Media recommendation method and apparatus, and computer device and storage medium

Abstract

The present application relates to a media recommendation method and apparatus, and a computer device, a storage medium and a computer program product. The method comprises: acquiring an interaction feature set between an object set and a media set, wherein the interaction feature set is a set generated on the basis of historical media transaction records of objects in the object set, and the historical media transaction records are item transaction records of the objects with regard to recommended items corresponding to historical interactive media (step S202); decomposing the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set (step S204); determining a target object from the object set, acquiring, from the media set, media with which the target object has not interacted, using said media as target media, acquiring from the object feature set a target object feature of the target object, and acquiring from the media feature set a target media feature of the target media (step S206); on the basis of the target object feature and the target media feature, determining the matching degree of transaction between the target object and the target media (step S208); and on the basis of the matching degree of transaction, determining, from among the target media, recommended media corresponding to the target object (step S210).

Description

Media recommendation method and device, computer device, and storage medium

[0001] The present application claims priority to the Chinese patent application No. 2024110099996, filed on July 25, 2024, and entitled "Media recommendation method and device, computer device, and storage medium", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] The present application relates to the technical field of computer, in particular to a media recommendation method and device, computer device, storage medium and computer program product. BACKGROUND

[0003] With the development of computer technology, people are exposed to more and more media in daily life, such as videos, music, articles, and graphic information. Internet has also emerged with recommendation functions, for example, in video applications, users are recommended videos that they may be interested in. The recommendation function supports personalized recommendation for users.

[0004] In the traditional technology, media recommendation is usually based on the user side or the media side. For the user side, media recommendation is usually based on user similarity, and media that similar users have interacted with is recommended to the user. For the media side, media recommendation is usually based on media similarity, and other media similar to the media that the user has interacted with is recommended to the user. However, these one-sided media recommendation methods currently have limitations, resulting in low effectiveness of media recommendation. SUMMARY

[0005] The embodiments of the present application provide a media recommendation method, device, computer device, computer readable storage medium and computer program product, which can improve the effectiveness of media recommendation.

[0006] The present application provides a media recommendation method, which is executed by a computer device, comprising:

[0007] obtaining an interaction feature set between an object set and a media set; the interaction feature set is a set generated based on historical media transaction records of objects in the object set, and the historical media transaction records are object transaction records of corresponding recommended items for historical interaction media;

[0008] decomposing the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set;

[0009] determine a target object from the object set, obtain media that the target object has not interacted with from the media set as target media, obtain a target object feature of the target object from the object feature set, and obtain a target media feature of the target media from the media feature set;

[0010] determine a transaction matching degree between the target object and the target media based on the target object feature and the target media feature;

[0011] determine, based on the transaction matching degree, recommended media corresponding to the target object from each target media.

[0012] The application further provides a media recommendation device, comprising:

[0013] an interaction feature set obtaining module, configured to obtain an interaction feature set between an object set and a media set; the interaction feature set is a set generated based on historical media transaction records of objects in the object set, and the historical media transaction records are object transaction records of recommended items corresponding to historical interaction media by the objects;

[0014] an interaction feature set decomposition module, configured to decompose the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set;

[0015] a target feature determining module, configured to determine a target object from the object set, obtain media that the target object has not interacted with from the media set as target media, obtain a target object feature of the target object from the object feature set, and obtain a target media feature of the target media from the media feature set;

[0016] a transaction matching degree determining module, configured to determine a transaction matching degree between the target object and the target media based on the target object feature and the target media feature;

[0017] a recommended media determining module, configured to determine, based on the transaction matching degree, recommended media corresponding to the target object from each target media.

[0018] The application further provides a computer device, comprising a memory and a processor, wherein the memory stores computer readable instructions, and the processor implements the steps of the above media recommendation method when executing the computer readable instructions.

[0019] The application further provides a computer readable storage medium, which stores computer readable instructions, and the computer readable instructions implement the steps of the above media recommendation method when executed by a processor.

[0020] The application further provides a computer program product comprising computer readable instructions which, when executed by a processor, implement the steps of the media recommendation method.

[0021] Details of one or more embodiments of the application are set forth in the accompanying drawings and description below. Other features, objects, and advantages of the application will be apparent from the description, drawings, and claims. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0023] Fig. 1 is a diagram of the application environment of the media recommendation method in one embodiment;

[0024] Fig. 2 is a flowchart of the media recommendation method in one embodiment;

[0025] Fig. 3 is a diagram of the decomposition of the interaction matrix in one embodiment;

[0026] Fig. 4 is a flowchart of the media recommendation method in another embodiment;

[0027] Fig. 5 is a diagram of the data processing of the target matching model in one embodiment;

[0028] Fig. 6 is a flowchart of the training of the matching model in one embodiment;

[0029] Fig. 7 is a flowchart of the media recommendation method in another embodiment;

[0030] Fig. 8 is a flowchart of the media recommendation method in another embodiment;

[0031] Fig. 9 is a flowchart of the media recommendation method in another embodiment;

[0032] Fig. 10 is a diagram of the interface of the short video platform in one embodiment;

[0033] Fig. 11 is a structural block diagram of the media recommendation device in one embodiment;

[0034] Fig. 12 is an internal structure diagram of the computer device in one embodiment;

[0035] Fig. 13 is an internal structure diagram of the computer device in another embodiment. DETAILED DESCRIPTION

[0036] In order to make the purposes, technical solutions and advantages of the present application clearer, further detailed description will be given below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0037] The media recommendation method provided by the embodiments of the present application can be applied in the application environment as shown in FIG. 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required to be processed by the server 104. The data storage system can be separately arranged, can be integrated on the server 104, or can be placed on the cloud or other devices. Among them, the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 can be realized by an independent server or a server cluster composed of multiple servers or a cloud server.

[0038] The terminal and the server can be used alone to execute the media recommendation method provided in the embodiments of the present application. The terminal and the server can also be used cooperatively to execute the media recommendation method provided in the embodiments of the present application.

[0039] For example, the server obtains an interaction feature set between an object set and a media set, and decomposes the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set. The interaction feature set is generated based on the historical media transaction records of the objects in the object set, and the historical media transaction records are the transaction records of the recommended items corresponding to the historical interaction media. The server determines a target object from the object set, obtains media that the target object has not interacted with as a target media from the media set, obtains target object features of the target object from the object feature set, and obtains target media features of the target media from the media feature set. The server determines a transaction matching degree between the target object and the target media based on the target object features and the target media features, and determines the recommended media corresponding to the target object from each target media based on the transaction matching degree.

[0040] In one embodiment, as shown in FIG. 2, a media recommendation method is provided, which is taken as an example applied to a computer device, which can be a terminal or a server. It can be understood that the method can be executed by the terminal or the server alone, or can be realized through the interaction between the terminal and the server. Among them:

[0041] In step S202, an interaction feature set between the object set and the media set is obtained. The interaction feature set is a set generated based on historical media transaction records of the objects in the object set. The historical media transaction record is an item transaction record of a recommended item corresponding to a historical interaction media by the object.

[0042] The object set includes a plurality of objects. The object refers to a user. The media set includes a plurality of media. The media refers to content presented through at least one media form (such as text, picture, video, audio, etc.). For example, the media can be a video, a picture-text information, an audio, an article, etc.

[0043] The interaction feature set between the object set and the media set includes interaction features between the plurality of objects in the object set and the plurality of media in the media set. The interaction feature set is a set generated based on historical media transaction records of the objects in the object set. The historical media transaction record of the object refers to an item transaction record of a recommended item corresponding to a historical interaction media by the object. The historical interaction media of the object refers to a media interacted by the user. For example, if the media is a video, the historical interaction media of the object can be a video watched by the user, a video liked by the user, a video commented by the user, a video shared by the user, etc. In the process of browsing the media, the user can be recommended an item. The recommended item corresponding to the media refers to an item recommended through the media. The recommended item corresponding to the media can be an item contained in the media itself. For example, if the media is an item advertisement, the item corresponding to the media is the item advertised by the media. The recommended item corresponding to the media can be an item associated with the media and displayed. For example, the recommended item corresponding to the media can be an item with a corresponding item transaction entry displayed in a media display interface. The item transaction entry is an entry for triggering a transaction of the item. For example, the item transaction entry can be an item purchase control. If the media is a video, an item purchase control is displayed in a video display interface, and the user can purchase the corresponding item through the item purchase control. The item transaction record of the object refers to a record generated by the object in a transaction of the item. For example, the item transaction record of the object refers to a consumption record of the object in the transaction of the item. The item transaction record is used to record transaction information of the transaction of the item. For example, the item transaction record includes data such as transaction time, transaction amount, transaction channel, transaction platform, etc. The historical interaction media belongs to the media set. The historical interaction media of the object is a media interacted by the object. The historical interaction media has a corresponding recommended item. The historical media transaction record of the object is an item transaction record of the object in the transaction of the corresponding recommended item (the recommended item corresponding to the historical interaction media). For example, if the user A watches the video B, the recommended item corresponding to the video B is the item C, and the historical media transaction record of the user A includes an item transaction record of the user in the transaction of the item C.

[0044] It can be understood that the item can be a physical item, for example, clothing, food, etc., and the item can also be a virtual item, for example, a service, a virtual prop, etc. The virtual prop can be a social prop (such as an expression, a background, a bubble, etc.), a game prop (such as an instrument, equipment, a pet, etc.), etc.

[0045] For any object in the object set, the media set can include media that the object has interacted with and media that the object has not interacted with. For the media that the object has interacted with, the interaction feature between the object and the media is generated based on the relevant transaction record of the item recommended by the object for the media that has been interacted with. For example, if the media is a video, for video B watched by user A, the interaction feature between user A and video B is generated based on the item transaction record of the recommended item corresponding to video B by user A. For the media that the object has not interacted with, the interaction feature between the object and the media is unknown.

[0046] Further, the interaction feature between the object and the media can include at least one dimension of feature data, and different feature data can represent different transaction features. For example, if the media is a video, for video B watched by user A, the interaction feature between user A and video B can be the number of transactions, the total transaction amount, the transaction score determined based on the transaction amount (at least one of the single transaction amount and the total transaction amount), etc.

[0047] Specifically, the computer device can obtain the interaction feature set between the object set and the media set locally or from other devices, select media that can be recommended to the object from the media set based on the interaction feature set for the object in the object set, and recommend the media to the object.

[0048] In one of the embodiments, the historical media transaction record of the object can be the item transaction record of the recommended item corresponding to the historical interaction media of the object on at least one transaction platform.

[0049] The historical media transaction record of the object can be an item transaction record of the object for a recommended item corresponding to the historical interaction media on each transaction platform. For example, if the media is music, the user plays music in the music application, and the user plays the music with a corresponding recommended item, the historical media transaction record of the user can be the item transaction record of the user for the recommended item on each transaction platform. If the user has not traded the recommended item in the music application, but the user has traded the recommended item in other applications, the historical media transaction record of the user can be the item transaction record of the user for the recommended item in other applications. If the user has traded the recommended item in the music application, and the user has also traded the recommended item in other applications, the historical media transaction record of the user can be the item transaction record of the user for the recommended item in the music application and other applications. In this way, by aggregating the transaction records of each transaction platform to generate the interaction feature between the object and the media, the content richness of the interaction feature can be improved, and the calculation accuracy of the subsequent transaction matching degree can be improved, and the media recommendation effectiveness, recommendation accuracy, and media conversion rate can be improved.

[0050] Of course, the historical media transaction record of the object can be an item transaction record for a recommended item corresponding to the media during the interaction of the object and the media. That is, the historical media transaction record is an item transaction record generated in a single transaction platform due to the interaction behavior of the object for the media. For example, if the media is music, the user plays music in the music application, and the historical media transaction record of the user can be a related transaction record of the user for the recommended item of the played music in the music application during the user playing the music. That is, the media set can include media that the object has interacted with and media that the object has not interacted with, and the media that the object has interacted with can be media that the object has interacted with and traded. In this way, according to the item transaction record generated by the user in a single platform due to the interaction with the media, the recommended media is determined from the media that the user has not interacted with in the platform through data analysis, and the recommended media is recommended to the user in the platform, so that the consumption habits, consumption tendencies, and consumption willingness of the user in the single platform are analyzed to recommend media displayed in the platform to the user, which can effectively improve the recommendation effectiveness, recommendation accuracy, and media conversion rate in the platform.

[0051] Step S204, decomposing the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set.

[0052] The object feature set corresponding to the object set includes object features corresponding to each object in the object set respectively. The object feature is a latent feature representation of the object. The interaction feature set is generated based on the historical media transaction records of the objects in the object set, and therefore, the object feature obtained by decomposing the interaction feature set can reflect the transaction characteristics of the object. For example, the object feature can reflect the transaction preferences of the user. The media feature set corresponding to the media set includes media features corresponding to each media in the media set respectively. The media feature is a latent feature representation of the media. The media feature can reflect the intrinsic characteristics of the media.

[0053] Specifically, after obtaining the interaction feature set between the object set and the media set, the computer device can perform feature decomposition on the interaction feature set to obtain the object feature set corresponding to the object set and the media feature set corresponding to the media set.

[0054] It can be understood that the interaction features between part of the objects in the object set and part of the media in the media set are unknown, but the object features of each object in the object set and the media features of each media in the media set can be obtained through feature decomposition with reference to the interaction features between the reference object and other media.

[0055] In one of the embodiments, the interaction feature set is an interaction matrix, part of the data in the interaction matrix is known, and part of the data is unknown. Matrix decomposition is performed on the interaction matrix to obtain two feature matrices, one of which represents the latent features of each user, and the other of which represents the latent features of each media.

[0056] In step S206, a target object is determined from the object set, media that the target object has not interacted with is obtained from the media set as target media, a target object feature of the target object is obtained from the object feature set, and a target media feature of the target media is obtained from the media feature set.

[0057] The target object can be any object in the object set. The target media is media in the media set that the target object has not interacted with. For example, if the media is a video, the target media can be a video that the target user has not watched; if the media is music, the target media can be music that the target user has not played; and if the media is an article, the target media can be an article that the target user has not browsed.

[0058] Specifically, the computer device can select any object from the object set as the target object, and obtain the object feature of the target object from the object feature set corresponding to the object set as the target object feature. The computer device can obtain media that the target object has not interacted with from the media set as the target media, and obtain the media feature of the target media from the media feature set corresponding to the media set as the target media feature.

[0059] It can be understood that there can be at least one target media for the same target object.

[0060] In step S208, a transaction matching degree between the target object and the target media is determined based on the target object feature and the target media feature.

[0061] Specifically, the target object feature can reflect a transaction characteristic of the target object, the target media feature can reflect a potential characteristic of the target media, and the computer device can predict a transaction willingness of the target object for the target media based on the target object feature and the target media feature, to obtain the transaction matching degree between the target object and the target media. The higher the transaction matching degree between the target object and the target media, the higher the transaction willingness of the target object for the recommended item corresponding to the target media.

[0062] In one of the embodiments, a feature similarity between the target object feature and the target media feature can be calculated as the transaction matching degree between the target object and the target media. For example, the target object feature and the target media feature are feature vectors, and a cosine similarity between the target object feature and the target media feature can be calculated as the transaction matching degree between the target object and the target media.

[0063] In one embodiment, the target object feature and the target media feature can be input into a target matching model, and the target matching model outputs the transaction matching degree between the target object and the target media. The target matching model is a trained matching model. The target matching model can be obtained by supervised training based on training samples and training labels corresponding to the training samples. The training samples include training object features corresponding to training objects and training media features corresponding to training media, and the training labels are divided into positive labels and negative labels. The positive label indicates that the training object has traded a recommended item corresponding to the training media for an interactive training media, and the negative label indicates that the training object has not traded the recommended item corresponding to the training media for the interactive training media. It can be understood that the training object features corresponding to the training objects can be obtained from the object feature set, and the training media features corresponding to the training media can be obtained from the media feature set.

[0064] In step S210, the recommended media corresponding to the target object is determined from the target media based on the transaction matching degree.

[0065] The recommended media corresponding to the target object is the target media that needs to be recommended to the target object.

[0066] Specifically, after determining the transaction matching degrees between the target object and the target media, the computer device can determine the recommended media corresponding to the target object from the target media based on the transaction matching degrees. For example, the target media with a transaction matching degree greater than a matching degree threshold is selected as the recommended media; the target media with the greatest transaction matching degree is selected as the recommended media; the target media are sorted in descending order of the transaction matching degrees, and the recommended order of the target media is determined according to the sorting result, and the target media with a higher ranking are preferentially recommended to the user; and the like.

[0067] In one of the embodiments, the recommendation in the method of the present application can be active recommendation. When the target user enters the target platform, the media not interacted with by the target object is obtained from the media set of the target platform as the target media, the transaction matching degrees between the target media and the target user are respectively determined, and the recommended media corresponding to the target user is determined from the target media based on the transaction matching degrees. When the target user enters the target platform, the recommended media is actively displayed to the target user.

[0068] The recommendation in the method of the present application can also be passive recommendation (also referred to as search). The search information (such as search words, search sentences, search images, etc.) input by the target user in the target platform is obtained, and the media matching the search information is obtained from the media set as candidate media, that is, the preliminary search result is determined. For example, the media similar in semantics to the search information is obtained as the candidate media. Further, the media not interacted with by the target user is obtained from the candidate media as the target media, the transaction matching degrees between the target media and the target user are respectively determined, and the recommended media corresponding to the target user is determined from the target media based on the transaction matching degrees, that is, the preliminary search result is screened to obtain the final search result. The recommended media are sorted in descending order of the transaction matching degrees to obtain a recommended media sequence. The target platform displays the recommended media sequence to the target user as the final search result.

[0069] It can be understood that the target platform can be any platform for displaying content. For example, the target platform can be a short video platform, a live broadcast platform, a social platform, etc. Taking the short video platform as an example, the media in the short video platform is a short video.

[0070] In one of the embodiments, a recommended item matched with the recommended media is obtained, the recommended media and the recommended item are sent to a target terminal corresponding to the target object, so that the target terminal displays the recommended media and the recommended item in association. The recommended item matched with the recommended media can be at least one of an item related to the content of the recommended media or an item complementary to the recommended media. The item related to the content of the recommended media can be an item displayed, introduced or used in the recommended media. The item complementary to the recommended media can be an item capable of being used in cooperation with the item related to the content of the recommended media. In this way, the recommended media is a media with a relatively high transaction matching degree, and the display of the recommended media and the recommended item in association can further attract the user to trigger an interactive operation on the recommended media, improve the media conversion rate of the recommended media, further attract the user to trigger an interactive operation on the recommended item, and improve the item conversion rate of the recommended item.

[0071] It can be understood that the interaction feature set can be updated as needed, for example, the interaction feature set can be updated at a regular time. The latest object feature set and the media feature set are obtained by decomposing the latest interaction feature set. When it is necessary to determine the transaction matching degree, the latest object feature set and the media feature set are obtained to determine the transaction matching degree.

[0072] In the above media recommendation method, the transaction matching degree obtained by comprehensively analyzing the respective features of the target object and the target media can more accurately reflect the matching degree between the target object and the target media, and the determination of whether to recommend the target media to the target object based on the transaction matching degree can effectively improve the effectiveness and accuracy of media recommendation, and achieve more accurate and effective personalized recommendation. Further, the interaction feature set can reflect the interaction and transaction relationship between the user set and the media set, the object features in the object feature set obtained by decomposing the interaction feature set can reflect the consumption preferences of the user, and the media features in the media feature set obtained by decomposing the interaction feature set can reflect the characteristics of the media. Based on the object features of the target user and the media features of the target media, the consumption willingness of the user to the uninteracted target media is predicted, the transaction matching degree is obtained, and the media is recommended based on the transaction matching degree, which can improve the consumption interest of the target user to the recommended media, and further improve the effectiveness and accuracy of media recommendation. It can be understood that improving the recommendation effectiveness and the recommendation accuracy can reduce the waste of physical resources required for implementing the recommendation, and improve the utilization rate of the physical resources. The physical resources herein include server resources, network bandwidth resources, etc.

[0073] In one of the embodiments, the interaction feature set between the object set and the media set is obtained, including:

[0074] obtain a historical media transaction record of each object in the object set; determine an item transaction statistical value of the object for a corresponding historical interaction medium based on the historical media transaction record of the object; generate an interaction matrix based on the item transaction statistical value of each object for the respective historical interaction medium; a first dimension of the interaction matrix represents the object in the object set, and a second dimension of the interaction matrix represents the medium in the medium set; and the interaction matrix is taken as an interaction feature set between the object set and the medium set.

[0075] The item transaction statistical value of the object for the historical interaction medium is a transaction statistical value of a recommended item corresponding to the historical interaction medium by the object. The historical media transaction record of the object includes an item transaction record of a recommended item corresponding to at least one historical interaction medium by the object. Each item transaction record of the recommended item corresponding to the same historical interaction medium in the historical media transaction record is counted to obtain the transaction statistical value. For example, the transaction statistical value can be a transaction frequency, a total transaction amount, an average transaction amount, a transaction frequency, a transaction score, etc. of the recommended item corresponding to the historical interaction medium. The transaction score is an index for quantitatively evaluating the item transaction record. The transaction score can be a score calculated based on the transaction frequency, the transaction amount, the transaction item quantity, etc.

[0076] Specifically, the computer device can obtain a historical media transaction record of each object in the object set, and generate an interaction matrix based on the historical media transaction record of each object, and take the interaction matrix as an interaction feature set between the object set and the medium set.

[0077] For the interaction matrix, the item transaction statistical value of the object for the corresponding historical interaction medium is determined based on the historical media transaction record of the object, for example, the transaction scores of the object for multiple historical interaction media are calculated based on the historical media transaction record of the object, and the transaction scores are taken as the item transaction statistical values to obtain the item transaction statistical values of the object for the multiple historical interaction media. Then, the interaction matrix is generated based on the item transaction statistical value of each object for the respective historical interaction medium. The first dimension of the interaction matrix represents the object in the object set, the second dimension of the interaction matrix represents the medium in the medium set, and the matrix value positioned by the object and the medium in the interaction matrix represents the item transaction statistical value of the object for the medium.

[0078] It can be understood that if the object has not interacted with a certain medium, the item transaction statistical value of the object for the medium is unknown data, and the matrix value positioned by the object and the medium in the interaction matrix is unknown data.

[0079] In the above embodiments, based on the historical media transaction records of the objects, the object transaction statistical values of the objects for the corresponding historical interaction media are determined, and the object transaction statistical values are quantifications of the historical media transaction records and are convenient for calculation. Based on the object transaction statistical values of the respective objects for the respective historical interaction media, an interaction matrix is generated, the interaction matrix is taken as an interaction feature set between an object set and a media set, and thus the object feature set and the media feature set can be quickly obtained by performing matrix decomposition on the interaction matrix.

[0080] In a specific application, the interaction matrix is illustrated by taking a video as an example of the media. Referring to FIG. 3, the interaction matrix is an m-row-n-column matrix, the rows of the interaction matrix represent users, and there are m users in total, and the columns of the interaction matrix represent videos, and there are n videos in total. The matrix values in the interaction matrix can be payment scores determined according to payment amounts of recommended objects of the users for the videos, and the higher the payment amount, the higher the payment score. The matrix values in the interaction matrix can be determined according to payment histories of the users. If a user has not watched a video, the corresponding matrix value is unknown data. By performing matrix decomposition on the interaction matrix by using a matrix decomposition algorithm (for example, a singular value decomposition algorithm, an alternating least squares method, etc.), the interaction matrix can be decomposed into two matrices, namely an object feature matrix (namely, an object feature set) and a video feature matrix (namely, a media feature set). One matrix represents the latent features of the users, and the other matrix represents the latent features of the videos, and these latent features can capture the characteristics of the videos preferred by the users. The object feature matrix is an m-row-k-column matrix, the rows of the object feature matrix represent the users, and one row of data represents the latent features of one user. The video feature matrix is a k-row-n-column matrix, the columns of the video feature matrix represent the videos, and one column of data represents the latent features of one video.

[0081] In one embodiment, based on the object transaction statistical values of the respective objects for the respective historical interaction media, the interaction matrix is generated, including:

[0082] The target protection parameter for the object transaction statistical values is acquired as the transaction protection parameter; the target transaction statistical value is determined from the object transaction statistical values of the respective objects for the respective historical interaction media; the noise data is added to the target transaction statistical value based on the transaction protection parameter, to obtain an encrypted transaction statistical value; and the interaction matrix is generated based on the encrypted transaction statistical value and the remaining object transaction statistical values.

[0083] The transaction protection parameter is the target protection parameter for the object transaction statistical values. The target protection parameter is used to add noise to the data to protect the security of the data and avoid data leakage. It can be understood that the target protection parameter can be a preset protection parameter or a protection parameter obtained by adjusting the preset protection parameter. In addition, the target protection parameter can be the same or different for different types of data.

[0084] The target transaction statistical value refers to a transaction statistical value of an object to be protected. The target transaction statistical value can be selected as needed. For example, each transaction statistical value of an object can be selected as the target transaction statistical value; a part of the transaction statistical values of the objects can be randomly selected as the target transaction statistical value; a transaction statistical value corresponding to an object whose activity is greater than an activity threshold can be selected as the target transaction statistical value; and the like. The activity of an object refers to the activity of the object in the application. For example, if the media is a video in a video application, the activity of the object refers to the activity of the object in the video application, and the activity of the object can be determined according to the login times, use time, transaction amount, and the like of the object in the video application. In this way, the object whose activity is greater than the activity threshold is an important object, and the transaction statistical value of the important object corresponding to the historical interaction media is selected as the target transaction statistical value, so that the data security of the important object can be protected, data leakage can be avoided, and the important object can be prevented from being lost due to data leakage.

[0085] The encrypted transaction statistical value is obtained by adding noise data to the target transaction statistical value according to the transaction protection parameter. It can be understood that the target transaction statistical value is sensitive data, and an external person cannot obtain the original target transaction statistical value according to the encrypted transaction statistical value, so that the data security of the target transaction statistical value is effectively protected.

[0086] Specifically, in order to improve the data security, the computer device can add noise data to the transaction statistical value of the object to hide the real transaction statistical value of the object when generating the interaction matrix, and generate the interaction matrix according to the transaction statistical value of the object to which the noise data is added.

[0087] First, the target protection parameter for the transaction statistical value of the object is obtained as the transaction protection parameter, the target transaction statistical value is determined from the transaction statistical value of each object corresponding to each historical interaction media of the object, and noise data is added to the target transaction statistical value based on the transaction protection parameter to obtain the encrypted transaction statistical value. For example, a part of the transaction statistical values of the objects is selected as the target transaction statistical value, the interaction protection parameter includes a value range of the noise data, one noise data is selected from the value range, and the sum of the target transaction statistical value and the noise data is taken as the encrypted transaction statistical value. Different noise data can be selected for different target transaction statistical values. Further, the interaction matrix is generated based on the encrypted transaction statistical value and the remaining transaction statistical values of the objects.

[0088] In the above embodiment, when generating the interaction matrix, noise data is added to the transaction statistical value of the object based on the transaction protection parameter, so that the data security of the transaction statistical value of the object can be protected, and data leakage of the transaction statistical value of the object can be avoided.

[0089] In an embodiment, as shown in FIG. 4, based on the target object feature and the target media feature, the transaction matching degree between the target object and the target media is determined, including:

[0090] In step S402, the object attribute feature of the target object is obtained, the target object feature and the object attribute feature are combined, and the comprehensive object feature of the target object is obtained.

[0091] In step S404, the media attribute feature of the target media is obtained, the target media feature and the media attribute feature are combined, and the comprehensive media feature of the target media is obtained.

[0092] In step S406, based on the comprehensive object feature and the comprehensive media feature, the transaction matching degree between the target object and the target media is determined.

[0093] The object attribute feature is a feature representation of the object attribute. The object attribute refers to the attribute of the user, and the object attribute is a description or summary of the personal information of the user. For example, the object attribute can be an object portrait, and the object portrait is a summary and description of the features such as personal information, interests, and behavior habits of the user. The media attribute feature is a feature representation of the media attribute. The media attribute is a description or summary of the media content. For example, the media attribute can be a media title, a media introduction, a media tag, and the like.

[0094] Specifically, in order to further improve the accuracy of the transaction matching degree, the computer device can combine the latent features obtained through feature decomposition with other available metadata to better capture the relationship between the user interest and the media.

[0095] The computer device can obtain the object attribute feature of the target object, combine the target object feature and the object attribute feature, and obtain the comprehensive object feature of the target object. For example, the target object feature and the object attribute feature can be spliced to obtain the comprehensive object feature. The computer device can obtain the media attribute feature of the target media, combine the target media feature and the media attribute feature, and obtain the comprehensive media feature of the target media. For example, the target media feature and the media attribute feature can be spliced to obtain the comprehensive media feature. Further, the computer device can predict the transaction intention of the target object for the target media based on the comprehensive object feature and the comprehensive media feature, and obtain the transaction matching degree between the target object and the target media. For example, the feature similarity between the comprehensive object feature and the comprehensive media feature is calculated to obtain the transaction matching degree between the target object and the target media.

[0096] In one of the embodiments, the integrated object feature and the integrated media feature are input into a target matching model, and the target matching model outputs a transaction matching degree between the target object and the target media. The target matching model is a trained matching model. The target matching model can be obtained through supervised training based on training samples and training labels corresponding to the training samples. The training samples include training object features corresponding to training objects and training media features corresponding to training media. The training labels are divided into positive labels and negative labels. The positive label indicates that the training object has traded a recommended item corresponding to the training media through the training media that has been interacted with. The negative label indicates that the training object has not traded the recommended item corresponding to the training media through the training media that has been interacted with. It can be understood that the training object features corresponding to the training objects can be obtained from the object feature set, and the training media features corresponding to the training media can be obtained from the media feature set. The training object features corresponding to the training objects can also be the integrated object features corresponding to the training objects, and the training media features corresponding to the training media can also be the integrated media features corresponding to the training media.

[0097] In one of the embodiments, referring to FIG. 5, in the target matching model, the integrated object feature and the integrated media feature are fused to obtain a first matching feature, and the integrated object feature and the integrated media feature are spliced to obtain an integrated spliced feature. The integrated spliced feature is extracted to obtain a second matching feature, the first matching feature and the second matching feature are spliced to obtain a target matching feature, and the transaction matching degree between the target object and the target media is determined based on the target matching feature. The target matching model outputs the transaction matching degree between the target object and the target media. Specifically, the integrated object feature and the integrated media feature can be fused by multiplying corresponding elements of the integrated object feature and the integrated media feature, adding, and the like. The integrated spliced feature can be extracted through network layers of the model, such as convolution layers, fully connected layers, and the like. In this way, the target matching model fully interacts the integrated object feature and the integrated media feature through multiple ways to determine the transaction matching degree, which can improve the accuracy of the transaction matching degree.

[0098] In the above embodiments, the target object feature and the object attribute feature are combined to obtain the integrated object feature of the target object, and the integrated object feature has more abundant information. The target media feature and the media attribute feature are combined to obtain the integrated media feature of the target media, and the integrated media feature has more abundant information. The transaction matching degree between the target object and the target media is determined based on the integrated object feature and the integrated media feature, which can effectively improve the accuracy of the transaction matching degree, and further improve the accuracy of subsequent media recommendation and the media conversion rate.

[0099] In one of the embodiments, the object attribute feature of the target object is obtained, including:

[0100] Obtain a plurality of attribute data of a target object; determine an attribute protection parameter corresponding to target attribute data in the plurality of attribute data based on a data type of the target attribute data, add noise data to the target attribute data based on the attribute protection parameter to obtain encrypted attribute data; and extract features from the encrypted attribute data and other attribute data in the plurality of attribute data except the target attribute data to obtain object attribute features of the target object.

[0101] The target object has a plurality of corresponding attribute data. The target attribute data refers to the attribute data to be protected. It can be understood that if inappropriate noise data is added to the data, it will affect the accuracy of subsequent data processing. Therefore, for different types of attribute data, corresponding target protection parameters can be set to balance data security and data accuracy.

[0102] The attribute protection parameter corresponding to the target attribute data is a target protection parameter set for the data type of the target attribute data. The encrypted attribute data is obtained by adding noise data to the target attribute data according to the attribute protection parameter. It can be understood that the target attribute data is sensitive data, and it is difficult for external personnel to obtain the original target attribute data from the encrypted attribute data, thereby effectively protecting the data security of the target attribute data.

[0103] Specifically, in addition to data protection when generating the interaction matrix, the computer device can also perform data protection when generating the object attribute features to further protect the personal information of the user.

[0104] In order to generate the object attribute feature of the target object, the computer device can obtain a plurality of attribute data of the target object, determine target attribute data from the plurality of attribute data, for example, randomly obtain attribute data from the plurality of attribute data as the target attribute data; obtain attribute data with a sensitivity greater than a sensitivity threshold (i.e., high-sensitivity attribute data) from the plurality of attribute data as the target attribute data; and the like. The sensitivity corresponding to each attribute data can be set as needed. Further, the computer device determines the attribute protection parameter corresponding to the target attribute data based on the data type of the target attribute data in the plurality of attribute data, that is, obtains the target protection parameter for the data type of the target attribute data as the attribute protection parameter corresponding to the target attribute data. The computer device adds noise data to the target attribute data based on the attribute protection parameter to obtain encrypted attribute data, for example, the attribute protection parameter includes a value range of the noise data, a noise data is selected from the value range, and the sum of the target attribute data and the noise data is taken as the encrypted attribute data. Finally, the computer device extracts features from the encrypted attribute data and other attribute data in the plurality of attribute data except the target attribute data to obtain the object attribute feature of the target object. For example, the attribute data is subjected to feature engineering to obtain the object attribute feature; the feature is extracted through a machine learning model to obtain the object attribute feature; and the like.

[0105] In the above embodiments, when generating the object attribute feature, the noise data is added to the target attribute data based on the attribute protection parameter, which can protect the data security of the target attribute data and avoid data leakage of the target attribute data.

[0106] In one embodiment, based on the attribute protection parameter, the noise data is added to the target attribute data to obtain encrypted attribute data, including:

[0107] The noise distribution is determined based on the noise generation mode and the noise intensity in the attribute protection parameter, and the noise data is sampled from the noise distribution based on the target attribute data; the sampled noise data is added to the target attribute data to obtain the encrypted attribute data.

[0108] The target protection parameter is used to add noise to the data to protect the security of the data and avoid data leakage. It can be understood that the target protection parameter can be a preset protection parameter, or a protection parameter obtained by adjusting the preset protection parameter.

[0109] The target protection parameter includes a noise generation mode and a noise intensity. The noise generation mode refers to a manner of generating noise, for example, the noise generation mode can be a noise generation algorithm adopted. The noise intensity is used to control the intensity and size of the noise. The noise distribution can be determined based on the noise generation mode and the noise intensity in the target protection parameter. The noise distribution refers to a distribution of noise data, for example, the noise distribution can be a noise function that inputs data to which noise is to be added and outputs noise data. The noise generation mode is used to determine statistical characteristics of the noise distribution, for example, the noise obeys a Laplace distribution, a Gaussian distribution, an exponential distribution, and the like. The noise intensity is used to determine a value range of the noise data.

[0110] It can be understood that for different types of attribute data, corresponding target protection parameters can be respectively set. Different target protection parameters can be different noise generation modes or different noise intensities.

[0111] Specifically, after the target attribute data is determined, the computer device can determine the data type of the target attribute data, obtain the target protection parameter corresponding to the data type as the attribute protection parameter corresponding to the target attribute data. Further, the computer device determines the noise distribution based on the noise generation mode and the noise intensity in the attribute protection parameter, and samples noise data from the noise distribution based on the target attribute data. For example, the noise distribution is a noise function generated based on the noise generation mode and the noise intensity, and the noise generation mode and the noise intensity are components of the noise function. The target attribute data is substituted into the noise function to obtain the noise data. Finally, the computer device adds the sampled noise data to the target attribute data to obtain the encrypted attribute data.

[0112] In the above embodiments, the noise distribution is determined based on the noise generation mode and the noise intensity in the attribute protection parameter, and the noise data is sampled from the noise distribution based on the target attribute data, so that different target attribute data has respective noise data. Adding the sampled noise data to the target attribute data can realize individualized data protection for the target attribute data and improve the data security of the target attribute data.

[0113] It can be understood that the specific process of adding noise data to the target transaction statistical value based on the transaction protection parameter to obtain the encrypted transaction statistical value can refer to the specific process of adding noise data to the target attribute data based on the attribute protection parameter to obtain the encrypted attribute data, which will not be described herein.

[0114] In an embodiment, the noise generation mode can be a differential privacy algorithm, and the noise strength is at least one of a privacy parameter ε, a privacy parameter δ. The differential privacy algorithm can be a Laplace mechanism, an exponential mechanism, an inexact counting, etc. The privacy parameter ε is used to quantify the strength of privacy protection, and generally a smaller ε means providing a higher level of privacy protection, but may sacrifice some data utility. The privacy parameter δ is used to describe the confidence of privacy protection. A δ close to 0 means that the algorithm provides almost complete differential privacy protection, but may introduce more noise.

[0115] In an embodiment, based on the integrated object features and the integrated media features, a transaction matching degree between the target object and the target media is determined, including:

[0116] The integrated object features and the integrated media features are input into a target matching model, and the target matching model outputs a transaction matching degree between the target object and the target media; the target matching model is obtained by training an initial matching model based on encrypted training samples and corresponding training labels, and the encrypted training samples are obtained by adding noise data to original training samples based on a target protection parameter.

[0117] The matching model is an artificial intelligence model for matching objects and media. The input data of the matching model includes integrated object features of the objects and integrated media features of the media, and the output data of the matching model includes a transaction matching degree between the objects and the media. The initial matching model is a matching model to be trained. The target matching model is a trained matching model.

[0118] The integrated object features corresponding to the target object and the integrated media features corresponding to the target media can be determined based on related data containing encrypted data. For example, the target object features in the integrated object features and the target media features in the integrated media features can be determined based on an interaction feature set containing encrypted data. For another example, the object attribute features in the integrated object features can be determined based on attribute data containing encrypted data. When the model is applied, the input data of the model can be generated based on encrypted data. In order to ensure the accuracy of the model, when the model is trained, the initial matching model is supervised trained based on the encrypted training samples and the training labels corresponding to the encrypted training samples, to obtain the target matching model. The encrypted training samples are input into the initial matching model to obtain the predicted labels corresponding to the encrypted training samples. Based on the difference between the training labels corresponding to the encrypted training samples and the predicted labels, the model parameters of the initial matching model are adjusted until the convergence condition is met, to obtain the target matching model.

[0119] The training sample includes a comprehensive object feature corresponding to a training object and a comprehensive media feature corresponding to a training media. A training label corresponding to the training sample is used to indicate an accurate transaction matching degree between the training object and the training media. For example, the training label is divided into a positive label and a negative label. The training media is a media that has been interacted with by the training object. The positive label indicates that the training object has transacted a recommended item corresponding to the training media, and the negative label indicates that the training object has not transacted the recommended item corresponding to the training media. The transaction matching degree corresponding to the positive label is greater than the transaction matching degree corresponding to the negative label.

[0120] The encrypted training sample is obtained by adding noise data to the original training sample. It can be understood that the original training sample is an original training sample, and the comprehensive object feature and the comprehensive media feature in the original training sample do not include encrypted data. At least one of the comprehensive object feature and the comprehensive media feature in the encrypted training sample is determined based on related data containing encrypted data. The encrypted training sample is obtained by adding noise data to the original training sample based on a target protection parameter. The target protection parameter is used to add noise to the data to protect the security of the data and avoid data leakage. The target protection parameter includes at least one of a transaction protection parameter and an attribute containing parameter.

[0121] Specifically, the computer device can determine the matching degree between the target object and the target media by means of the feature analysis capability of the artificial intelligence model. The computer device obtains a target matching model, inputs the comprehensive object feature and the comprehensive media feature into the target matching model, performs data processing on the input data through the target matching model, and outputs the transaction matching degree between the target object and the target media by the target matching model.

[0122] In the above embodiment, the comprehensive object feature and the comprehensive media feature are input into the target matching model, and the target matching model outputs the transaction matching degree between the target object and the target media. The target matching model is obtained by training and has strong prediction capability, and can output accurate transaction matching degree. In addition, the target matching model is obtained by training the initial matching model based on the encrypted training sample and the corresponding training label, and can process encrypted input data and output accurate transaction matching degree.

[0123] In one embodiment, as shown in FIG. 6, the media recommendation method further includes:

[0124] In step S602, a first original training sample and a training label corresponding to the first original training sample are obtained.

[0125] In step S604, based on the first original training sample and the training label corresponding to the first original training sample, an initial matching model is trained to obtain an intermediate matching model.

[0126] Specifically, the computer device can obtain a first original training sample and a training label corresponding to the first original training sample, perform supervised training on the initial matching model based on the first original training sample and the training label corresponding to the first original training sample, and obtain an intermediate matching model, so that the intermediate matching model has a certain degree of accuracy.

[0127] In step S606, a second original training sample and a training label corresponding to the second original training sample are obtained.

[0128] In step S608, noise data is added to the second original training sample based on the initial protection parameter to obtain an encrypted training sample.

[0129] The first original training sample and the second original training sample can be the same original training sample or different original training samples. It can be understood that there can be multiple first original training samples and multiple second original training samples. The initial protection parameter refers to an initial protection parameter. The initial protection parameter can be a preset protection parameter.

[0130] Specifically, the intermediate matching model has good performance in the transaction matching degree prediction task for the original training sample. In order to make the intermediate matching model also have good performance in the transaction matching degree prediction task for the encrypted training sample, the computer device can obtain a second original training sample and a training label corresponding to the second original training sample, add noise data to the second original training sample based on the initial protection parameter to obtain an encrypted training sample, and then further train the intermediate matching model based on the second original training sample, the encrypted training sample, and the training label corresponding to the second original training sample to obtain a target matching model. It can be understood that the training label corresponding to the second original training sample is also the training label corresponding to the encrypted training sample.

[0131] In step S610, the second original training sample and the encrypted training sample are input into the intermediate matching model to obtain prediction labels corresponding to the second original training sample and the encrypted training sample, respectively.

[0132] In step S612, based on the difference between the prediction labels corresponding to the second original training sample and the encrypted training sample, respectively, the initial protection parameter is adjusted, and based on the difference between the prediction label corresponding to the encrypted training sample and the training label corresponding to the second original training sample, the model parameters of the intermediate matching model are adjusted until a convergence condition is met, to obtain a target matching model and a target protection parameter.

[0133] Specifically, the computer device can input the second original training sample into the intermediate matching model, the intermediate matching model outputs the predicted label corresponding to the second original training sample, and input the encrypted training sample into the intermediate matching model, the intermediate matching model outputs the predicted label corresponding to the encrypted training sample. Although the initial protection parameter can protect the security of the data, the initial protection parameter will also affect the accuracy of data processing. In order to achieve the balance between data security and data accuracy, the initial protection parameter and the model parameter can be adjusted synchronously according to the related labels of the second original training sample and the encrypted training sample. The computer device adjusts the initial protection parameter based on the difference between the predicted labels corresponding to the second original training sample and the encrypted training sample, so that the predicted label corresponding to the encrypted training sample gradually approaches the predicted label corresponding to the second original training sample, so as to reduce the influence of the protection parameter on the model prediction result, and adjusts the model parameter of the intermediate matching model based on the difference between the predicted label corresponding to the encrypted training sample and the training label corresponding to the second original training sample, so that the predicted label corresponding to the encrypted training sample gradually approaches the training label corresponding to the encrypted training sample, so that the model can gradually have the processing ability for encrypted data. The computer device performs multiple rounds of iterative training, and through multiple rounds of protection parameter and model parameter adjustment, until the convergence condition is met, to obtain the target matching model and the target protection parameter.

[0134] It can be understood that multiple second original training samples can be obtained in each round. In a round of training, based on the difference between each second original training sample and the predicted label corresponding to the respective encrypted training sample of each second original training sample, a first loss is calculated, for example, the predicted label corresponding to the original training sample and the predicted label corresponding to the encrypted training sample of the original training sample form a group, the intra-group difference of each group of data is calculated, the sum of each group difference is calculated as the first loss, or the average of each group difference is calculated as the first loss, the initial protection parameter is adjusted based on the first loss to obtain the intermediate protection parameter, and the intermediate protection parameter is used as the initial protection parameter of the next round. In a round of training, based on the difference between the predicted label corresponding to each encrypted training sample and the corresponding training label, a second loss is calculated, for example, the predicted label corresponding to the encrypted training sample and the training label form a group, the intra-group difference of each group of data is calculated, the sum of each group difference is calculated as the second loss, or the average of each group difference is calculated as the second loss, the model parameter of the intermediate matching model is adjusted based on the second loss to obtain an updated matching model, and the updated matching model is used as the intermediate matching model of the next round.

[0135] The convergence condition refers to a condition for judging whether to end data adjustment. The convergence condition can be that the number of iterations is greater than a training round threshold, the first loss and the second loss are less than respective loss thresholds, the loss change rates of the first loss and the second loss are less than respective change rate thresholds, etc.

[0136] In the above embodiments, the initial matching model is trained based on the first original training sample and the training label corresponding to the first original training sample to obtain an intermediate matching model, so that the intermediate matching model has a certain data processing capability with accuracy. The second original training sample and the encrypted training sample are input into the intermediate matching model to obtain the prediction labels corresponding to the second original training sample and the encrypted training sample respectively, the initial protection parameter is adjusted based on the difference between the prediction labels corresponding to the second original training sample and the encrypted training sample, the model parameter of the intermediate matching model is adjusted based on the difference between the prediction label corresponding to the encrypted training sample and the training label corresponding to the second original training sample, until the convergence condition is met, to obtain the target matching model and the target protection parameter, so that the target matching model and the target protection parameter can realize effective balance of prediction accuracy and data security, and the accuracy of the model output data is ensured on the basis of ensuring the data security of the model input data.

[0137] In one specific application, the method of the present application is illustrated by taking video as an example. Referring to FIG. 7, first, data is collected, which can specifically include user payment history, viewing history, object portrait (such as user age, gender, geographic location, etc.), and video metadata (such as video tags, types, publishers, etc.). Then, an interaction matrix is constructed, which is constructed according to the user's payment history and viewing history. The rows of the interaction matrix represent users, and the columns represent videos. The values of the matrix can have multiple dimensions, such as the payment amount of recommended goods by the user to the video, the user's stay time on the video, the user's payment score for the video, etc. Further, the interaction matrix is decomposed into two low-dimensional matrices, one representing the user's latent features, and the other representing the video's latent features. These latent features capture the user's preferences and the video's characteristics. Further, the latent features generated by the matrix decomposition can be combined with other available metadata (such as object portraits, video metadata, etc.) to better capture the relationship between user interests and videos. By calculating the inner product of the user latent feature matrix and the video latent feature matrix, the payment willingness score of each user for all unwatched short videos is predicted. For any user, the predicted payment willingness scores are sorted, and the video with the highest payment willingness score is recommended to the user. In addition, other business strategies such as recommending coupons, promotions, and advertisements to users can be combined to further improve user payment willingness. Subsequently, offline verification or online testing is used to evaluate the quality and effectiveness of the recommendation results, using indicators such as accuracy, recall rate, or AUC (Area Under Curve, ROC curve area) to evaluate the quality and effectiveness of the recommendation results. According to the evaluation results, the recommendation strategy can be continuously adjusted to achieve better recommendation results.

[0138] In this way, valuable information is mined from the payment history of the user and combined with other data features to generate more accurate video recommendations, which will help attract users to watch the recommended videos and improve their willingness to pay.

[0139] In an embodiment, the media recommendation method further comprises:

[0140] obtaining recommendation feedback data of the target object for the recommended media; determining a recommendation quality based on the recommendation feedback data; and updating at least one of the target matching model and the target protection parameter when the recommendation quality is less than a quality threshold.

[0141] The recommendation feedback data is the satisfaction, opinion, suggestion or behavioral response expressed by the target object after the recommended media is pushed to the target object. For example, the recommendation feedback data can be whether the target object interacts with the recommended media after the recommended media is pushed to the target object.

[0142] After determining the recommended media corresponding to the target object, the recommendation feedback data of the target object for the recommended media is obtained, and the recommendation quality is determined according to the recommendation feedback data, which represents the recommendation effectiveness of the recommended media. The quality threshold is a threshold for the recommendation quality, which can be set according to actual needs.

[0143] Specifically, after the recommended media is pushed to the target object, the computer device can obtain the recommendation feedback data of the target object for the recommended media, analyze the recommendation feedback data, and obtain the recommendation quality. For example, there are multiple target objects, and the recommendation feedback data of each target object for the respective recommended media is obtained, and the proportion of positive feedback in the recommendation feedback data is counted to obtain the recommendation quality. Positive feedback indicates that the user is satisfied with the recommended media, for example, if the user opens the recommended media to browse the content of the recommended media, it is determined as positive feedback. Further, the computer device compares the recommendation quality with the quality threshold, if the recommendation quality is greater than or equal to the quality threshold, it means that the recommendation effect meets the expectation, and the recommendation effect is good, if the recommendation quality is less than the quality threshold, it means that the recommendation effect is not good. If the recommendation quality is less than the quality threshold, the computer device can select to update at least one of the target matching model and the target protection parameter to improve the recommendation quality.

[0144] It can be understood that the recommendation quality can be determined at a time, for example, the recommendation quality is determined according to the recommendation feedback data of each target object for the respective recommended media in the last month every month. The recommendation quality can also be determined at any time, for example, the recommendation feedback data of multiple target objects for the respective recommended media is obtained based on the needs of the operation personnel or the developer to determine the recommendation quality when the operation personnel or the developer needs data analysis.

[0145] It can be understood that updating the target matching model can be adjusting the model parameters of the target matching model, or replacing the model. For example, new training samples and corresponding training labels are obtained, and the target matching model is fine-tuned. Updating the target protection parameter can be adjusting at least one of the noise generation mode and the noise intensity in the target protection parameter.

[0146] In the above embodiments, the recommendation quality is determined based on the recommendation feedback data, and at least one of the target matching model and the target protection parameter is updated when the recommendation quality is less than the quality threshold. This can timely adjust and optimize the recommendation strategy when the recommendation effect is not good, so as to improve the recommendation effect.

[0147] In one specific application, taking a short video as an example, with reference to FIG. 8, the background server of the short video application collects the interaction data and personal data of the user. For example, the interaction data can be the short videos watched by the user in the short video application, payment history, etc., and the personal data can be the registration data of the user in the short video application, user activity, etc. The background server extracts user information and consumption features from the interaction data and personal data of the user. For example, the user information can be an object portrait constructed according to the interaction data and personal data of the user, and the consumption features can be object features obtained by decomposing the interaction matrix. The background server constructs a recommendation model according to the user information and consumption features, and recommends short videos and products to the user according to the recommendation model. For example, the recommendation model can be a machine learning model, and the comprehensive object features for the user and the comprehensive video features for the short video obtained according to the user information and consumption features are input into the recommendation model. The recommendation model outputs the consumption matching degree between the user and the short video, and the short video with a higher consumption matching degree is recommended to the user from the short videos not watched by the user. In addition, when the user watches the short video, the user is recommended related products, so as to facilitate the user to discover interested products through recommended content and improve the user's purchase decision. The background server can generate personalized content for the user according to the recommendation model, and the personalized content is a personalized recommendation list. The personalized recommendation list is pushed to the front end of the short video application, and the recommendation list contains short videos and related products. By providing appropriate short videos and products, the user's purchase intention can be stimulated. The background server can collect user feedback to evaluate the recommendation effect, and timely adjust the recommendation strategy when the recommendation effect is not good, so as to more efficiently recommend short videos and products. The user satisfaction, conversion rate, etc. can be determined according to the user feedback to evaluate the recommendation effect, and the user's consumption intention in the short video application can be effectively improved by continuously optimizing the recommendation algorithm.

[0148] In one embodiment, based on the target object features and the target media features, the transaction matching degree between the target object and the target media is determined, comprising:

[0149] The transaction matching degree between the target object and the target media is obtained based on the feature similarity between the target object feature and the target media feature.

[0150] Specifically, the computer device can obtain the transaction matching degree between the target object and the target media based on the feature similarity between the target object feature and the target media feature. For example, the cosine similarity between the target object feature and the target media feature is calculated as the feature similarity between the target object feature and the target media feature.

[0151] In the above embodiments, the transaction matching degree between the target object and the target media can be quickly obtained by calculating the feature similarity between the target object feature and the target media feature.

[0152] In one embodiment, based on the transaction matching degree, the target object corresponding recommended media is determined from each target media, including:

[0153] The target media with a transaction matching degree greater than a matching degree threshold is obtained as the target object corresponding recommended media; and each recommended media is sorted in descending order of the transaction matching degree to obtain a target object corresponding recommended media sequence.

[0154] The matching degree threshold is a threshold for the transaction matching degree and is used to filter the target media. The matching degree threshold can be set as needed.

[0155] Specifically, if there are multiple target media, the transaction matching degree between the target object and each target media can be obtained. The computer device filters the media based on the matching degree threshold and obtains the target media with a transaction matching degree greater than the matching degree threshold as the target object corresponding recommended media. If there are multiple recommended media, the computer device can sort each recommended media in descending order of the transaction matching degree to obtain a target object corresponding recommended media sequence. The recommended media sequence can indicate the recommended order of the recommended media, and the recommended media with a higher ranking is recommended to the target object first. Subsequently, the computer device can push the corresponding media to the target terminal corresponding to the target object according to the recommended media sequence.

[0156] In the above embodiments, the target media with a transaction matching degree greater than the matching degree threshold is obtained as the target object corresponding recommended media, which can guarantee the recommendation effect and improve the media conversion rate of the recommended media. Each recommended media is sorted in descending order of the transaction matching degree to obtain a target object corresponding recommended media sequence, and the media is recommended to the target object according to the recommended media sequence, which can guarantee the recommendation effect and improve the media conversion rate of the recommended media.

[0157] In one embodiment, the media recommendation method further includes:

[0158] acquire a recommended item matched with the recommended media; determine a preferential activity matched with the target object based on a historical media transaction record of the target object; and push the recommended media, the recommended item and the preferential activity to a target terminal corresponding to the target object in association, so as to cause the target terminal to display the recommended media, the recommended item and the preferential activity in association.

[0159] The recommended item matched with the recommended media can be at least one of an item related to the recommended media content or an item complementary to the recommended media. The item related to the recommended media content can be an item displayed, introduced or used in the recommended media. The item complementary to the recommended media can be an item capable of being used in cooperation with the item related to the recommended media content. For example, if the recommended media is a video introducing a smart home product A, the item related to the recommended media content can be the smart home product A, and the item complementary to the recommended media can be a smart home product B used in cooperation with the smart home product A. The item complementary to the recommended media is not a smart home product directly displayed in the video, but is helpful to improve the overall performance and convenience of the smart home system.

[0160] The preferential activity refers to an activity for providing a preferential treatment for an item transaction. For example, the preferential activity can be a coupon, a promotion activity, a discount activity, etc. The consumption habit of the target object is analyzed based on the historical media transaction record of the target object, and a preferential activity matched with the consumption habit is determined as the preferential activity matched with the target object.

[0161] Specifically, the computer device can acquire a recommended item matched with the recommended media, and push the recommended media and the recommended item to a target terminal corresponding to the target object in association, so as to cause the target terminal to display the recommended media and the recommended item in association. For example, the recommended item and an item transaction portal corresponding to the recommended item are displayed in a media display interface of the recommended media.

[0162] Further, the computer device can determine the preferential activity matched with the target object based on the historical media transaction record of the target object. For example, classify the items in the historical media transaction record, obtain the category coupon corresponding to the item type with the most transaction times as the preferential activity, or obtain the category coupon corresponding to the item type with the largest total transaction amount, and push the category coupon to the target object; obtain the discount activity or the promotion activity of the item with the most transaction times as the preferential activity, or obtain the discount activity or the promotion activity of the item with the largest total transaction amount as the preferential activity, and push the discount activity or the promotion activity to the target object; and the like. The computer device associates and pushes the recommended media, the recommended item and the preferential activity to the target terminal corresponding to the target object, so that the target terminal associates and displays the recommended media, the recommended item and the preferential activity. For example, the recommended item and the preferential activity are displayed in the media display interface of the recommended media, if the user triggers the recommended item, the item detail interface can be entered, and the item transaction can be triggered in the item detail interface, if the user triggers the preferential activity, the preferential qualification can be obtained or the activity detail interface can be entered, and the preferential qualification can be obtained in the activity detail interface.

[0163] In the above embodiments, the recommended item matched with the recommended media is obtained, the preferential activity matched with the target object is determined based on the historical media transaction record of the target object, the recommended media, the recommended item and the preferential activity are associated and pushed to the target terminal corresponding to the target object, which can realize personalized recommendation, and the transaction matching degree between the recommended media and the target object is high, the preferential activity is related to the historical media transaction record of the target object, and the consumption willingness of the target object can be effectively improved by associating and displaying the recommended media, the recommended item and the preferential activity, so as to improve the conversion rate of the recommended media.

[0164] In a specific application, the method of the present application can be applied to a short video platform. The present application provides a short video recommendation scheme combined with user consumption habits, which can help enterprises achieve business goals such as optimizing promotion, improving sales and increasing customer interaction while maintaining user satisfaction and optimizing user experience. The payment history of the user is introduced into the characteristics of the user, so that the short video recommendation can more accurately point to the consumption tendency of the user, improve the consumption willingness of the user, encourage more consumption type merchants to settle in the short video platform, and improve the content quality and rich video classification of the short video. And in the process of collecting user data, a privacy protection algorithm is used to fully protect the privacy and security of the user.

[0165] Referring to FIG. 9, the background server of the short video platform acquires user data, determines sensitive data in the user data, for example, payment history, personal information, and viewing records in the user data as sensitive data, and for the sensitive data, a differential privacy algorithm can be used to ensure the security of the sensitive data. The background server selects appropriate differential privacy algorithms and privacy parameters according to the data type and requirements of the sensitive data, adds a certain degree of noise to the sensitive data according to the selected differential privacy algorithms and privacy parameters, to balance privacy protection and data utility while protecting privacy. Further, the background server can perform aggregation processing on the user data, for example, calculating user activity based on user data, calculating transaction scores, etc., and can perform desensitization processing on the user data, for example, replacing the user's contact information with a specific string, masking part of the user's transaction account data, etc. Further, the background processor performs feature engineering on the processed user data, selects and generates features related to recommendation, for example, based on user data to establish an interaction matrix, performs matrix decomposition on the interaction matrix to obtain an object feature matrix and a short video feature matrix, combines the object features of the target user in the object feature matrix and the object features established based on other user data of the target user to obtain the comprehensive object features of the target user, acquires short videos that the target user has not watched as target short videos, and combines the short video features of the target short videos in the short video feature matrix and the short video features established based on other short video data of the target short videos to obtain the comprehensive short video features of the target short videos. The background server constructs a recommendation model based on the generated feature data, tests the recommendation effect of the recommendation model using the feature data protected by differential privacy, and compares it with the recommendation effect obtained using the original feature data, adjusts the model parameters and differential privacy related parameters to find the best balance between privacy and recommendation effect, and under the premise of protecting the privacy and safety of users, enables the model to understand the consumption habits of different users, such as being more accustomed to consuming in which industry or type of merchant, so that the short video platform can recommend short videos of merchants that are more in line with the consumption habits of the user to the user, thereby improving the user's consumption willingness.

[0166] When the user watches the recommended short video, the user can be recommended products related to or complementary to the short video, thereby facilitating the user to discover interesting products through recommended content, improving the user's purchase decision, and improving the conversion rate of the platform. When the user watches the recommended short video, in addition to recommending products to the user, the user's consumption habits can be analyzed based on the user's payment history, and the user can be recommended coupons, discounts, and promotional activities that match the user's consumption habits, for example, analyzing the user's payment history to obtain the most frequently traded item type of the user, and recommending the corresponding category coupon of the item type to the user. These incentive strategies can encourage consumers to generate more consumption behavior on the short video platform and improve the conversion rate of the platform.

[0167] For the method of the present application, when collecting user data, privacy protection technology can ensure the security and privacy of user data. According to the prediction result of the recommendation model, short videos and products highly related to the user's payment willingness are recommended to the user, improving the user's possibility of purchasing goods or services. Through the method of the present application, users can discover more short videos that meet their personal tastes and preferences, thereby increasing their dwell time and engagement on the platform. Through precise product promotion and incentive policies such as coupons and promotional activities, users can be stimulated to make purchases, improving the conversion rate of the platform.

[0168] It can be understood that the method of the present application can not only be applied to short video platforms, but also to other video platforms, such as long video platforms, mixed platforms of long videos and short videos, etc., and can also be applied to other platforms, such as music platforms, game platforms, social platforms, etc.

[0169] Further, the various platforms in the present application can be application programs or components of application programs. For example, the short video platform can be a short video platform in a social application program. In a specific application, the social application program includes a friend dynamic platform for displaying friend dynamics, a short video platform for displaying short videos, a live platform for displaying live broadcasts, etc. The short video platform can be a short video platform in the social application program. Referring to FIG. 10, a user enters the short video platform by triggering the entrance of the short video platform in the social application program, and the user can browse short videos in the short video platform. Through the method of the present application, short videos, products, and preferential activities (such as coupons, discounts, and promotional activities) related to the user's payment willingness are recommended to the user in the short video platform, which can effectively improve the effectiveness and accuracy of the recommendations, improve the user's willingness to consume, and improve the conversion rate of the video number platform.

[0170] It should be understood that although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0171] Based on the same inventive concept, the embodiments of the present application further provide a media recommendation device for implementing the media recommendation method described above. The device provides a solution to the implementation scheme as described in the above method, and therefore the specific limitations in one or more media recommendation device embodiments provided below can refer to the limitations of the media recommendation method described above, which will not be repeated here.

[0172] In one embodiment, as shown in FIG. 11, a media recommendation device is provided, comprising: an interaction feature set acquisition module 1102, an interaction feature set decomposition module 1104, a target feature determination module 1106, a transaction matching degree determination module 1108, and a recommended media determination module 1110, wherein:

[0173] The interaction feature set acquisition module 1102 is configured to acquire an interaction feature set between an object set and a media set; the interaction feature set is a set generated based on historical media transaction records of the objects in the object set, and the historical media transaction records are the transaction records of the recommended items corresponding to the historical interaction media by the objects.

[0174] The interaction feature set decomposition module 1104 is configured to decompose the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set.

[0175] The target feature determination module 1106 is configured to determine a target object from the object set, acquire media that the target object has not interacted with from the media set as a target media, acquire target object features of the target object from the object feature set, and acquire target media features of the target media from the media feature set.

[0176] The transaction matching degree determination module 1108 is configured to determine a transaction matching degree between the target object and the target media based on the target object features and the target media features.

[0177] The recommended media determination module 1110 is configured to determine a recommended media corresponding to the target object from each target media based on the transaction matching degree.

[0178] In one embodiment, the interaction feature set acquisition module 1102 is further configured to:

[0179] acquire historical media transaction records of each object in the object set;

[0180] determine item transaction statistical values of the objects for the corresponding historical interaction media based on the historical media transaction records of the objects;

[0181] generate an interaction matrix based on the item transaction statistical values of each object for respective historical interaction media; a first dimension of the interaction matrix represents the objects in the object set, and a second dimension of the interaction matrix represents the media in the media set;

[0182] The interaction matrix is taken as an interaction feature set between the object set and the media set.

[0183] In one embodiment, the interaction feature set acquisition module 1102 is further configured to:

[0184] Obtain a target protection parameter for the item transaction statistical value as a transaction protection parameter;

[0185] Determine a target transaction statistical value from the item transaction statistical values of each object for respective historical interaction media;

[0186] Based on the transaction protection parameter, add noise data to the target transaction statistical value to obtain an encrypted transaction statistical value;

[0187] Based on the encrypted transaction statistical value and the remaining item transaction statistical values, generate an interaction matrix.

[0188] In one embodiment, the interaction feature set acquisition module 1102 is further configured to:

[0189] From the item transaction statistical values of each object for respective historical interaction media, obtain the item transaction statistical values corresponding to the objects with an object activity greater than an activity threshold as target transaction statistical values.

[0190] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0191] Obtain an object attribute feature of the target object, combine the target object feature and the object attribute feature to obtain a comprehensive object feature of the target object;

[0192] Obtain a media attribute feature of the target media, combine the target media feature and the media attribute feature to obtain a comprehensive media feature of the target media;

[0193] Based on the comprehensive object feature and the comprehensive media feature, determine the transaction matching degree between the target object and the target media.

[0194] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0195] Obtain a plurality of attribute data of the target object;

[0196] Based on the data type of the target attribute data in the plurality of attribute data, determine an attribute protection parameter corresponding to the target attribute data, and based on the attribute protection parameter, add noise data to the target attribute data to obtain encrypted attribute data;

[0197] The features are extracted from the encrypted attribute data and the attribute data other than the target attribute data, to obtain object attribute features of the target object.

[0198] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0199] Determine a noise distribution based on the noise generation mode and the noise intensity in the attribute protection parameter, and sample noise data from the noise distribution based on the target attribute data;

[0200] Add the sampled noise data to the target attribute data to obtain the encrypted attribute data.

[0201] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0202] Input the comprehensive object features and the comprehensive media features into the target matching model, and the target matching model outputs a transaction matching degree between the target object and the target media; the target matching model is obtained by training an initial matching model based on encrypted training samples and corresponding training labels, and the encrypted training samples are obtained by adding noise data to original training samples based on target protection parameters.

[0203] In one embodiment, the media recommendation apparatus is further configured to:

[0204] Obtain a first original training sample and a training label corresponding to the first original training sample;

[0205] Train an initial matching model based on the first original training sample and the training label corresponding to the first original training sample, to obtain an intermediate matching model;

[0206] Obtain a second original training sample and a training label corresponding to the second original training sample;

[0207] Add noise data to the second original training sample based on the initial protection parameter, to obtain an encrypted training sample;

[0208] Input the second original training sample and the encrypted training sample into the intermediate matching model, to obtain predicted labels corresponding to the second original training sample and the encrypted training sample respectively;

[0209] Adjust the initial protection parameter based on the difference between the predicted labels corresponding to the second original training sample and the encrypted training sample respectively, and adjust model parameters of the intermediate matching model based on the difference between the predicted label corresponding to the encrypted training sample and the training label corresponding to the second original training sample, until a convergence condition is met, to obtain the target matching model and the target protection parameter.

[0210] In one embodiment, the media recommendation apparatus is further configured to:

[0211] obtaining recommendation feedback data of the target object for the recommended media;

[0212] determining recommendation quality based on the recommendation feedback data;

[0213] updating at least one of the target matching model and the target protection parameter when the recommendation quality is less than a quality threshold.

[0214] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0215] inputting the comprehensive object feature and the comprehensive media feature into the target matching model;

[0216] In the target matching model, the comprehensive object feature and the comprehensive media feature are fused to obtain first matching features, the comprehensive object feature and the comprehensive media feature are spliced to obtain comprehensive spliced features, features of the comprehensive spliced features are extracted to obtain second matching features, the first matching features and the second matching features are spliced to obtain target matching features, and the transaction matching degree between the target object and the target media is determined based on the target matching features.

[0217] The target matching model outputs the transaction matching degree.

[0218] In one embodiment, the transaction matching degree determination module 1108 is further configured to:

[0219] obtaining the transaction matching degree between the target object and the target media based on the feature similarity between the target object feature and the target media feature.

[0220] In one embodiment, the recommended media determination module 1110 is further configured to:

[0221] obtaining the target media with the transaction matching degree greater than a matching degree threshold as the recommended media corresponding to the target object;

[0222] sorting the recommended media in descending order of the transaction matching degree to obtain a recommended media sequence corresponding to the target object.

[0223] In one embodiment, the recommended media determination module 1110 is further configured to:

[0224] obtaining a recommended item matched with the recommended media;

[0225] associatively pushing the recommended media and the recommended item to a target terminal corresponding to the target object, so that the target terminal associatively displays the recommended media and the recommended item.

[0226] In one embodiment, the recommended media determination module 1110 is further configured to:

[0227] obtaining a recommended item matched with the recommended media;

[0228] determining a preferential activity matched with the target object based on the historical media transaction record of the target object;

[0229] pushing the recommended media, the recommended item and the preferential activity to a target terminal corresponding to the target object, so that the target terminal displays the recommended media, the recommended item and the preferential activity in association.

[0230] In an embodiment, the recommended media determining module 1110 is further configured to:

[0231] obtaining at least one of an item related to the recommended media content and an item complementary to the recommended media as the recommended item matched with the recommended media.

[0232] The media recommendation apparatus can more accurately reflect the matching degree between the target object and the target media by comprehensively analyzing the transaction matching degree obtained based on the respective characteristics of the target object and the target media, and can effectively improve the effectiveness and accuracy of media recommendation by determining whether to recommend the target media to the target object based on the transaction matching degree, so as to achieve more accurate and effective personalized recommendation. Further, the interaction feature set can reflect the interaction and transaction relationship between the user set and the media set, the object features in the object feature set obtained by decomposing the interaction feature set can reflect the consumption preferences of the user, the media features in the media feature set obtained by decomposing the interaction feature set can reflect the characteristics of the media, the consumption willingness of the target user for the un-interacted target media can be predicted based on the object features of the target user and the media features of the target media, the transaction matching degree can be obtained, and the media recommendation can be performed based on the transaction matching degree, so as to improve the consumption interest of the target user for the recommended media and further improve the effectiveness and accuracy of media recommendation.

[0233] Each module in the media recommendation apparatus can be realized by software, hardware and a combination thereof in whole or in part. Each module can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0234] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in FIG. 12. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer readable instructions in the non-volatile storage medium. The database of the computer device is used to store data such as a recommendation model and a protection parameter. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer readable instructions are executed by the processor to implement a media recommendation method.

[0235] In one embodiment, a computer device is provided, which can be a terminal, and an internal structure diagram thereof can be as shown in FIG. 13. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer readable instructions. The internal memory provides an environment for the operation of the operating system and computer readable instructions in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication), or other technologies. The computer readable instructions are executed by the processor to implement a media recommendation method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0236] Those skilled in the art can understand that the structures shown in FIG. 12 and FIG. 13 are only block diagrams of part of the structures related to the schemes of the present application, and do not constitute a limitation on the computer device to which the schemes of the present application are applied. The specific computer device can include more or fewer components than those shown in the figures, or combine certain components, or have a different arrangement of components.

[0237] In an embodiment, a computer device is also provided, including a memory and a processor, the memory storing computer readable instructions, and the processor implementing the steps in the above method embodiments when executing the computer readable instructions.

[0238] In an embodiment, a computer readable storage medium is provided, storing computer readable instructions, and the processor implementing the steps in the above method embodiments when executing the computer readable instructions.

[0239] In an embodiment, a computer program product is provided, including computer readable instructions, and the processor implementing the steps in the above method embodiments when executing the computer readable instructions.

[0240] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0241] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through computer readable instructions, and the computer readable instructions can be stored in a non-volatile computer readable storage medium. When the computer readable instructions are executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. The non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. The volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0242] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0243] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A media recommendation method characterized by, The method is executed by a computer device, and the method comprises: obtaining an interaction feature set between an object set and a media set; the interaction feature set is a set generated based on historical media transaction records of objects in the object set, and the historical media transaction records are object transaction records of recommended items corresponding to historical interaction media of objects; decomposing the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set; determining a target object from the object set, obtaining media that the target object has not interacted with from the media set as target media, obtaining target object features of the target object from the object feature set, and obtaining target media features of the target media from the media feature set; determining a transaction matching degree between the target object and the target media based on the target object features and the target media features; and determining recommended media corresponding to the target object from each target media based on the transaction matching degree.

2. The method of claim 1, wherein, The obtaining of the interaction feature set between the object set and the media set comprises: obtaining historical media transaction records of each object in the object set; determining object transaction statistical values of objects for corresponding historical interaction media based on the historical media transaction records of the objects; generating an interaction matrix based on the object transaction statistical values of each object for the respective historical interaction media; a first dimension of the interaction matrix represents the objects in the object set, and a second dimension of the interaction matrix represents the media in the media set; and taking the interaction matrix as the interaction feature set between the object set and the media set.

3. The method of claim 2, wherein, The generating of the interaction matrix based on the object transaction statistical values of each object for the respective historical interaction media comprises: obtaining a target protection parameter for the object transaction statistical values as a transaction protection parameter; determining a target transaction statistical value from the object transaction statistical values of each object for the respective historical interaction media; adding noise data to the target transaction statistical value based on the transaction protection parameter to obtain an encrypted transaction statistical value; and generating the interaction matrix based on the encrypted transaction statistical value and the remaining object transaction statistical values.

4. The method of claim 3, wherein, The determining of the target transaction statistical value from the object transaction statistical values of each object for the respective historical interaction media comprises: obtaining object transaction statistical values corresponding to objects with an object activity greater than an activity threshold from the object transaction statistical values of each object for the respective historical interaction media as the target transaction statistical value.

5. The method according to any one of claims 1 to 4, characterized in that, The determining of the transaction matching degree between the target object and the target media based on the target object features and the target media features comprises: obtaining object attribute features of the target object, combining the target object features and the object attribute features to obtain comprehensive object features of the target object; obtaining media attribute features of the target media, combining the target media features and the media attribute features to obtain comprehensive media features of the target media; and Determine a transaction matching degree between the target object and the target media based on the comprehensive object feature and the comprehensive media feature.

6. The method of claim 5, wherein, The object attribute feature of the target object is obtained by: Obtaining a plurality of attribute data of the target object; Based on the data type of the target attribute data in the plurality of attribute data, determine the attribute protection parameter corresponding to the target attribute data, based on the attribute protection parameter, add noise data to the target attribute data to obtain encrypted attribute data; and Extract features from the encrypted attribute data and other attribute data in the plurality of attribute data except the target attribute data to obtain the object attribute feature of the target object.

7. The method of claim 6, wherein, The encrypted attribute data is obtained by adding noise data to the target attribute data based on the attribute protection parameter, including: Based on the noise generation mode and noise intensity in the attribute protection parameter, determine the noise distribution, sample noise data from the noise distribution based on the target attribute data; and Add the sampled noise data to the target attribute data to obtain the encrypted attribute data.

8. The method of claim 5, wherein, The transaction matching degree between the target object and the target media is determined based on the comprehensive object feature and the comprehensive media feature, including: Input the comprehensive object feature and the comprehensive media feature into a target matching model, and the target matching model outputs the transaction matching degree between the target object and the target media; the target matching model is obtained by training an initial matching model based on encrypted training samples and corresponding training labels, and the encrypted training samples are obtained by adding noise data to original training samples based on target protection parameters.

9. The method of claim 8, wherein, The method further comprises: Obtaining a first original training sample and a training label corresponding to the first original training sample; Based on the first original training sample and the training label corresponding to the first original training sample, train an initial matching model to obtain an intermediate matching model; Obtaining a second original training sample and a training label corresponding to the second original training sample; Based on the initial protection parameter, add noise data to the second original training sample to obtain an encrypted training sample; Input the second original training sample and the encrypted training sample into the intermediate matching model to obtain the corresponding predicted labels of the second original training sample and the encrypted training sample; and Based on the difference between the predicted labels corresponding to the second original training sample and the encrypted training sample, adjust the initial protection parameter, and based on the difference between the predicted label corresponding to the encrypted training sample and the training label corresponding to the second original training sample, adjust the model parameters of the intermediate matching model until the convergence condition is met, to obtain the target matching model and the target protection parameter.

10. The method of claim 8, wherein, The method further comprises: Obtaining recommendation feedback data of the target object for the recommended media; Determine the recommendation quality based on the recommendation feedback data; and When the recommendation quality is less than the quality threshold, update at least one of the target matching model and the target protection parameter.

11. The method of claim 5, wherein, The determining the transaction matching degree between the target object and the target media based on the comprehensive object feature and the comprehensive media feature comprises: inputting the comprehensive object feature and the comprehensive media feature into a target matching model; in the target matching model, fusing the comprehensive object feature and the comprehensive media feature to obtain a first matching feature, splicing the comprehensive object feature and the comprehensive media feature to obtain a comprehensive splicing feature, extracting a feature from the comprehensive splicing feature to obtain a second matching feature, splicing the first matching feature and the second matching feature to obtain the target matching feature, and determining the transaction matching degree between the target object and the target media based on the target matching feature; and the target matching model outputs the transaction matching degree.

12. The method according to any one of claims 1 to 11, characterized in that, The determining the transaction matching degree between the target object and the target media based on the target object feature and the target media feature comprises: obtaining the transaction matching degree between the target object and the target media based on a feature similarity between the target object feature and the target media feature.

13. The method according to any one of claims 1 to 12, characterized in that, The determining the recommended media corresponding to the target object from the target media based on the transaction matching degree comprises: obtaining a target media with a transaction matching degree greater than a matching degree threshold as the recommended media corresponding to the target object; and sorting the recommended media in descending order of the transaction matching degree to obtain a recommended media sequence corresponding to the target object.

14. The method according to any one of claims 1 to 13, characterized in that, The method further comprises: obtaining a recommended item matched with the recommended media; and associating and pushing the recommended media and the recommended item to a target terminal corresponding to the target object, so that the target terminal displays the recommended media and the recommended item in association.

15. The method according to any one of claims 1 to 13, characterized in that, The method further comprises: obtaining a recommended item matched with the recommended media; determining a preferential activity matched with the target object based on a historical media transaction record of the target object; and associating and pushing the recommended media, the recommended item and the preferential activity to a target terminal corresponding to the target object, so that the target terminal displays the recommended media, the recommended item and the preferential activity in association.

16. The method according to claim 14 or 15, characterized in that, The obtaining a recommended item matched with the recommended media comprises: obtaining at least one of an item related to the content of the recommended media and an item complementary to the recommended media as the recommended item matched with the recommended media.

17. A media recommendation apparatus, characterized by comprising: The device comprises: an interaction feature set acquisition module configured to acquire an interaction feature set between an object set and a media set, the interaction feature set being a set generated based on a historical media transaction record of an object in the object set, the historical media transaction record being an item transaction record of a recommended item corresponding to a historical interaction media by the object; an interaction feature set decomposition module configured to decompose the interaction feature set into an object feature set corresponding to the object set and a media feature set corresponding to the media set; and an object feature set acquisition module configured to acquire an object feature set corresponding to a target object from the object feature set corresponding to the object set; and a media feature set acquisition module configured to acquire a media feature set corresponding to a target media from the media feature set corresponding to the media set. determine a target object from the object set, acquire media that the target object has not interacted with from the media set as target media, acquire a target object feature of the target object from the object feature set, and acquire a target media feature of the target media from the media feature set; determine a transaction matching degree between the target object and the target media based on the target object feature and the target media feature; and determine a recommended media corresponding to the target object from each target media based on the transaction matching degree.

18. A computer device comprising a memory and a processor, the memory storing computer readable instructions, wherein, The processor executes the computer readable instructions to implement the steps of the method in any one of claims 1 to 16.

19. A computer-readable storage medium having stored thereon computer-readable instructions, wherein, The computer readable instructions are executed by the processor to implement the steps of the method in any one of claims 1 to 16.

20. A computer program product comprising computer readable instructions, characterized in that, The computer readable instructions are executed by the processor to implement the steps of the method in any one of claims 1 to 16.

Images