Method and device for determining labels of media resources, storage medium, and electronic device

Abstract

The present invention discloses a method and device for determining labels for media resources, a storage medium, and an electronic device. The method includes: in a media resource push system, the purpose of the media resources preferred by the target user group can be determined based on the user profile characteristics (basic user information) of each user in the target user group and the resource content characteristics of the media resources. Based on the labels of the media resources, the users who prefer the media resources can be identified, and the media resources can be effectively pushed to the users, thereby solving the technical problem of the inability to effectively push media resources to users in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of computer processing, and in particular to a method and device for determining a label of a media resource, a storage medium, and an electronic device. Background Art

[0002] Nowadays, more and more people use fragmented time to watch media resources on mobile terminal devices, such as watching media resources through an application software on the way to and from get off work. The application software usually pushes various media resources to users, such as articles, short videos, videos, audio, etc.

[0003] When pushing media resources, applications can usually push the media resources to relevant users based on the media resource's tags. For example, if the media resource is an article about games, the article will be pushed to users who play the game based on the game tag. How to accurately determine the tag information of the media resource is becoming increasingly important in the process of pushing media resources.

[0004] Existing solutions employ deep learning models that take a user's historical click behavior and interest profiles as input, along with article content features, to learn user / article feature representations for article recall. This requires users to have a relatively rich set of interest profiles and historical click behavior. However, this approach is not user-friendly for users with limited click behavior and interest profiles, such as new users of certain apps. Because new users have relatively limited click behavior and interest profiles, the articles recalled by the model are often trending articles that do not align with user preferences. This means that user data cannot accurately identify the preferred group for an article, making it difficult to push media resources to interested users.

[0005] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0006] The embodiments of the present invention provide a method and apparatus for determining a tag of a media resource, a storage medium, and an electronic device, so as to at least solve the technical problem in the prior art that media resources cannot be effectively pushed to users.

[0007] According to one aspect of an embodiment of the present invention, a method for determining a label for a media resource is provided, comprising: obtaining user portrait features of each user in a target user group, wherein the group label of the target user group is a target group label, and the user portrait features of each user match the target group label; obtaining resource content features of each media resource in a target media resource set; combining the user portrait features of each user and the resource content features of each media resource to form a plurality of input information, wherein each input information includes a user portrait feature of a user and a resource content feature of a media resource; inputting the plurality of input information into a target neural network model respectively to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether a user corresponding to an input information will click on a corresponding media resource; determining a target media resource matching the target user group in the target media resource set based on the plurality of prediction results, and determining the target group label as the label of the target media resource.

[0008] According to another aspect of an embodiment of the present invention, a device for determining a label of a media resource is also provided, including: a first acquisition unit, used to obtain user portrait features of each user in a target user group, wherein the group label of the target user group is a target group label, and the user portrait features of each user match the target group label; a second acquisition unit, used to obtain resource content features of each media resource in a target media resource set; a composition unit, used to combine the user portrait features of each user and the resource content features of each media resource to form a plurality of input information, wherein each input information includes a user portrait feature of a user and a resource content feature of a media resource; an output unit, used to input the plurality of input information into a target neural network model respectively to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to a piece of input information will click on the corresponding media resource; a first determination unit, used to determine a target media resource matching the target user group in the target media resource set based on the plurality of prediction results, and determine the target group label as the label of the target media resource.

[0009] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored. The computer program is configured to execute the above-mentioned method for determining a label of a media resource when running.

[0010] According to another aspect of an embodiment of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-mentioned method for determining a label of a media resource through the computer program.

[0011] In an embodiment of the present invention, by obtaining the user portrait features of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait features of each user are matched with the target group label; obtaining the resource content features of each media resource in the target media resource set; combining the user portrait features of each user and the resource content features of each media resource to form multiple input information, wherein each input information includes the user portrait features of a user and the resource content features of a media resource; inputting the multiple input information into the target neural network model respectively, obtaining multiple prediction results output by the target neural network model, wherein each prediction result is used to It indicates whether the user corresponding to an input message will click on the corresponding media resource; based on multiple prediction results, the target media resource matching the target user group is determined in the target media resource set, and the target group label is determined as the label of the target media resource, thereby achieving the purpose of determining the media resources preferred by this type of users in the target user group based on the user portrait characteristics (basic information of the user) and the resource content characteristics of the media resource of each user in the target user group. According to the label of the media resource, the user who prefers the media resource can be known, and the media resource can be effectively pushed to the user, thereby solving the technical problem in the existing technology that media resources cannot be effectively pushed to users. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0013] Figure 1 is a schematic diagram of an application environment of an optional method for determining a label of a media resource according to an embodiment of the present invention;

[0014] Figure 2 This is a schematic structural diagram of an optional distributed system applied to a blockchain system according to an embodiment of the present invention;

[0015] Figure 3 is a schematic diagram of an optional block structure according to an embodiment of the present invention;

[0016] Figure 4 is a flowchart of an optional method for determining a label of a media resource according to an embodiment of the present invention;

[0017] Figure 5is a structural block diagram of an optional method for determining a label of a media resource according to an embodiment of the present invention;

[0018] Figure 6 1 is a schematic diagram of a scenario application of an optional method for determining a tag of a media resource according to an embodiment of the present invention;

[0019] Figure 7 This is a schematic diagram of a structure for determining an optional feature combination of a target user group according to an embodiment of the present invention;

[0020] Figure 8 is a schematic structural diagram of an optional target neural network model according to an embodiment of the present invention;

[0021] Figure 9 is a schematic structural diagram of an optional target neural network model according to an embodiment of the present invention;

[0022] Figure 10 is a schematic diagram of an optional recursive process of a cross network according to an embodiment of the present invention;

[0023] Figure 11 is a schematic diagram of a function image of an optional negative sample sampling according to an embodiment of the present invention;

[0024] Figure 12 is a schematic diagram of an optional salient feature according to an embodiment of the present invention;

[0025] Figure 13 is a schematic diagram of an optional construction feature combination label according to an embodiment of the present invention;

[0026] Figure 14 is a schematic structural diagram of an optional device for determining a label of a media resource according to an embodiment of the present invention;

[0027] Figure 15 FIG. 4 is a schematic structural diagram of an optional electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] According to one aspect of an embodiment of the present invention, a method for determining a label of a media resource is provided. Optionally, as an optional implementation, the above-mentioned method for determining a label of a media resource can be applied to, but is not limited to, Figure 1 In the environment shown, the media resource tag determination system may include but is not limited to a terminal device 102, a network 110, and a server 112. The terminal device 102 runs a target client for a user to browse media resources.

[0031] The terminal device 102 may include, but is not limited to, a human-computer interaction screen 104, a processor 106, and a memory 108. The human-computer interaction screen 104 is used to receive human-computer interaction instructions through a human-computer interaction interface and to present target media resources. The processor 106 is used to respond to these human-computer interaction instructions and assist the user in completing operations on the target media resources (such as turning pages and adjusting the screen). The memory 108 is used to store resource content attribute information of the target media resources. Here, the server may include, but is not limited to, a database 114 and a processing engine 116, wherein the processing engine 116 is configured to call user portrait features of users and resource content features of media resources stored in the database 114, and combine the user portrait features of each user and the resource content features of each media resource to form a plurality of input information, wherein each input information includes a user portrait feature of a user and a resource content feature of a media resource; the plurality of input information are respectively input into the target neural network model to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to a piece of input information will click on the corresponding media resource; based on the plurality of prediction results, a target media resource matching the target user group is determined in the target media resource set, and the target group label is determined as the label of the target media resource, thereby achieving the purpose of determining the media resources preferred by the target user group based on the user portrait features (basic information of the user) and the resource content features of the media resources of each user in the target user group; based on the label of the media resource, the user who prefers the media resource can be known, and the media resource can be effectively pushed to the user, thereby solving the technical problem in the prior art of not being able to effectively push media resources to the user.

[0032] The specific process is as follows: Steps S102-S112, where the user portrait features of the user and the resource content features of the media resources are obtained and sent to the server 112 via the network 110. In the server 112, the user portrait features of each user and the resource content features of each media resource are combined into a plurality of input information, wherein each input information includes a user portrait feature of a user and a resource content feature of a media resource; the plurality of input information are respectively input into the target neural network model to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to a piece of input information will click on the corresponding media resource; based on the plurality of prediction results, the target media resource that matches the target user group is determined in the target media resource set, and the target group label is determined as the label of the target media resource. The above-mentioned determined result is then returned to the terminal device 102.

[0033] Then, as shown in steps S114-S116, the terminal device 102 pushes the target media resource to other users based on the tag information of the target media resource when the target media resource is tagged, thereby avoiding the inaccurate determination of the media resource tag information due to the inability to obtain other user features, that is, insufficient other features, in the process of determining the media resource tag based on other user features, thereby solving the technical problem in the existing technology that media resources cannot be effectively pushed to users.

[0034] Optionally, in this embodiment, the above-mentioned terminal device can be a terminal device configured with a target client, which can include but is not limited to at least one of the following: a mobile phone (such as an Android phone, an iOS phone, etc.), a laptop computer, a tablet computer, a PDA, an MID (Mobile Internet Devices), a PAD, a desktop computer, a smart TV, etc. The target client can be a video client, an instant messaging client, a browser client, an education client, etc. The above-mentioned network can include but is not limited to: a wired network, a wireless network, wherein the wired network includes: a local area network, a metropolitan area network and a wide area network, and the wireless network includes: Bluetooth, WIFI and other networks that realize wireless communication. The above-mentioned server can be a single server, or it can be a server cluster composed of multiple servers, or a cloud server. The above is only an example, and this embodiment does not impose any limitation on this.

[0035] Optionally, the system involved in the embodiment of the present invention may be a distributed system formed by connecting a client and multiple nodes (any form of computing devices in an access network, such as a server and a user terminal) through network communication.

[0036] Taking the distributed system as the blockchain system as an example, see Figure 2 , Figure 2 This is a schematic diagram of an optional architecture for a distributed system 100, provided in an embodiment of the present invention, applied to a blockchain system. The system consists of multiple nodes (any type of computing device connected to a network, such as a server or user terminal) and clients, forming a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol running on top of the Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or terminal, can join and become a node. A node comprises a hardware layer, an intermediate layer, an operating system layer, and an application layer.

[0037] See also Figure 1 The functions of each node in the blockchain system shown include:

[0038] 1) Routing: A basic function of a node, used to support communication between nodes.

[0039] In addition to the routing function, nodes can also have the following functions:

[0040] 2) Applications, deployed in the blockchain, implement specific services based on actual business needs, record data related to the implementation of functions to form record data, carry digital signatures in the record data to indicate the source of the task data, and send the record data to other nodes in the blockchain system for other nodes to add the record data to a temporary block when they successfully verify the source and integrity of the record data.

[0041] In this embodiment, the tag of the determined media resource can be written into the blockchain for storage. As the user changes, the user corresponding to the media resource may change, and the tag operation of the media resource can be modified or deleted in the blockchain.

[0042] 3) Blockchain, including a series of blocks that are connected to each other in the order of their generation. Once a new block is added to the blockchain, it will not be removed. The block records the record data submitted by the nodes in the blockchain system.

[0043] See also Figure 3 , Figure 3 This is an optional schematic diagram of the block structure provided by an embodiment of the present invention. Each block includes the hash value of the transaction records stored in the block (the hash value of the current block) and the hash value of the previous block. The blocks are connected by hash values ​​to form a blockchain. In addition, the block may also include information such as the timestamp when the block was generated. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each data block contains relevant information used to verify the validity of the information (anti-counterfeiting) and generate the next block.

[0044] Alternatively, as an optional implementation, Figure 4 As shown, the above method for determining the label of the media resource includes:

[0045] Step S402: obtaining a user portrait feature of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait feature of each user matches the target group label;

[0046] Step S404: Obtain resource content features of each media resource in the target media resource set.

[0047] Step S406: Combine the user portrait features of each user and the resource content features of each media resource to form multiple pieces of input information, wherein each piece of input information includes the user portrait features of a user and the resource content features of a media resource.

[0048] In step S408, the plurality of input information are respectively input into the target neural network model to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to a piece of input information will click on the corresponding media resource.

[0049] Step S410 : determining a target media resource that matches the target user group in the target media resource set according to the multiple prediction results, and determining the target group label as a label of the target media resource.

[0050] Optionally, in this embodiment, the above-mentioned method for determining the label of the media resource may include but is not limited to the push process of media resources applied to various applications, that is, according to the label of the media resource, the media resource is pushed to relevant users. Since the media resource is pushed to users who are interested in the media resource, the exposure rate of the media resource is effectively improved.

[0051] The aforementioned media resources may include, but are not limited to, articles, short videos, videos, and audio. For example, some apps will push articles to users interested in them based on their tags. Alternatively, some short video apps can push videos to users interested in them based on their tags. If the video is an advertisement for a product, the app can push it to valid users, meaning users interested in the product, to increase its effective exposure.

[0052] In this embodiment, user portrait features may include 26 demographic attributes such as age, gender, occupation, zodiac sign, and province, and article features may include title, category, tag, and topic.

[0053] In this embodiment, the above-mentioned user portrait features may include feature labels and feature values. For example, the user portrait features of a user are expressed as: Gender_Male, where "gender" can be understood as a feature label and "Male" can be understood as a value. The user portrait features of a user may include multiple features. For example, the user portrait features of user a include: Gender_Female, Age_20, Occupation_Student. That is to say, the user portrait features of user a are identified by three features.

[0054] In this embodiment, the user profile feature refers to the user's own information, which can also be understood as the user's attribute information or basic information. This information can be the information that the user fills in when registering an application account before using an application. For example, when registering a WeChat account, the user can fill in information such as gender, age, occupation, and place of origin. The information that the user fills in when registering an account is understood as the user's basic information, which constitutes the user's user profile feature.

[0055] In this embodiment, the user portrait features of each user in the target user group include features that match the target group label. If the target group label of the target user group is: gender_male, life status_middle-aged, occupation_sales, then the user portrait features of each user in the target user group include the three features of gender, age, and occupation.

[0056] The target group label can be used to distinguish the target user group from other user groups. The target user group includes 8 users, as shown in Table 1, and the user portrait features of each user.

[0057] Table 1

[0058]

[0059] It should be noted that each user in the target user group can have multiple user portrait features. The types of each user portrait feature can be different or the same, but each user includes the three features of gender, age, and occupation.

[0060] In this embodiment, the user portrait features of each user and the resource content features of each media resource are combined to form multiple input information. For example, if the target user group includes 4 users and the target media resource set includes 5 resources, the input information shown in Table 2 can be obtained, a total of 20 input information.

[0061] Table 2

[0062] User 1 User 2 User 3 User 4 Media Resources 1 1-1 1-2 1-3 1-4 Media Resources 2 2-1 2-2 2-3 2-4 Media Resources 3 3-1 3-2 3-3 3-4 Media Resources 4 4-1 4-2 4-3 4-4 Media Resources 5 5-1 5-2 5-3 5-4

[0063] According to the content shown in Table 2, user 1 may include 10 features, user 2 may include 8 features, user 3 may include 7 features, and user 4 may include 5 features. Among them, users 1 to 4 all include the three features of gender, age, and occupation.

[0064] Ten features of user 1 and N features of media resource 1 are used as input information and output to the target neural network model to obtain a prediction result, where N can include, but is not limited to, one or more. The prediction result indicates whether the user corresponding to the input information will click on the corresponding media resource, predicting that user 1 will click on media resource 1.

[0065] In this embodiment, three of the four users in the target user group will click on media resource 1. The target group tag is determined as the tag of the target media resource. Furthermore, the tag of media resource 1 is determined based on the users who like media resource 1. For example, if the target user group's group tags are Gender_Male, Life Status_Middle-Aged, and Occupation_Sales, this indicates that media resource 1 will be of interest to this type of user.

[0066] like Figure 5 As shown in FIG, a structural diagram of a method for determining a label of a media resource is shown, Figure 5 The specific application process is as follows.

[0067] Step 1: Obtain basic information of users in a historical time period, and form a first user set from the users in the historical time period;

[0068] Step 2: Cluster the first user set based on the user portrait features of each user in the first user set, group users with similar features into a group, and determine a target user group from the first user set. Since users with similar features are determined as the target group, a target group label for the target user group can be obtained;

[0069] Step 3: Combine each media resource in the media resource set and each user in the target user group into an input message, input this input message into the target neural network model, and obtain the prediction result output by the target neural network model, that is, whether the user in the target user group clicked on the media resource;

[0070] It should be noted that the first input information includes the user's user portrait features and resource content features. If the user's portrait features include 6 features, gender_male, age_30, occupation_game development engineer, education_graduate student, travel mode_private car, constellation_Taurus, these 6 features are input into the target neural network model. If the resource content features include 3 features, title_shooting game, category_game category, and theme topic_game skill analysis, these 3 features are input into the target neural network model.

[0071] One piece of input information of the input layer of the target neural network model is: Gender_Male, Age_30, Occupation_Game Development Engineer, Education_Graduate, Travel Mode_Private Car, Constellation_Taurus; Title_Shooting Game, Category_Game Category, Theme_Game Skill Analysis.

[0072] It should also be noted that the representation layer of the target neural network model is a dual-tower structure, that is, the user portrait features of the user correspond to the first neural network model, and the resource content features of the media resources correspond to the second neural network model.

[0073] Step 4: inputting daily input information from the plurality of input information consisting of media resources in the media resource set and users in the target user group into the target neural network model to obtain a set of prediction results;

[0074] Step 5: Determine the label information of the media resource based on a set of prediction results.

[0075] In step 6, the media resource can be pushed to users in the target user group based on the tag information of the media resource, and can also be pushed to other users. That is, the preference of the media resource can be determined based on the user's own information (attribute information), so that the media resource can be effectively pushed to the user, which can increase the effective exposure rate of the media resource.

[0076] like Figure 6 As shown in FIG, a schematic diagram of a scenario application of a method for determining a tag of a media resource is shown, Figure 6 As shown, the method for determining the label of media resources can be applied to various APPs, web pages, websites, etc., such as shopping APPs, instant messaging APPs, and short video APPs. APP can only use the basic attribute features of the user through the target neural network model (deep learning model) to represent the user portrait combination features into a vector, and use the resource content features of the media resource to represent it into a vector through the model. Through vector retrieval, the resource media (article) vector similar to the user vector is retrieved, and then the user portrait combination feature label is output to the media resource, and the label of the media resource is used to recall the user media resource.

[0077] It should be noted that in this embodiment, the preferences of users in the target user group can be determined only based on the user portrait characteristics (basic information) of each user in the target user group and the resource content characteristics of the media resources. In other words, it is also possible to determine what type of media resources are liked by which type of users, and then effectively push the media resources of interest to them.

[0078] Through the embodiments provided by the present application, by obtaining the user portrait features of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait features of each user are matched with the target group label; obtaining the resource content features of each media resource in the target media resource set; combining the user portrait features of each user and the resource content features of each media resource to form multiple input information, wherein each input information includes the user portrait features of a user and the resource content features of a media resource; inputting the multiple input information into the target neural network model respectively, and obtaining multiple prediction results output by the target neural network model, wherein each prediction result is The invention is used to indicate whether the user corresponding to an input information will click on the corresponding media resource; based on multiple prediction results, the target media resource matching the target user group is determined in the target media resource set, and the target group label is determined as the label of the target media resource, thereby achieving the purpose of determining the media resources preferred by this type of users in the target user group based on the user portrait characteristics (basic information of the user) and the resource content characteristics of the media resource of each user in the target user group. According to the label of the media resource, the user who prefers the media resource can be known, and the media resource can be effectively pushed to the user, thereby solving the technical problem in the existing technology that the media resource cannot be effectively pushed to the user.

[0079] Optionally, in this embodiment, before obtaining the user portrait features of each user in the target user group, the above may also include: obtaining the user portrait features of each user in the first user set, wherein the user portrait features are used to represent the user's own information; and determining the target user group in the first user set based on the preset target group label and the user portrait features of each user in the first user set.

[0080] In this embodiment, users matching the preset target group label are selected from the first user set. That is, users matching the target group label are grouped into a user group, thus obtaining a target user group.

[0081] In this embodiment, the preset target user group label can be manually determined. For example, if the target user group label is: gender_male, age_23, occupation_student, then users matching the target group label are selected from the first user set according to the target group label to obtain the target user group.

[0082] Optionally, determining the target user group in the first user set based on a preset target group label and user portrait features of each user in the first user set may include: searching for the target user in the first user set, wherein the target group label includes M labels, the user portrait features of the target user include M features corresponding to the M labels, the values ​​of the M features match the values ​​of the M labels, and M is 1 or 2 or a natural number greater than or equal to 3; when the target user is found, determining that the user portrait features of the target user match the target group label, and determining the target user as a user in the target user group.

[0083] In this embodiment, if M is 3, the target group tag includes three tags: gender, age, and occupation. Users corresponding to the characteristic gender, age, and occupation are searched in the first user set, and the values ​​of the M features match the values ​​of the M tags.

[0084] Among them, searching for the target user in the first user set may include: corresponding to each user in the first user set, performing the following operations, wherein each user is the current user in the following operations: obtaining the values ​​of the current M features corresponding to the M tags in the user portrait features of the current user; when the values ​​of the current M features are equal to the values ​​of the M tags, or are within the value range of the M tags, determining that the values ​​of the current M features match the values ​​of the M tags, and determining the current user as the target user.

[0085] In this embodiment, based on the M features of the target group label, each user in the first user set is traversed to obtain users whose user portrait features match the M features, and the target user group is determined based on the matching users.

[0086] Optionally, in this embodiment, the above method may further include: determining M features based on the user portrait features of each user in the first user set, wherein the target group label includes M labels, and M is 1 or a natural number greater than or equal to 2; setting the M labels to M features, wherein the values ​​or value ranges of the M labels are preset values, or are determined based on the values ​​of the M features.

[0087] Among them, determining M features based on the user portrait features of each user in the first user set can include: determining a first feature combination set based on the user portrait features of each user in the first user set, wherein the number of features in each feature combination in the first feature combination set is M; determining a second feature combination set based on a pre-acquired click log, wherein the number of features in each feature combination in the second feature combination set is M, and the click log records the user portrait features of each user in the second user set; and determining a feature combination in the intersection of the first feature combination set and the second feature combination set as the M features.

[0088] In this embodiment, the first user set includes five users, user a, user b, user c, user d, and user e. Each user may include the same feature tag or a different feature tag, as shown in Table 3. The feature tags of each user are the same, but the feature values ​​may be the same or different.

[0089] Table 3

[0090] User A Gender_Male Age_30 Occupation_Xa Place of residence_ya Education_Ca User B Gender_Female Age_25 Occupation_Xb Place of residence_yb Education_Cb User c Gender_Male Age_35 Occupation_Xc Place of residence_yc Travel Mode_Cc User d Gender_Male Age_30 Occupation_Xd Place of residence_yd Educational background_Cd User e Gender_Female Age_40 Occupation_Xe Place of residence_ye Education_Ce

[0091] As shown in Table 3, a set of feature sets may include: combining any M features with feature labels such as gender, age, occupation, place of residence, and educational background to obtain a first feature combination set. For example, for any 3 feature combinations, 120 ternary feature groups can be calculated in the first feature group set based on permutations and combinations.

[0092] In this embodiment, a second feature combination set can be derived from the user profile features of each user in the second user set. Each feature group in the second feature combination set is also composed of three features, a ternary feature group. This ternary feature group is determined based on the click logs of each user in the second user set. For example, if seven users click on media resource m, after ten users click on media resource m, seven click logs corresponding to these seven users will be generated. From these seven click logs, the ternary features of each user can be accurately obtained.

[0093] It should be noted that after obtaining the first feature combination set and the second feature combination set, a feature combination may be determined in the intersection of the first feature combination set and the second feature combination set as the M features.

[0094] like Figure 7 As shown in Figure 1, the structural diagram of the target user group is determined by the combination of characteristics. Figure 7 As shown, the feature combination of the target user group is the intersection of the feature combinations determined by the two methods.

[0095] Typically, the first feature combination set is obtained by enumerating all possible combinations of user portrait features. Some illegal combinations may appear in the first feature combination set, such as "gender_female, age_45, life status_marriageable", where age and life status conflict with each other.

[0096] The second feature combination set is obtained by counting the actual feature combinations from the click logs and taking the top N combinations to form the label set. However, some of the counted combinations contain significant features and need to be filtered out.

[0097] In this embodiment, the intersection of the first feature combination set and the second feature combination set is used as the final label set, and a feature combination can be determined from the label set.

[0098] After obtaining the tag set, the user tower model is used for inference to obtain a vector for each profile combination in the set. Media resources are then processed through the media resource tower to obtain a media resource vector. Using the user vector, similar media resource vectors are retrieved to obtain articles that match the preferences of the group. The profile feature combination and label combination are then output to the article.

[0099] Optionally, in this embodiment, multiple input information are respectively input into the target neural network model to obtain multiple prediction results output by the target neural network model, which may include: for each input information, executing the following steps to obtain the corresponding prediction result, wherein, when executing the following steps, each input information is the current input information, and the current input information includes the user portrait features of the current user and the resource content features of the current media resource: inputting the user portrait features of the current user into the first neural network model to obtain the user vector of the current user output by the first neural network model; inputting the resource content features of the current media resource into the second neural network model to obtain the media resource vector of the current media resource output by the second neural network model, wherein the target neural network model includes the first neural network model and the second neural network model; determining the current prediction result based on the user vector and the media resource vector, wherein the current prediction result is used to indicate whether the current user will click on the current media resource when the current media resource is pushed to the current user.

[0100] Determining the current prediction result according to the user vector and the media resource vector may include: determining a vector distance between the user vector and the media resource vector; and determining the current prediction result according to the vector distance.

[0101] In this embodiment, as shown in Table 2, each of the multiple input information includes the user portrait features of the user and the resource content features of the current media resource, and each input information is input into the target neural network to obtain multiple prediction results output by the target neural network model, where each preset result corresponds to one input information.

[0102] It should be noted that, in this embodiment, Figure 8 As shown in FIG, a schematic diagram of the structure of the target neural network model, the target neural network model structure adopts the structure of the DSSM (Deep Structured Semantic Model) deep semantic matching model, which includes an input layer, a representation layer, and a matching layer.

[0103] In this embodiment, the representation layer in the DSSM model is a dual-tower model, which can include two independent network models: a first neural network model (user tower) and a second neural network model (media resource tower). The first neural network model is used to obtain user feature vectors based on user profile features, and the second neural network model is used to obtain media resource vectors based on resource content features of media resources. The specific implementation process of this model is as follows.

[0104] Step S41: Input user portrait features and media resource content features to both sides of the input layer.

[0105] Step S42: obtaining a user feature vector / media resource vector after learning at the representation layer;

[0106] Step S43: Calculate the distance between the vectors at the matching layer, and finally determine the label of the media resource.

[0107] Optionally, determining a target media resource matching the target user group in the target media resource set based on multiple prediction results may include:

[0108] S1, determining a group of prediction results corresponding to the target media resource from a plurality of prediction results, wherein the group of prediction results is prediction results obtained by respectively using resource content features of the target media resource and user profile features of each user in the target user group as input information;

[0109] S2. When the number of target prediction results in a group of prediction results is greater than or equal to a preset threshold, determine that the target media resource matches the target user group, wherein the target prediction result is used to indicate that users in the target user group will click on the target media resource.

[0110] Optionally, the above method may also include: obtaining user portrait features of each user in the first user set, wherein the user portrait features are used to represent the user's own information; dividing the first user set into multiple user groups based on multiple group tags obtained in advance and the user portrait features of each user in the first user set, wherein each user group corresponds to a group tag, and the user portrait features of the users in each user group are matched with a corresponding group tag, and the multiple user groups include the target user group.

[0111] Optionally, the above method may further include: pushing the target media resource to a third user set, wherein the user portrait features of each user in the third user set match the label of the target media resource, and the label of the target media resource includes a target group label.

[0112] Optionally, the above method may also include: searching for a third user set, wherein the target group label includes M labels, the user portrait features of each user in the third user set include M features corresponding to the M labels, the values ​​of the M features match the values ​​of the M labels, and M is 1 or 2 or a natural number greater than or equal to 3.

[0113] In this embodiment, after obtaining the tag of the media resource, the media resource is pushed to the user matching the tag according to the tag, thereby effectively recommending the media resource that the user is interested in to the user.

[0114] As an optional embodiment, a method for determining article tags is described by taking an article resource as an example. The article tag determination process may include two parts: a target neural network training process and a target neural network model use process.

[0115] In this embodiment, the structure of the target neural network model uses the DSSM model framework, and the model structure is as follows: Figure 8 As shown in the figure. After training, the DSSM model can produce two independent network models, namely the first neural network model and the second neural network model. The first neural network model and the second neural network model are used to obtain user / article vectors respectively. The DSSM model has a three-layer structure. The user features and article features are input on both sides of the input layer. After learning in the representation layer, the user / article vectors are obtained. The distance between the vectors is calculated in the matching layer. Finally, the cross-entropy loss is calculated. When the cross-entropy loss value meets the preset conditions, the target neural network model is obtained.

[0116] In this embodiment, Figure 9 The specific structure of the target neural network model is shown in Figure 2. In order to enhance the representation learning ability of the representation layer, a display feature cross network CN (Cross Network) is added to the representation layer. Figure 9As shown in the figure, the representation layer includes a deep network and a cross network. The cross network can realize feature crossover of the bit dimension of the vector, and the order of crossover is determined by the number of layers of the network.

[0117] In this embodiment, Figure 10 The recursive process of the cross network is shown, where the output x of the k+1 layer is k+1 (For example, Figure 10 The y in the k layer is the output x k (For example, Figure 10 The residual mechanism of this formula ensures that the cross-features of the next layer contain the cross-features of the current layer, so the final output vector contains cross-features of all orders.

[0118] refer to Figure 10 The recursive relationship in the cross network is shown in the following formula (1):

[0119]

[0120] In this embodiment, the training data for the target neural network model uses real online production data. A training data sample consists of user profile features and content comprehension features of the clicked article. User profile features include 26 demographic attributes such as age, gender, occupation, zodiac sign, and province. Article features include title, category, tag, and topic. Training data preprocessing consists of the following two steps.

[0121] Training sample construction involves abstracting crowd preference mining into a classification task. An article clicked in a click log is used as a positive sample in the training data. K negative samples need to be constructed for this training data. Negative sample construction can be done in the following three ways:

[0122] Method a: Randomly sample K articles from the positive article library as negative examples.

[0123] Method b: negative sample sampling is performed according to the click frequency of the article, as shown in the sampling formula (2):

[0124]

[0125] Method c: Articles clicked by other users in the same batch are used as negative samples of the current user. Compared with method b, this method reduces the redundancy of repeatedly calculating the article vectors.

[0126] Among them, due to the large differences in click distribution of different articles, some popular articles are clicked more times. If the positive samples are not sampled, the recalled articles will be biased towards the popular ones. The formula (3) for positive sample sampling is as follows: ran = ((x / sample_rate)^0.5+1)*(sample_rate / x)(3)

[0127] Among them, ran is the probability of the article being sampled, x is the click frequency of the article if it is sampled according to the article dimension, and the click frequency of the third-level channel if it is sampled according to the category. sample_rate is a sampling hyperparameter. The smaller the sample_rate value, the greater the penalty for articles with high clicks. Figure 11 As shown, the function image diagram of negative sample sampling is shown in Figure 11 The function graph of the sampling formula sample_rate when it takes different values. The horizontal axis is the click frequency of the article, and the vertical axis is the probability of the article being sampled.

[0128] A tag set is constructed, where each tag in the tag set corresponds to a feature combination, and the feature corresponds to the user portrait feature of the user, that is, the user group tag of the target user group can be obtained.

[0129] After obtaining the user / article representation model from the DSSM model, the user vector can be obtained by inputting the user features, and the article vector can be obtained by inputting the article features. Using the user vector to retrieve similar article vectors, a set of articles preferred by the user group can be obtained. The user group combination features are output to the article, which is the group preference of the article. Therefore, it is necessary to construct a combination feature label set. The combination feature is to arrange and combine k independent features together to form a label set. The more features in the combination, the more restrictions there are, and the fewer people can be covered. In order to expand the population that each feature combination can cover, in this embodiment, the combination features are all ternary combinations, and the combination feature form is "age, gender, X", where X represents a feature other than age or gender.

[0130] Significant feature screening: Before constructing the ternary combination feature, it is necessary to clarify which features can significantly distinguish between groups of people, that is, different values ​​of the features correspond to groups of people with different preferences. For some features, the articles liked by user groups corresponding to different values ​​are not significantly different, such as "zodiac sign, education level", so this feature cannot be used to construct a combination. When analyzing the significance of features, a mask strategy is used. When predicting, the features of the portrait to be analyzed in the test set are masked separately, and the performance of the model on the test set is compared. From the 26 features, 10 features with significant group preferences are selected, such as Figure 12 As shown, a schematic diagram of the salient features is shown in Figure 12The circled features are the significant features (salient features in user portrait features), including age, occupational segmentation, rural population, gender, province, car owners, parenting, life status, college population, and high-profile population.

[0131] Constructing feature combination labels (that is, determining user group labels) can include the following methods:

[0132] The first method is to enumerate all possible feature combinations. However, some illegal combinations may appear in the feature combinations obtained in this way, such as "gender_female, age_45, life status_marriageable", because age and life status conflict with each other.

[0133] The second method is to count the actual combinations from the training data and select the top N combinations to form the label set. However, some of the counted combinations contain significant features and need to be filtered out.

[0134] In this embodiment, the intersection of the feature combinations obtained by method 1 and method 2 is used as the final label set, such as Figure 13 As shown in , the schematic diagram of constructing feature combination labels. Figure 13 As shown, through method one, we can get the feature combination set list on the left, that is, the constructed combination portrait, and through method two, we can get the feature combination set list on the right, that is, the statistical combination portrait. The intersection of the constructed combination portrait and the statistical combination portrait is taken as the final label set.

[0135] After obtaining the tag set, the user tower model is used for inference to obtain the vectors for each profile combination in the set. The positively ranked articles are then processed through the article tower to obtain the article vectors. Using the user vectors, similar article vectors are retrieved to find articles that match the preferences of the group. The profile combination and label combination are then output to the article.

[0136] In this embodiment, a deep learning model can be trained using only basic user attribute features. The model learns independent user / article representation models, represents the user profile combined features as a vector, and uses the article content understanding features to represent them as a vector through the model. Through vector retrieval, article vectors similar to the user vector are retrieved, and the user profile combined feature labels are then output to the articles for user article recall.

[0137] In this embodiment, only basic user attribute features are used to train the deep learning model. The model learns independent user / article representation models, represents the user profile combined features as a vector, and uses the article content understanding features to represent them as a vector. Through vector retrieval, article vectors similar to the user vector are retrieved, and the user profile combined feature labels are output to the articles for user article recall.

[0138] It should also be noted that the use process of the target neural network model is the same as the training process of the above-mentioned target neural network model, and the use process of the target neural network model will not be repeated here.

[0139] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0140] According to another aspect of the embodiments of the present invention, there is also provided a device for determining a label of a media resource for implementing the above-mentioned method for determining a label of a media resource. Figure 14 As shown, the apparatus for determining a label of a media resource includes: a first obtaining unit 1401 , a second obtaining unit 1403 , a composition unit 1405 , an output unit 1407 , and a first determining unit 1409 .

[0141] The first acquiring unit 1401 is configured to acquire a user portrait feature of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait feature of each user matches the target group label.

[0142] The second acquiring unit 1403 is configured to acquire resource content features of each media resource in the target media resource set.

[0143] The composition unit 1405 is used to combine the user portrait characteristics of each user and the resource content characteristics of each media resource into multiple input information, wherein each input information includes the user portrait characteristics of a user and the resource content characteristics of a media resource.

[0144] The output unit 1407 is used to input multiple input information into the target neural network model respectively to obtain multiple prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to a piece of input information will click on the corresponding media resource.

[0145] The first determining unit 1409 is configured to determine, based on the plurality of prediction results, a target media resource that matches the target user group in the target media resource set, and determine the target group label as a label of the target media resource.

[0146] Through the embodiment provided by the present application, the first acquisition unit 1401 acquires the user portrait features of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait features of each user match the target group label; the second acquisition unit 1403 acquires the resource content features of each media resource in the target media resource set; the composition unit 1405 combines the user portrait features of each user and the resource content features of each media resource to form multiple input information, wherein each input information includes the user portrait features of a user and the resource content features of a media resource; the output unit 1407 inputs the multiple input information into the target neural network model respectively to obtain multiple prediction results output by the target neural network model, wherein each prediction result is used to indicate whether the user corresponding to an input information will click on the corresponding media resource; the first determination unit 1409 determines the target media resource that matches the target user group in the target media resource set based on the multiple prediction results, and determines the target group label as the label of the target media resource. The purpose of determining the media resources preferred by the target user group based on the user portrait characteristics (basic information of the user) and the resource content characteristics of the media resources of each user in the target user group is achieved. According to the label of the media resource, the users who prefer the media resource can be known, and the media resource can be effectively pushed to the user, thereby solving the technical problem in the existing technology that the media resources cannot be effectively pushed to the user.

[0147] Optionally, the above-mentioned device may also include: a third acquisition unit, used to obtain user portrait features of each user in the first user set, wherein the user portrait features are used to represent the user's own information; a second determination unit, used to determine the target user group in the first user set based on a preset target group label.

[0148] Among them, the above-mentioned second determination unit may include: a search module, used to search for the target user in the first user set, wherein the target group label includes M labels, the user portrait features of the target user include M features corresponding to the M labels, the values ​​of the M features match the values ​​of the M labels, and M is 1 or a natural number greater than or equal to 2; the first determination module is used to determine that the user portrait features of the target user match the target group label when the target user is found, and determine the target user as a user in the target user group.

[0149] Among them, the above-mentioned search module can be used to perform the following operations: corresponding to each user in the first user set, perform the following operations, wherein, in the following operations, each user is the current user: obtain the values ​​of the current M features corresponding to the M tags in the user portrait features of the current user; when the values ​​of the current M features are equal to the values ​​of the M tags, or are within the value range of the M tags, determine that the values ​​of the current M features match the values ​​of the M tags, and determine the current user as the target user.

[0150] Optionally, the above-mentioned device may also include: a third determination unit, used to determine M features based on the user portrait features of each user in the first user set, wherein the target group label includes M labels, and M is 1 or a natural number greater than or equal to 2; a setting unit, used to set the M labels to M features, wherein the values ​​or value ranges of the M labels are preset values, or are determined based on the values ​​of the M features.

[0151] Among them, the above-mentioned third determination unit may include: a second determination module, used to determine the first feature combination set based on the user portrait features of each user in the first user set, wherein the number of features in each feature combination in the first feature combination set is M; a third determination module, used to determine the second feature combination set based on the pre-acquired click log, wherein the number of features in each feature combination in the second feature combination set is M, and the click log records the user portrait features of each user in the second user set; a fourth determination module, used to determine a feature combination in the intersection of the first feature combination set and the second feature combination set as M features.

[0152] Optionally, the above-mentioned output unit can also perform the following operations: for each piece of input information, perform the following steps to obtain a corresponding prediction result, wherein, when performing the following steps, each piece of input information is current input information, and the current input information includes the user portrait features of the current user and the resource content features of the current media resource: input the user portrait features of the current user into the first neural network model to obtain the user vector of the current user output by the first neural network model; input the resource content features of the current media resource into the second neural network model to obtain the media resource vector of the current media resource output by the second neural network model, wherein the target neural network model includes the first neural network model and the second neural network model; determine the current prediction result based on the user vector and the media resource vector, wherein the current prediction result is used to indicate whether the current user will click on the current media resource when the current media resource is pushed to the current user.

[0153] The output unit may further perform the following operations: determining a vector distance between the user vector and the media resource vector; and determining a current prediction result based on the vector distance.

[0154] Optionally, the above-mentioned first determination unit may include: determining a group of prediction results corresponding to the target media resource from multiple prediction results, wherein a group of prediction results are prediction results obtained by taking the resource content characteristics of the target media resource and the user portrait characteristics of each user in the target user group as input information; when the number of target prediction results in a group of prediction results is greater than a preset threshold, determining that the target media resource matches the target user group.

[0155] Optionally, the above-mentioned device may also include: a fourth acquisition unit, used to obtain user portrait features of each user in the first user set, wherein the user portrait features are used to represent the user's own information; a division unit, used to divide the first user set into multiple user groups based on multiple group labels obtained in advance and the user portrait features of each user in the first user set, wherein each user group corresponds to a group label, and the user portrait features of the users in each user group satisfy a corresponding group label, and the multiple user groups include the target user group.

[0156] Optionally, the above-mentioned device may also include: a pushing unit, used to push the target media resource to a third user set, wherein the user portrait characteristics of each user in the third user set meet the label of the target media resource, and the label of the target media resource includes a target group label.

[0157] Optionally, the above-mentioned device may also include: a search unit for searching a third user set, wherein the target group label includes M labels, the user portrait features of each user in the third user set include M features corresponding to the M labels, the values ​​of the M features match the values ​​of the M labels, and M is 1 or a natural number greater than or equal to 2.

[0158] According to another aspect of the embodiment of the present invention, an electronic device for implementing the above-mentioned method for determining a label of a media resource is also provided. The electronic device may be Figure 1 The terminal device or server shown in FIG. This embodiment is described by taking the electronic device as a server as an example. Figure 15 As shown, the electronic device includes a memory 1502 and a processor 1504. The memory 1502 stores a computer program, and the processor 1504 is configured to execute the steps in any of the above method embodiments through the computer program.

[0159] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0160] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0161] S1, obtaining the user portrait features of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait features of each user are matched with the target group label;

[0162] S2, obtaining resource content features of each media resource in the target media resource set;

[0163] S3, combining the user profile features of each user and the resource content features of each media resource into multiple pieces of input information, where each piece of input information includes the user profile features of a user and the resource content features of a media resource;

[0164] S4, inputting the plurality of input information into the target neural network model respectively, and obtaining a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether a user corresponding to a piece of input information will click on a corresponding media resource;

[0165] S5 , determining a target media resource that matches the target user group in the target media resource set based on the multiple prediction results, and determining the target group label as a label of the target media resource.

[0166] Alternatively, those skilled in the art will appreciate that Figure 15 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 15 It does not limit the structure of the electronic device. For example, the electronic device may also include Figure 15 More or fewer components (such as network interfaces, etc.) as shown in, or with Figure 15 Different configurations shown.

[0167] Among them, the memory 1502 can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for determining the label of media resources in the embodiment of the present invention. The processor 1504 executes various functional applications and data processing by running the software programs and modules stored in the memory 1502, that is, realizing the above-mentioned method for determining the label of media resources. The memory 1502 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1502 may further include a memory remotely located relative to the processor 1504, and these remote memories can be connected to the terminal via a network. Examples of the above-mentioned networks include but are not limited to the Internet, corporate intranets, local area networks, mobile communication networks and combinations thereof. Among them, the memory 1502 can be used to store, but is not limited to, user portrait features, resource content features, target group labels and other information of the user. As an example, such as Figure 15 As shown, the memory 1502 may include, but is not limited to, the first acquisition unit 1401, the second acquisition unit 1403, the composition unit 1405, the output unit 1407, and the first determination unit 1409 in the apparatus for determining a label for a media resource. Furthermore, the memory 1502 may also include, but is not limited to, other module units in the apparatus for determining a label for a media resource, which will not be described in detail in this example.

[0168] Optionally, the transmission device 1506 is used to receive or send data via a network. Specific examples of the network may include a wired network and a wireless network. In one embodiment, the transmission device 1506 includes a network interface controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In one embodiment, the transmission device 1506 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0169] In addition, the electronic device further includes: a display 1508 for displaying the media resources; and a connection bus 1510 for connecting various module components in the electronic device.

[0170] In other embodiments, the terminal device or server may be a node in a distributed system, wherein the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting multiple nodes through network communication. The nodes may form a peer-to-peer (P2P) network, and any computing device, such as a server, terminal, or other electronic device, may become a node in the blockchain system by joining the peer-to-peer network.

[0171] According to one aspect of the present application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method for determining a label for a media resource provided in the aforementioned aspect of determining a label for a media resource or in various optional implementations of the aspect of determining a label for a media resource. The computer program is configured to execute the steps of any of the aforementioned method embodiments when executed.

[0172] Optionally, in this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0173] S1, obtaining the user portrait features of each user in the target user group, wherein the group label of the target user group is the target group label, and the user portrait features of each user are matched with the target group label;

[0174] S2, obtaining resource content features of each media resource in the target media resource set;

[0175] S3, combining the user profile features of each user and the resource content features of each media resource into multiple pieces of input information, where each piece of input information includes the user profile features of a user and the resource content features of a media resource;

[0176] S4, inputting the plurality of input information into the target neural network model respectively, and obtaining a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether a user corresponding to a piece of input information will click on a corresponding media resource;

[0177] S5 , determining a target media resource that matches the target user group in the target media resource set based on the multiple prediction results, and determining the target group label as a label of the target media resource.

[0178] Optionally, in this embodiment, a person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing the hardware related to the terminal device through a program, and the program may be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0179] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0180] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0181] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0182] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, and can be electrical or other forms.

[0183] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0184] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0185] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for determining a label of a media resource, characterized in that: include: Obtaining a user portrait feature of each user in a target user group, wherein the group label of the target user group is a target group label, and the user portrait feature of each user matches the target group label; Obtain resource content characteristics of each media resource in the target media resource collection; Combining the user portrait feature of each user and the resource content feature of each media resource into multiple pieces of input information, wherein each piece of input information includes the user portrait feature of one user and the resource content feature of one media resource; Inputting the plurality of input information into a target neural network model respectively to obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether a user corresponding to a piece of input information will click on a corresponding media resource; According to the multiple prediction results, a target media resource matching the target user group is determined in the target media resource set, and the target group label is determined as a label of the target media resource.

2. The method according to claim 1, characterized in that Before obtaining the user portrait features of each user in the target user group, the method further includes: Obtaining a user portrait feature of each user in the first user set, wherein the user portrait feature is used to represent the user's own information; The target user group is determined in the first user set according to the preset target group label and the user portrait feature of each user in the first user set.

3. The method according to claim 2, characterized in that The determining the target user group in the first user set according to the preset target group label and the user portrait feature of each user in the first user set includes: Searching for a target user in the first user set, wherein the target group tag includes M tags, the user profile features of the target user include M features corresponding to the M tags, the values ​​of the M features match the values ​​of the M tags, and M is 1, 2, or a natural number greater than or equal to 3; When the target user is found, it is determined that the user portrait feature of the target user matches the target group label, and the target user is determined as a user in the target user group.

4. The method according to claim 3, characterized in that The searching for a target user in the first user set includes: For each user in the first user set, perform the following operations, wherein each user is a current user in the following operations: Obtaining values ​​of the current M features corresponding to the M tags from the user portrait features of the current user; When the values ​​of the current M features are equal to the values ​​of the M tags, or are within the value range of the M tags, it is determined that the values ​​of the current M features match the values ​​of the M tags, and the current user is determined as the target user.

5. The method according to claim 2, characterized in that The method further comprises: Determining M features based on the user portrait features of each user in the first user set, wherein the target group label includes M labels, and M is 1, 2, or a natural number greater than or equal to 3; The M tags are set as the M features, wherein the values ​​or value ranges of the M tags are preset values, or are determined based on the values ​​of the M features.

6. The method according to claim 5, characterized in that The determining M features according to the user portrait features of each user in the first user set includes: Determine a first feature combination set based on the user portrait feature of each user in the first user set, wherein the number of features in each feature combination in the first feature combination set is M; Determine a second feature combination set based on the pre-acquired click log, wherein the number of features in each feature combination in the second feature combination set is M, and the click log records the user portrait features of each user in the second user set; A feature combination is determined in the intersection of the first feature combination set and the second feature combination set as the M features.

7. The method according to claim 1, characterized in that The step of inputting the plurality of pieces of input information into the target neural network model to obtain a plurality of prediction results output by the target neural network model includes: For each piece of input information, perform the following steps to obtain a corresponding prediction result. When performing the following steps, each piece of input information is current input information, and the current input information includes the user profile features of the current user and the resource content features of the current media resource: Inputting the user portrait feature of the current user into a first neural network model to obtain a user vector of the current user output by the first neural network model; Inputting the resource content feature of the current media resource into a second neural network model to obtain a media resource vector of the current media resource output by the second neural network model, wherein the target neural network model includes the first neural network model and the second neural network model; A current prediction result is determined according to the user vector and the media resource vector, wherein the current prediction result is used to indicate whether the current user will click on the current media resource when the current media resource is pushed to the current user.

8. The method according to claim 7, characterized in that The determining a current prediction result according to the user vector and the media resource vector includes: determining a vector distance between the user vector and the media resource vector; The current prediction result is determined according to the vector distance.

9. The method according to any one of claims 1 to 8, characterized in that Determining, in the target media resource set according to the multiple prediction results, a target media resource that matches the target user group includes: Determining a group of prediction results corresponding to the target media resource from the multiple prediction results, wherein the group of prediction results is prediction results obtained by respectively using resource content features of the target media resource and user profile features of each user in the target user group as input information; When the number of target prediction results in the set of prediction results is greater than or equal to a preset threshold, it is determined that the target media resource matches the target user group, wherein the target prediction result is used to indicate that users in the target user group will click on the target media resource.

10. The method according to any one of claims 1 to 8, characterized in that The method further comprises: Obtaining a user portrait feature of each user in the first user set, wherein the user portrait feature is used to represent the user's own information; Based on multiple pre-acquired group tags and user portrait features of each user in the first user set, the first user set is divided into multiple user groups, wherein each user group corresponds to a group tag, and the user portrait features of the users in each user group match the corresponding group tag, and the multiple user groups include the target user group.

11. The method according to any one of claims 1 to 8, characterized in that The method further comprises: The target media resource is pushed to a third user set, wherein the user portrait feature of each user in the third user set matches the tag of the target media resource, and the tag of the target media resource includes the target group tag.

12. The method according to claim 11, characterized in that The method further comprises: Search the third user set, wherein the target group label includes M labels, the user portrait features of each user in the third user set include M features corresponding to the M labels, the values ​​of the M features match the values ​​of the M labels, and M is 1 or 2 or a natural number greater than or equal to 3.

13. A device for determining a label of a media resource, characterized in that: include: A first acquiring unit is configured to acquire a user portrait feature of each user in a target user group, wherein the group label of the target user group is a target group label, and the user portrait feature of each user matches the target group label; A second acquisition unit is used to acquire resource content characteristics of each media resource in the target media resource set; a composition unit, configured to combine the user portrait feature of each user and the resource content feature of each media resource into a plurality of pieces of input information, wherein each piece of input information includes the user portrait feature of one user and the resource content feature of one media resource; an output unit, configured to input the plurality of input information into a target neural network model respectively, and obtain a plurality of prediction results output by the target neural network model, wherein each prediction result is used to indicate whether a user corresponding to a piece of input information will click on a corresponding media resource; A first determining unit is configured to determine, in the target media resource set, a target media resource that matches the target user group according to the multiple prediction results, and determine the target group label as a label of the target media resource.

14. A computer-readable storage medium comprising a stored program, wherein: When the program is executed, the method described in any one of claims 1 to 12 is executed.

15. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 12 through the computer program.

Images