obtained from the user profile of user 1, are used to... i And the feature matrix B of the adoption results of multiple media asset tags to be recommended, obtained based on user 1's behavioral data. i,j The feature matrix C of the number of adoptions of multiple media asset tags. i,j The data can be input into the target recommended media asset tag model, which can then perform probability prediction on multiple media asset tags to be recommended, and obtain the predicted probability corresponding to each media asset tag to be recommended.

[0080] It should be noted that the target recommendation media asset tagging model can be obtained by training a machine learning algorithm model such as a deep learning neural network model or a convolutional neural network model using training samples. This embodiment of the invention does not limit the specific type of the target recommendation tagging model.

[0081] S503. Based on the multiple predicted probabilities, determine at least one target recommended media asset tag among the multiple media asset tags to be recommended corresponding to the user, and send it to the terminal device so that the terminal device displays at least one of the target recommended media asset tags.

[0082] In some embodiments, after obtaining the predicted probabilities corresponding to the plurality of media asset tags to be recommended, the degree to which each tag matches the user can be determined based on the magnitude of the predicted probabilities corresponding to each recommended media asset tag. In this embodiment, the recommended media asset tag with the highest predicted probability can be fed back to the user as the target media asset tag based on the magnitude of the predicted probabilities corresponding to each recommended media asset tag. Alternatively, a predicted probability threshold can be set, and tags with a predicted probability greater than the threshold can be fed back to the user as the target tags.

[0083] Optionally, the predicted probabilities corresponding to each of the media asset tags to be recommended can be sorted, and at least one target recommended media asset tag can be determined from the plurality of media asset tags to be recommended based on the sorting results.

[0084] Specifically, corresponding weight coefficients can be pre-set for the media asset tags among the multiple media asset tags to be recommended. If you want to obtain media asset tags that are closer to the user profile, you can set the weight coefficient of the corresponding media asset tags in the user profile to be relatively large. If you want to obtain media asset tags that are closer to the user's most recent behavior data, you can set the weight coefficient of the media asset tags obtained from the user behavior data to be relatively large. This is to meet the user needs in different scenarios.

[0085] In this embodiment of the application, after obtaining the target recommended media asset tag, based on the target recommended media asset tag and the tags corresponding to each target recommended media asset in the target recommended media asset set, at least one target recommended media asset corresponding to the target recommended media asset tag is determined in the target recommended media asset set and sent to the terminal device so that the terminal device displays the at least one target recommended media asset.

[0086] Reference Figure 6 As shown, this is a tag recommendation interface provided in an embodiment of this application. After obtaining the at least one target recommendation media asset tag, the at least one target recommendation tag is displayed on... Figure 6 The target recommended tag display interface 61 includes two touch buttons, one for "accept" and one for "disappear," allowing users to choose whether to accept or reject at least one target recommended media asset tag. If accepted, the system updates the target recommended media asset set based on the current tag and provides a new media asset tag. If rejected, the system uses the target recommended media asset tag model to mark the media asset tags that the user has not accepted to avoid recommending the same or similar media asset tags in the next round. The next round of tag calculation continues until the user accepts a media asset tag and clicks on the target media asset, or exits the current recommendation process.

[0087] In the technical solution provided in this embodiment, the server controller determines multiple media asset tags to be recommended for a user based on the user profile. These media asset tags are used to characterize user preferences. The media asset tags to be recommended are input into a target media asset tag model to obtain the predicted probabilities corresponding to each media asset tag output by the target media asset tag model. The target media asset tag model is trained based on multiple training samples, including: recommended media asset tags corresponding to the sample user profile and user behavior data corresponding to the recommended media asset tags. The user behavior data includes: the number of clicks on each recommended media asset tag and the adoption result. Based on the multiple predicted probabilities, at least one target recommended media asset tag is determined from the multiple media asset tags to be recommended for the user and sent to the terminal device so that the terminal device displays at least one target recommended media asset tag. In the above technical solution, multiple media asset tags to be recommended can be obtained by combining user profiles. Then, the predicted probability corresponding to each media asset tag to be recommended can be calculated by using a target recommended media asset tag model. Finally, at least one target recommended media asset tag is determined and sent to the terminal device. This allows the user to be provided with at least one media asset tag that matches the user's preferences, eliminating the need for the user to select tags from multiple fixed tags for media asset filtering. This application can recommend tags that match the user's preferences based on user preferences. Based on this, the accuracy of media asset filtering by media asset tags can be improved, thus ensuring that more accurate candidate media assets are provided to the user when filtering media assets by media asset recommendation tags, thereby improving the user experience.

[0088] As an extension and refinement of the above embodiments, refer to Figure 7 As shown, the method for obtaining the target recommendation media asset tag model can refer to the following steps:

[0089] S701. Obtain user profiles for multiple users, as well as user behavior data for multiple users within a preset period.

[0090] The user behavior data of each user within a preset period includes the number of clicks on the recommended media asset tags and the adoption results.

[0091] In some embodiments, the preset period can be the number of clicks on the recommended media asset tags within the most recent month, as well as the adoption results; wherein, the adoption results can be divided into adoption and non-adoption.

[0092] In this embodiment, a large amount of user profiles, user behavior data, and target tags are collected as sample data to train the initial recommendation media asset tag model. For example, assume the user set U = [u1, u2, ... u...]. n ], where u iThis represents the sequence of media assets clicked by the i-th user within a preset time period, M = [m i，1 m i，2 ...m i，n For media asset j in the media asset sequence, its tag set is T = [t] j，1 t j，2 ...t j，n ].

[0093] S702. Construct the multiple training samples based on the multiple recommended media asset tags, the click counts corresponding to the multiple recommended media asset tags, and the adoption results.

[0094] S703. Input multiple training samples into the initial recommended media asset tag model, and train the initial recommended media asset tag model using the multiple training samples to obtain the target recommended media asset tag model.

[0095] During model training, it is also necessary to input the feature matrices of multiple recommended media asset tags obtained from user profiles of multiple users, the feature matrices of the adoption results of multiple recommended media asset tags obtained from the behavioral data of multiple users, and the feature matrices of the adoption times of multiple media asset tags; and the corresponding recommended media asset tags, into the initial recommended media asset tag model for training.

[0096] During training, an auxiliary model can be set up. For example, the auxiliary model can be a Deep Q-Network (DQN) target network model with the same structure as the recommended media asset tag model. The optimized model parameters can be obtained by calculating the model loss. The parameters of the recommended media asset tag model can be updated once every preset number of training iterations using the optimized model parameters obtained from the DQN target network model. This completes the model update and makes the model more optimized.

[0097] Specifically, the formula for calculating the loss is as follows:

[0098] loss = loss_func(model) e (S)Score+γ*model t (S′))

[0099] Where loss represents the magnitude of the loss, γ represents the preset number of training iterations, and model e This represents the current recommended media asset tagging model, where S represents model. e Current model state, model t Let s' represent the auxiliary model, and s' represent model. eIn the updated model state based on the simulated user adoption results, Score represents the evaluation parameter obtained from the simulated user adoption results, i.e., the reward size.

[0100] As an extension and refinement of the above embodiments, in the embodiments of this application, reference is made to... Figure 8 As shown, the server stores multiple media resources to be recommended, and the controller is further configured to:

[0101] S801. Based on the number of clicks and the number of tags corresponding to each media asset to be recommended within a preset period, obtain the popular parameters corresponding to each media asset to be recommended.

[0102] The popularity parameter is used to characterize the popularity of the media asset to be recommended.

[0103] In this embodiment, when obtaining at least one target recommended media asset based on the at least one target recommendation tag, it is obtained from the plurality of media assets to be recommended. Therefore, when recommending at least one target recommended media asset, it is easily affected by the current popularity of the media asset. Under the condition of satisfying the at least one target recommendation tag, the user is often fed back media assets with high popularity while ignoring less popular media assets. However, it is possible that some users want to find less popular media assets. Therefore, in order to make the media asset data including multiple media assets to be recommended more fair, this embodiment calculates the scarcity of multiple media assets to be recommended and determines to recommend less popular media assets with lower popularity, so as to increase the popularity parameter of less popular media assets, so that the popularity parameter of multiple media assets to be recommended is consistent, and there are no media assets with high or low popularity.

[0104] In some embodiments, the specific method for obtaining the popularity parameters corresponding to each media asset to be recommended based on the number of clicks and the number of tags corresponding to each media asset to be recommended within a preset period, as described in step S801 above, can be referred to as follows:

[0105] S8011. Obtain the number of clicks for each of the multiple media assets to be recommended.

[0106] S8012. Obtain the number of media asset tags corresponding to the multiple media assets to be recommended, and multiply them by the corresponding media asset click count to obtain the media asset tag click count.

[0107] S8013. For each media asset to be recommended, based on the number of clicks on the media asset, the number of clicks on the media asset tags, the number of media asset tags, and the number of media assets to be recommended, obtain the popularity parameters corresponding to each of the multiple media assets to be recommended.

[0108] Specifically, the popularity parameter is used to characterize the popularity of the media asset to be recommended. The larger the popularity parameter, the lower the popularity of the media asset and the higher its scarcity. The popularity parameter can be represented by Q, and the formula for calculating the popularity parameter is as follows:

[0109]

[0110] Among them, c i,1 c represents the number of times the media asset has been clicked by users. i,2 This represents the number of times the media asset's tags have been clicked by users, where m represents the number of media assets to be recommended, and k represents the number of tags associated with the media asset. Based on the above formula, the popularity parameters corresponding to the media asset are calculated to measure its popularity.

[0111] S802. Based on the popularity parameter and the preset threshold, at least one first initial media asset to be recommended is determined from the plurality of media assets to be recommended, and the click count and media asset tag click count corresponding to each first initial media asset to be recommended are increased until the popularity parameter equals the preset popularity parameter. It should be noted that the preset popularity parameter may be based on the media asset click count corresponding to the media asset with the highest popularity among the plurality of media assets to be recommended; or it may be a click count set by the developers based on the click counts of the plurality of media assets to be recommended, with the popularity being relatively in the middle. This application does not impose any limitation on this.

[0112] In some embodiments, a threshold for the number of media assets can be set to specify the number of media assets to be recommended each time from the plurality of media assets to be recommended. For example, the threshold can be set to 100, that is, 100 media assets to be recommended are selected each time. The popularity parameters of these 100 media assets to be recommended are calculated. Then, based on the size of the popularity parameters, the media assets to be recommended are selected from the 100 media assets to be recommended in descending order of popularity parameters, and the number of media assets to be recommended is 100 multiplied by the preset threshold number of media assets to be recommended, to generate the first initial media assets to be recommended. If the preset threshold is 0.1, then the top 10 media assets to be recommended with the largest popularity parameters among the 100 media assets to be recommended are finally selected as the second initial media assets to be recommended. It should be noted that those skilled in the art can set the preset threshold according to the actual situation.

[0113] S803. Construct a target set of media assets to be recommended based on the at least one first initial media asset to be recommended, at least one tag corresponding to the first initial media asset to be recommended, the number of clicks and the number of clicks on the media asset tag, at least one second initial media asset to be recommended, at least one tag corresponding to the second initial media asset to be recommended, the number of clicks and the number of clicks on the media asset tag.

[0114] Wherein, the at least one second initial media asset to be recommended is the other media assets to be recommended among the plurality of media assets to be recommended, excluding at least one first initial media asset to be recommended.

[0115] After the above steps, the final target set of recommended media assets meets the user's needs without ignoring media assets with low popularity, avoiding the extreme practice of only recommending highly popular media assets, and providing the user with a more comprehensive set of target recommended media assets.

[0116] Optionally, based on the above embodiments, in some embodiments of this disclosure, since users have different interests and preferences and different needs for media assets at different times, in order to ensure that the target recommendation media asset tagging model can provide users with accurate target recommendation media assets and improve the user experience, the media asset tagging recommendation method provided in this application further includes the following steps:

[0117] S1001. Obtain the evaluation parameters corresponding to the target recommended media asset tags.

[0118] In some embodiments, evaluation parameters corresponding to the target recommended media asset tags can be obtained through user feedback. Specifically, evaluation can be based on whether the user adopts or does not adopt the tag, and further evaluation can be based on three conditions: the adopted tag matches the media asset tag to be recommended corresponding to the user profile; the adopted tag matches the media asset tag to be recommended corresponding to the user behavior data; and the adopted tag is a tag that has already been recommended. The target recommended media asset tags are scored according to these three conditions to obtain the evaluation parameters, so as to measure whether the current target recommended media asset tags meet the user's needs. The target recommended media asset tag model can be further optimized using these evaluation parameters.

[0119] Specifically, by calculating the scores corresponding to the above three conditions and adding them together, the scores can be used as the evaluation parameters corresponding to the target recommended media asset tags.

[0120] Optionally, corresponding weight coefficients can be set for the above three conditions to adjust the focus of the model's output target recommended media asset tags. If you want the model's output target recommended media asset tags to focus more on user profiles, that is, user features, then you can adjust the weight coefficient of the adopted tags that match the user profiles to be recommended media asset tags so that the weight coefficient of this condition has the largest proportion among the three conditions.

[0121] S1002. Update multiple training samples based on the evaluation parameters and the target recommended media asset tags.

[0122] In some embodiments, by combining the evaluation parameters corresponding to the target recommended media asset tags, it is possible to more clearly obtain the user's current needs for media asset tags, so as to update the training samples, adjust the model strategy in a timely manner, and make the model more closely match the current needs of users.

[0123] S1003. Using the updated training samples, periodically update the target recommendation media asset label model.

[0124] The term "periodic" refers to the ability to update the target recommendation media asset tag model using multiple updated training samples at preset intervals.

[0125] Specifically, the server controller periodically acquires user information and a set of popular media assets, and updates multiple training samples based on the user information and the set of popular media assets. Then, using the updated multiple training samples, the target recommendation media asset tag model is periodically updated.

[0126] In the technical solution provided by this disclosure, during the above process, multiple training samples can be periodically updated based on user information and popular media asset sets. Using the updated multiple training samples, the target recommendation media asset tag model is periodically updated, thereby ensuring that the target recommendation media asset tag model can provide users with accurate target recommendation media assets and improve the user experience.

[0127] This disclosure also provides a storage medium containing computer-executable instructions. When executed by a computer processor, the computer-executable instructions implement the various processes of the methods provided in any of the above embodiments and achieve the same technical effects. To avoid repetition, further details are omitted here.

[0128] The computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0129] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the discussion in some embodiments above is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be obtained based on the above teachings. The selection and description of the above embodiments are for the purpose of better explaining the principles and practical applications, thereby enabling those skilled in the art to better utilize the embodiments and various different variations of the embodiments suitable for specific application considerations.

Claims

1. A server, characterized by The method comprises the following steps: The controller is configured to: According to the user portrait, determine a plurality of to-be-recommended media tags corresponding to the user, wherein the recommended media tags are used to represent the user's preferences; Input the to-be-recommended media tags into a target recommended media tag model, and obtain a predicted probability corresponding to each to-be-recommended media tag output by the target recommended media tag model, wherein the target recommended media tag model is obtained by training a plurality of training samples, and the training samples comprise a recommended media tag corresponding to a sample user portrait and user behavior data corresponding to the recommended media tag, wherein the user behavior data comprises a click number of each recommended media tag and an adoption result; According to a plurality of predicted probabilities, determine at least one target recommended media tag from the plurality of to-be-recommended media tags corresponding to the user, and send the at least one target recommended media tag to a terminal device, so that the terminal device displays the at least one target recommended media tag. The server stores a plurality of to-be-recommended media, and the controller is configured to: According to a click number and a number of tags corresponding to each to-be-recommended media in a preset period, obtain a popular parameter corresponding to each to-be-recommended media, wherein the popular parameter is used to represent a recommended popularity of the to-be-recommended media; According to the popular parameter and a preset threshold, determine at least one first initial to-be-recommended media from the plurality of to-be-recommended media, and increase a click number and a media tag click number corresponding to each first initial to-be-recommended media until the popular parameter is equal to a preset popular parameter, wherein the first initial to-be-recommended media is at least one media with low recommended popularity selected from the plurality of to-be-recommended media according to a size of the popular parameter; According to the at least one first initial to-be-recommended media, at least one tag corresponding to the first initial to-be-recommended media, the click number and the media tag click number, at least one second initial to-be-recommended media, at least one tag corresponding to the second initial to-be-recommended media, the click number and the media tag click number, construct a target to-be-recommended media set. The at least one second initial to-be-recommended media is other to-be-recommended media in the plurality of to-be-recommended media except the at least one first initial to-be-recommended media.

2. The server of claim 1, wherein, The controller is specifically configured to: Sort the predicted probabilities corresponding to each to-be-recommended media tag, and determine at least one target recommended media tag from the plurality of to-be-recommended media tags according to a sorting result.

3. The server of claim 1, wherein, The controller is further configured to: Obtain a user portrait corresponding to each of a plurality of users and user behavior data of the plurality of users in a preset period, wherein the user behavior data of the plurality of users in the preset period comprises a click number and an adoption result of the recommended media tag; According to a plurality of recommended media tags, a click number and an adoption result corresponding to each of the plurality of recommended media tags, construct the plurality of training samples; Input the plurality of training samples into an initial recommended media tag model, train the initial recommended media tag model by using the plurality of training samples, and obtain the target recommended media tag model.

4. The server of claim 3, wherein, The controller is specifically configured to: Obtain the number of clicks of the plurality of media to be recommended respectively corresponding to the media; Obtain the number of tags corresponding to the plurality of media to be recommended respectively, and multiply the number of tags by the number of clicks of the corresponding media to obtain the number of clicks of the media tags; For each of the media to be recommended, the number of clicks of the media, the number of clicks of the media tags, the number of media tags, and the number of media to be recommended are obtained. The hot parameter corresponding to the plurality of media to be recommended is obtained.

5. The server of claim 3, wherein, The controller is further configured to: After obtaining the target recommended media tag, according to the target recommended media tag, and the target recommended media tag corresponding to each target recommended media in the target recommended media set, at least one target recommended media corresponding to the target recommended media tag is determined in the target recommended media set, and is sent to the terminal device, so that the terminal device displays the at least one target recommended media.

6. The server of claim 1, wherein, The controller is further configured to: Obtain the evaluation parameter corresponding to the target recommended media tag; According to the evaluation parameter and the target recommended media tag, update a plurality of training samples; Periodically update the target recommended media tag model using the updated plurality of training samples.

7. A method of media tag recommendation, the method comprising: The method comprises: According to the user portrait, a plurality of recommended media tags corresponding to the user are determined, and the recommended media tags are used to represent the user's preferences; Input the recommended media tags into the target recommended media tag model to obtain the prediction probability of each recommended media tag output by the target recommended media tag model, wherein the target recommended media tag model is obtained by training a plurality of training samples, and the training samples include: recommended media tags corresponding to sample user portraits, user behavior data corresponding to the recommended media tags, and the user behavior data includes: the number of clicks of each recommended media tag, and the adoption result; According to a plurality of prediction probabilities, at least one target recommended media tag is determined from the plurality of recommended media tags corresponding to the user, and is sent to a terminal device, so that the terminal device displays at least one target recommended media tag; The server stores a plurality of recommended media; According to the number of clicks, the number of tags corresponding to each of the plurality of recommended media in a preset period, obtain the hot parameter corresponding to each of the plurality of recommended media, and the hot parameter is used to represent the recommended hot degree of the recommended media; According to the hot parameter and the preset threshold, at least one first initial recommended media is determined from the plurality of recommended media, and the number of clicks and the number of clicks of the media tags corresponding to each first initial recommended media are increased until the hot parameter is equal to the preset hot parameter; According to the at least one first initial recommended media, the at least one tag corresponding to the first initial recommended media, the number of clicks and the number of clicks of the media tags, at least one second initial recommended media, at least one tag corresponding to the second initial recommended media, the number of clicks and the number of clicks of the media tags, a target recommended media set is constructed; The at least one second initial recommended media asset is other than the at least one first initial recommended media asset in the plurality of recommended media assets.

8. The method of claim 7, wherein, After inputting the recommended media asset tags into a target recommended media asset tag model and obtaining a predicted probability corresponding to each of the recommended media asset tags output by the target recommended media asset tag model, the method further comprises: sorting the predicted probabilities corresponding to each of the recommended media asset tags, and determining at least one target recommended media asset tag from the plurality of recommended media asset tags according to a sorting result.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the media asset tag recommendation method of any one of claims 7-8.

Images