Content pushing method and device, equipment, readable storage medium and program product

Abstract

The invention provides a content pushing method and device, equipment, a readable storage medium and a program product. The method comprises the following steps: receiving a content push request carrying a login object sent by a game client; obtaining object data including basic data and game data of the login object; performing feature extraction processing on the object data to obtain feature information of the login object; performing object type prediction processing according to the feature information to obtain an object type prediction probability of the login object, the object type prediction probability being a probability that the object type of the login object is a target object type; the probability that the object of the target object type executes conversion operation on the push content is greater than or equal to a first probability threshold; and performing content pushing processing on the login object according to the object type prediction probability. By adopting the embodiment of the invention, flexible content pushing processing can be performed according to the object data of the login object, personalized content pushing is realized, and improvement of the conversion rate of the pushed content is facilitated.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a content push method, a content push device, a computer device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] To enhance the user experience, game servers can push specific content to game clients, allowing game clients to quickly learn specific information or perform specific actions based on the pushed content.

[0003] Content delivery strategies typically involve scheduled pushes or event-specific pushes. Scheduled pushes send content to logged-in users on the game client at fixed intervals, such as sending a content push to all logged-in users weekly. Event-specific pushes send content after a logged-in user completes a specific event, such as sending a content push after each login. However, among all logged-in users, some have lower demand for the pushed content and are less likely to perform conversion actions based on it, resulting in lower conversion rates for the pushed content under both of these strategies. Summary of the Invention

[0004] This application provides a content push method, apparatus, device, readable storage medium, and program product, which can flexibly push content based on the object data of the logged-in object, realize personalized content push, and help improve the conversion rate of the pushed content.

[0005] On the one hand, embodiments of this application provide a content push method, which includes:

[0006] Receive a content push request sent by the game client, the content push request carrying a login object, the login object being the object currently logged into the game client;

[0007] In response to the content push request, the object data of the logged-in object is obtained. The object data includes basic data and game data. The game data includes one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0008] The object data is subjected to feature extraction processing to obtain the feature information of the login object;

[0009] Based on the feature information, object type prediction processing is performed to obtain the object type prediction probability of the login object. The object type prediction probability is the probability that the object type of the login object is the target object type. The probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0010] Content is pushed to the logged-in object based on the predicted probability of the object type.

[0011] Accordingly, embodiments of this application provide a content push device, which includes:

[0012] A receiving unit is used to receive a content push request sent by a game client, wherein the content push request carries a login object, and the login object is the object currently logged into the game client.

[0013] The acquisition unit is configured to, in response to the content push request, acquire object data of the logged-in object, wherein the object data includes basic data and game data, and the game data includes one or more of game scene description data, game scene statistics data, and game object description data;

[0014] The processing unit is used to perform feature extraction processing on the object data to obtain the feature information of the logged-in object;

[0015] The processing unit is further configured to perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, wherein the object type prediction probability is the probability that the object type of the login object is the target object type; the probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0016] The processing unit is also used to perform content push processing on the logged-in object based on the probability predicted by the object type.

[0017] In one embodiment, when the processing unit performs content push processing on the logged-in object based on the predicted probability of the object type, it specifically performs the following steps:

[0018] The predicted probability of the object type is compared with the second probability threshold corresponding to the login object;

[0019] If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then the target push content associated with the game is determined, and the target push content is pushed to the logged-in object;

[0020] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then no content push processing will be performed on the logged-in object.

[0021] In one embodiment, the processing unit is further configured to perform the following steps:

[0022] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then the push frequency of the login object is determined based on the number of times content is pushed to the login object within the target time.

[0023] The push frequency of the logged-in object is compared with the push frequency threshold to obtain the push frequency comparison result;

[0024] If the push frequency comparison result indicates that the push frequency of the logged-in object is less than or equal to the push frequency threshold, then the second probability threshold is reduced to obtain a reduced second probability threshold.

[0025] In one embodiment, the basic data includes target location data. When the processing unit performs feature extraction processing on the object data to obtain the feature information of the logged-in object, it specifically performs the following steps:

[0026] Obtain location mapping information, which indicates the linear mapping relationship between location data and location feature values ​​under the location feature dimension;

[0027] Based on the target location data, query the location mapping information to determine the location feature value under the location feature dimension corresponding to the target location data.

[0028] In one embodiment, the game data includes game scene description data, which includes the target game duration; when the processing unit performs feature extraction processing on the object data to obtain the feature information of the logged-in object, it specifically performs the following steps:

[0029] Obtain the duration mapping polynomial, which is used to indicate the nonlinear mapping relationship between game duration and duration feature values ​​under the game duration feature dimension;

[0030] Using the duration mapping polynomial, the target game duration is nonlinearly mapped to obtain the duration feature value under the game duration feature dimension corresponding to the target game duration.

[0031] In one embodiment, when the processing unit performs object type prediction processing based on the feature information to obtain the object type prediction probability of the logged-in object, it specifically performs the following steps:

[0032] The feature information is input into the object type prediction model for object type prediction processing to obtain the object type prediction probability of the login object.

[0033] The object type prediction model is obtained by adjusting the initial type prediction model using difference data. The difference data is determined based on the sample label and sample probability of the sample feature information. The sample label is used to indicate the probability that the object type of the sample object is the target object type. The sample object is the object corresponding to the sample feature information. The sample probability is obtained by performing object type prediction processing on the sample feature information using the initial type prediction model. The sample feature information is determined based on the object data of the sample object.

[0034] In one embodiment, the processing unit is further configured to perform the following steps:

[0035] Obtain a sample dataset, which includes multiple sample data. The sample data includes sample feature information and sample labels for the sample feature information. The sample feature information with positive sample labels is determined based on the object data of positive sample objects. The sample feature information with negative sample labels is determined based on the object data of negative sample objects. The positive sample objects are those for which no conversion operation was performed on the push content in the first time period and which was performed on the push content in the second time period. The negative sample objects are those for which a conversion operation was performed on the push content in both the first and second time periods, with the first time period being earlier than the second time period.

[0036] The sample feature information included in each sample data in the sample dataset is input into the initial type prediction model for object type prediction processing to obtain the sample probability corresponding to each sample feature information;

[0037] The difference data is determined based on the sample probability corresponding to each of the sample feature information and the sample label of each of the sample feature information;

[0038] The model parameters of the initial type prediction model are adjusted based on the difference data, and the object type prediction model is determined based on the adjusted initial type prediction model.

[0039] In one embodiment, the sample dataset includes multiple first sample data and second sample data, wherein the sample labels included in the first sample data are positive sample labels; the processing unit is further configured to perform the following steps:

[0040] From the sample feature information included in the plurality of first sample data, the target feature value under the target feature dimension is obtained, wherein the target feature dimension is any one of the N feature dimensions included in the sample feature information, and N is a positive integer;

[0041] Based on the target feature values ​​obtained under the target feature dimension, determine the feature value set corresponding to the target feature dimension;

[0042] The feature value set corresponding to the target feature dimension is subjected to feature value extraction processing to obtain the extracted feature value under the target feature dimension;

[0043] The second sample data is determined based on the extracted feature values ​​under the target feature dimension, and the sample labels included in the second sample data are positive sample labels.

[0044] In one embodiment, the model parameters of the initial type prediction model include weight parameters and bias parameters; when the processing unit adjusts the model parameters of the initial type prediction model based on the difference data, it specifically performs the following steps:

[0045] Based on the difference data, determine the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters in the initial type prediction model;

[0046] The gradient data of the weight parameters and the gradient data of the bias parameters are averaged to obtain gradient mean data, and the gradient data of the weight parameters and the gradient data of the bias parameters are varianced to obtain gradient variance data.

[0047] The gradient mean data and gradient variance data are corrected to obtain corrected gradient mean data and gradient variance data.

[0048] Based on the corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters, and the adjusted gradient data of the bias parameters, the weight parameters and the bias parameters are adjusted.

[0049] Accordingly, embodiments of this application provide a computer device, which includes:

[0050] A processor is a tool for implementing computer programs.

[0051] A computer-readable storage medium storing a computer program adapted for loading and execution by a processor using the aforementioned content push method.

[0052] Accordingly, this application also provides a computer-readable storage medium storing a computer program, which, when read and executed by the processor of a computer device, causes the computer device to perform the above-described content push method.

[0053] Accordingly, this application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the aforementioned content push method.

[0054] In this application, the game server can respond to content push requests from game clients and perform content push processing on the login object based on the object data of the login object corresponding to the game client. The content push method provided in this application allows for content push processing based on the object data of the login object, enabling the game server to perform different content push operations for different login objects, thus achieving personalized content push. It can obtain the object data of the login object logged into the game client and perform object type prediction processing based on the feature information of the object data to obtain the object type probability of the login object. This allows for accurate determination of the object type of the login object through the feature information of the object data, facilitating different content push operations based on different object types. This makes the pushed content more suitable for the needs of the login object, improves the game user experience, and increases the conversion rate of the pushed content. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the system architecture of a content push system provided in an embodiment of this application;

[0057] Figure 2 This is a flowchart illustrating a content push method provided in an embodiment of this application;

[0058] Figure 3 This is a schematic diagram of object data of a login object provided in an embodiment of this application;

[0059] Figure 4 This is a schematic diagram of an object type prediction model provided in an embodiment of this application;

[0060] Figure 5 This is a schematic diagram of a content push method provided in an embodiment of this application;

[0061] Figure 6This is a schematic diagram of another content push method provided in an embodiment of this application;

[0062] Figure 7 This is a schematic flowchart of a model training method provided in an embodiment of this application;

[0063] Figure 8 This is a schematic diagram of a feature derivation method provided in an embodiment of this application;

[0064] Figure 9 This is a structural block diagram of a content push device provided in an embodiment of this application;

[0065] Figure 10 This is a structural block diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0066] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0067] It should be noted that the terms "first," "second," etc., used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature specified with "first" or "second" may explicitly or implicitly include at least one of those features.

[0068] To enable game client users to quickly access specific information or perform specific actions, game servers can implement content push strategies, sending content to logged-in users on the game client. Specifically, game servers can use timed push strategies or event-driven push strategies. However, both of these strategies push content to all logged-in users. Some users, however, have a lower need for the pushed content and a lower probability of performing a conversion action. This results in a lower conversion rate for the pushed content using the aforementioned two methods.

[0069] Based on this, embodiments of this application provide a content push method, which can receive a content push request sent by a game client, the content push request carrying a login object, the login object being the object currently logged into the game client; in response to the content push request, obtain object data of the login object, the object data including basic data and game data, the game data including one or more of game scene description data, game scene statistical data, and game object description data; perform feature extraction processing on the object data to obtain feature information of the login object; perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, the object type prediction probability being the probability that the object type of the login object is the target object type; the probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold; and perform content push processing on the login object based on the object type prediction probability. Through the method provided by embodiments of this application, flexible content push processing can be performed based on the object data of the login object, realizing personalized content push and improving the conversion rate of pushed content.

[0070] The content push method provided in this application can be applied to the field of artificial intelligence. Artificial intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain optimal results. When using artificial intelligence technology to implement game operation and management, the content push method provided in this application can be used for content push processing. The content push method provided in this application can be added to the game management logic. Specifically, during content push processing, the game server with the game management logic deployed can receive a content push request carrying the login object sent by the game client; in response to the content push request, the game server can obtain the object data of the login object, perform feature extraction processing on the object data to obtain the feature information of the login object; perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, and perform content push processing on the login object based on the object type prediction probability. Using the method provided in this application, content push processing can be flexibly performed, realizing personalized content push, which is beneficial to improving the conversion rate of pushed content.

[0071] The content push method provided in this application can also be applied to the field of cloud computing. Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The cloud computing resource pool mainly includes: computing devices (virtualized machines containing operating systems), storage devices, and network devices. The content push method provided in this application can be implemented using cloud computing technology. Specifically, the game server can receive a content push request carrying the login object sent by the game client; in response to the content push request, the game server can obtain the object data of the login object from the cloud database, and use cloud computing technology to perform feature extraction processing on the object data to obtain the feature information of the login object; it can use cloud computing technology to perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, and perform content push processing on the login object based on the object type prediction probability. Through the method provided in this application, content push processing can be flexibly performed, realizing personalized content push, which is beneficial to improving the conversion rate of pushed content.

[0072] The architecture of the content push system provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0073] Please see Figure 1 The figure is a schematic diagram of the system architecture of a content push system provided in an embodiment of this application. The content push system includes a game client 101 (e.g., Figure 1 As shown, the game client 101 can interact with business objects, the game server 102, and the database 103. The game server device 102 can interact with the game client 101 and the database 103. Among them:

[0074] The game client 101 can run game applications and interact with business objects; one or more login objects (e.g., game accounts) can be logged in on the game client 101. The game client 101 can send requests to the game server 102, and can also receive and display content sent by the game server 102 (e.g., text or images). The game client 101 can be a handheld device (e.g., smartphone, tablet), computing device (e.g., personal computer (PC), in-vehicle terminal, intelligent voice interaction device, wearable device, or other intelligent device with display and communication functions, but is not limited to these).

[0075] Game server 102 can receive requests sent by game client 101, retrieve data from database 103, and perform corresponding operations based on the requests and the retrieved data. Game server 102 may also include an object type prediction model, which can be trained by game server 102 itself or sent from other devices different from game server 102. Game server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms.

[0076] Database 103 is used to store relevant data of game server 102, such as object information of logged-in objects and the predicted probability of object type of logged-in objects. Database 103 can be a local database in game server 102, or a cloud database associated with game server 102 (i.e., a database deployed in the cloud). Specifically, it can be deployed based on any of the following: private cloud, public cloud, hybrid cloud, edge cloud, etc., thus allowing the cloud database to focus on different functions. For example, a database deployed in a private cloud uses the basic cloud hardware of a user's personal device, focusing more on serving a small number of users, while a database deployed in a public cloud is based on a third-party cloud platform, allowing data sharing. Any user's data can be stored in this database, and any user can use the data in the database.

[0077] The following will elaborate on such matters. Figure 1 The working principle of the content push system shown is as follows:

[0078] Business objects can use the login object to log in to the game client 101. After the login is completed, the game client 101 can interact with the business object. In specific scenarios (e.g., after the login is completed or after a game session), the game client 101 can generate a content push request, which can include the login object.

[0079] The game client 101 can send a content push request to the game server 102. The game server 102 can receive the content push request sent by the game client 101 and respond to the content push request by retrieving the object data of the logged-in object from the database 103. The object data may include basic data and game data. The game data may include one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0080] After obtaining the object data of the login object, the game server 102 can perform feature extraction processing on the object data to obtain the feature information of the login object; based on the feature information, it can perform object type prediction processing to obtain the object type prediction probability of the login object. This object type prediction probability is the probability that the object type of the login object is the target object type; the probability that an object of the target object type will perform a conversion operation on the pushed content is greater than or equal to a first probability threshold. The game server 102 can perform content push processing on the login object based on the object type prediction probability (e.g., Figure 1 As shown, the game server 102 pushes the target content to the logged-in object on the game client 101. Through the content push method provided in this application embodiment, content push processing can be flexibly performed based on the object data of the logged-in object, realizing personalized content push and improving the conversion rate of the pushed content.

[0081] It is understood that the schematic diagrams of the content push system described in the embodiments of this application are for the purpose of more clearly illustrating the content push method of the embodiments of this application, and do not constitute a limitation on the content push method provided in the embodiments of this application. For example, the content push method provided in the embodiments of this application can be executed not only by the game server 102, but also by other devices different from the game server 102 that can communicate with the game client 101 and the database 103. Those skilled in the art will understand that... Figure 1 The number of game clients 101, game servers 102, and databases 103 shown in the examples is merely illustrative. Any number of devices can be configured according to business needs. Furthermore, as system architecture evolves and new business scenarios emerge, the content push method provided in this application embodiment is also applicable to similar technical problems.

[0082] It should be noted that the collection and processing of relevant data (e.g., object data of logged-in users) in this application should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0083] Please see Figure 2 , Figure 2 This is a flowchart illustrating a content push method provided in an embodiment of this application. This content push method can be implemented by the aforementioned game server 102, or by other devices capable of implementing the same method. The following description uses the implementation of the content push method by the aforementioned game server 102 as an example. The flow of the content push method provided in this embodiment includes, but is not limited to:

[0084] S201. Receive a content push request sent by the game client, wherein the content push request carries a login object, and the login object is the object currently logged in on the game client.

[0085] In this embodiment, the game server can receive a content push request sent by the game client. This content push request can be used to request the game server to perform content push processing for the game client. The content push request may carry a login object, which is the object currently logged into the game client. For example, the login object can be a game account.

[0086] It should be noted that content push requests can be sent by the game client in a specific scenario or after a specific operation is completed. For example, a content push request can be sent by the game client after the login operation is completed based on the logged-in object, or it can be sent after a game scene or game match is completed within the game client. The conditions for the game client to send content push requests can be adaptively adjusted according to different application scenarios and push content.

[0087] S202. In response to the content push request, obtain the object data of the logged-in object. The object data includes basic data and game data. The game data includes one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0088] In this embodiment, the game server can respond to a content push request and obtain object data of the logged-in object. This object data may include basic data and game data. The basic data is basic descriptive data associated with the logged-in object, such as the object's corresponding setting type data, location data, network environment data, etc. The game data is data associated with the game. Game data may include one or more of the following: game scene description data, game scene statistical data, and game object description data.

[0089] Game scene description data can include detailed data from one or more game scenes, such as game duration, game scene mode, and the objects to be manipulated within the game scene. Game scene statistics can include overall statistics for multiple game scenes over a period of time, such as the number of all game scenes involved in the logged-in object within a target time period and the number of game scenes of various types. Game object description data can include the logged-in object's game ability data, such as the logged-in object's game ability score and game level. The method provided in this application embodiment can acquire object data containing multiple dimensions, which is beneficial for subsequently determining the feature information of the logged-in object based on the object data, and then accurately determining the object type prediction probability of the logged-in object based on the feature information.

[0090] Please see Figure 3 This figure is a schematic diagram of object data for a login object provided in an embodiment of this application. The object data for the login object may include basic data and game data. The basic data may include setting type data (…). Figure 3 In the context of login objects, the data type is set to "X", and the location data is... Figure 3 In the data, the location data of the logged-in object is "City A"), and the network environment data is... Figure 3 In the context of the login object, the network environment data is defined as "traffic environment".

[0091] The game data includes game scene description data, game scene statistics data, and game object description data. The game scene description data may include detailed data on multiple game scenes involving the logged-in object (e.g., the 10 most recent game scenes). Figure 3 Taking a game scene as an example, the detailed data for each game scene includes the game duration (…). Figure 3 In the text, the game duration is "25 minutes", and the game scene type is (…). Figure 3 In the middle, the game scene type is "Type 1"), and the game scene capability data ( Figure 3 In the game, the scene capability data is "2-3-5".

[0092] Game scene statistics can include statistics on all game scenes involved by logged-in users within a certain period (e.g., 30 days), including type 1 scene time statistics. Figure 3 In the data, the time statistics for Type 1 scenarios are "300 minutes", and the number of times Type 1 scenarios are counted ( Figure 3 In the data, the statistics for the number of times in Type 1 scenarios are "12"), and the statistics for the objects operated on in Type 1 scenarios are ( Figure 3 In the context of game scene type 1, the statistical data for the operation objects is "3,5,2". Game scene statistics can also include relevant statistics for other types of game scenes, which are related to... Figure 3 The relevant statistics for game scenarios of type 1 are similar.

[0093] Game object description data can include the login object's game capabilities for different types of game scenarios. Game object description data can include capability data for Type 1 scenarios (…). Figure 3 In the middle, the ability data for type 1 scenario is "1002"), and the level data for type 1 scenario is ( Figure 3 In the context of Type 1 scenarios, the level data is "Level 1"; in Type 2 scenarios, the ability data is... Figure 3 In the middle, the ability data for type 2 scenario is "980", and the level data for type 2 scenario is ( Figure 3In the context of Type 2 scenarios, the level data is designated as "Second Level". The game object description data can also include relevant data for other types of game scenarios, which are related to... Figure 3 The relevant data for game scenarios in type 1 are similar.

[0094] It should be noted that the data included in the basic data and game data can be adaptively adjusted according to different actual application scenarios. The game server can obtain the object data of the logged-in object from the game log information of the logged-in object.

[0095] S203. Perform feature extraction processing on the object data to obtain the feature information of the logged-in object.

[0096] In this embodiment, the object data includes data of different dimensions and formats, such as text and numerical values. The game server can perform feature extraction processing on the object data to obtain the feature information of the logged-in object. This feature information can comprehensively and accurately represent the features in the object data, while removing useless data and achieving data format uniformity. This feature information can be a feature vector containing feature values ​​under multiple feature dimensions.

[0097] In one embodiment, the game server can utilize linear mapping information to perform feature extraction processing on object data, that is, linearly convert the data in the object data into feature values ​​under the feature dimension. The following explanation uses location data as an example: Since the basic data includes target location data, the implementation method for performing feature extraction processing on the object data to obtain the feature information of the logged-in object can be as follows: Obtain location mapping information, which indicates the linear mapping relationship between location data and location feature values ​​under the location feature dimension; query the location mapping information based on the target location data to determine the location feature values ​​under the location feature dimension corresponding to the target location data.

[0098] Location mapping information can be determined based on application requirements and location data. For example, location mapping information can indicate that the value corresponding to "City A" is "1" and the value corresponding to "City B" is "2". By querying location mapping information based on target location data, assuming the target location data in the object data is "City A", the location feature value under the corresponding location feature dimension can be determined to be "1". It should be noted that data in other dimensions of the object data can also be processed using the method provided in this application embodiment to obtain feature values ​​under the corresponding feature dimensions, thereby determining the feature information of the logged-in object. Through the method provided in this application embodiment, feature values ​​under feature dimensions can be determined using mapping information, which helps improve the efficiency of determining feature information and thus improves the efficiency of content push by the game server.

[0099] In one embodiment, the game can also utilize a mapping polynomial to perform feature extraction processing on the object data, that is, to non-linearly convert the data in the object data into feature values ​​under the feature dimension. The following uses game duration as an example: the game data in the object data includes game scene description data, which includes the target game duration; the implementation method for performing feature extraction processing on the object data to obtain the feature information of the logged-in object can be: obtaining a duration mapping polynomial, which indicates the non-linear mapping relationship between the game duration and the duration feature value under the game duration feature dimension; using the duration mapping polynomial, performing non-linear mapping processing on the target game duration to obtain the duration feature value under the game duration feature dimension corresponding to the target game duration. The mapping polynomial can be a non-first-order polynomial, that is, the mapping relationship between the data in the object data and the feature value under the feature dimension can be a non-linear relationship. For example: if the game duration is 10 minutes, the expression of the duration mapping polynomial can be y = 20 / x, then using the duration mapping polynomial, the duration feature value under the game duration feature dimension can be determined to be 2. It should be noted that data in other dimensions of the object data can also be processed using the method provided in this application embodiment, employing a mapping polynomial (the mapping polynomials corresponding to different feature dimensions can be different), thereby obtaining feature values ​​under the corresponding feature dimensions, and thus determining the feature information of the logged-in object. Through the method provided in this application embodiment, feature values ​​under feature dimensions can be determined using a non-linear mapping approach, enabling the feature values ​​to more accurately express the relevant features, thereby facilitating the acquisition of more accurate feature information.

[0100] It should be noted that when determining feature values ​​under a feature dimension using linear or non-linear transformation methods, missing values ​​can be added to ensure that the feature information comprehensively represents the object data. For example, suppose the feature value under the feature dimension corresponding to the set type data "X" is "0", the feature value under the feature dimension corresponding to the set type data "Y" is "1", and the missing feature value under the feature dimension corresponding to missing set type data is "2". When the set type data of the logged-in object is missing, the missing feature value can be determined as the feature value under the feature dimension of the set type.

[0101] In some cases, a missing feature dimension can be added to the end of the feature information, and the feature value under the missing feature dimension can be determined according to the degree of data missing in the object data. For example, the feature information of a login object can be a 21-dimensional feature vector, where the first 20 dimensions are obtained by feature extraction processing of the object data, and the last dimension of the feature vector is the missing feature dimension. If there is data missing in the object data, the feature value under the missing feature dimension can be determined to be "1", and if there is no data missing in the object data, the feature value under the missing feature dimension can be determined to be "0". Through the method provided by the embodiments of this application, missing feature values ​​can be introduced, so that the feature information of the login object can more completely and comprehensively express the relevant features of the object data, and it is also beneficial to determine the accurate object type prediction probability based on the feature information.

[0102] S204. Perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object. The object type prediction probability is the probability that the object type of the login object is the target object type. The probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0103] In this embodiment, the game server can perform object type prediction processing based on feature information to obtain the predicted object type probability of the logged-in object. This object type prediction refers to the probability that the logged-in object's object type is the target object type. For example, the object type probability can be 0.8, meaning the probability that the logged-in object's object type is the target object type is 0.8. The probability that an object of the target object type will perform a conversion operation on the pushed content is greater than or equal to a first probability threshold. This first probability threshold can be adaptively adjusted according to different application scenarios and pushed content. The target object type is the object type with a higher probability of performing a conversion operation on the pushed content. For example, if the pushed content indicates information retrieval, the target object type is the object with a higher probability of performing an information retrieval operation on the pushed content. If the pushed content indicates resource transfer, the target object type is the object with a higher probability of performing a resource transfer operation on the pushed content. The method provided in this embodiment can accurately determine the predicted object type probability of the logged-in object, which is beneficial for subsequent content push processing based on the predicted object type probability, thereby achieving personalized content push.

[0104] In one embodiment, the method for obtaining the object type prediction probability of a logged-in object by performing object type prediction processing based on feature information can be as follows: Feature information is input into an object type prediction model for object type prediction processing to obtain the object type prediction probability of the logged-in object. The object type prediction model is obtained by adjusting an initial type prediction model using difference data. The difference data is determined based on the sample labels and sample probabilities of the sample feature information. The sample labels indicate the probability that the sample object's object type is the target object type. The sample object is the object corresponding to the sample feature information. The sample probability is obtained by performing object type prediction processing on the sample feature information using the initial type prediction model. The sample feature information is determined based on the object data of the sample object. The game server can use the object type prediction model to process the feature information to obtain the object type prediction probability of the logged-in object. Using the method provided in this embodiment, the trained model can be used to determine the object type prediction probability, which can effectively improve the accuracy and processing efficiency of the object type prediction probability, thereby improving the overall efficiency of content push processing.

[0105] In one embodiment, the feature information includes feature values ​​under N feature dimensions, where N is a positive integer. The method for inputting the feature information into an object type prediction model to obtain the object type prediction probability of the logged-in object can be as follows: A weight data set is obtained, comprising M first weight data groups and a second weight data group. The first weight data group includes N weight data and a first bias data, where M is a positive integer. The second weight data group includes M weight data and a second bias data. The object type prediction model can be a Multilayer Perceptron (MLP). A MLP can include an input layer, one or more hidden layers, and an output layer. The following example illustrates a MLP containing only one hidden layer: The obtained weight data set can include M first weight data groups and a second weight data group. The first weight data group contains weight data between the input layer and the hidden layer, and the second weight data group contains weight data between the hidden layer and the output layer. M and N can be the same or different.

[0106] The N weight data, the first bias data, and the N feature values ​​included in the feature information in the target first weight data group are combined to obtain the comprehensive feature value corresponding to the target first weight data group. The target first weight data group can be any first weight data group in the weight data set. For any first weight data group (i.e., the target first weight data group) in the M first weight data groups, the N weight data, the first bias data, and the N feature values ​​included in the feature information in the target first weight data group can be combined to obtain the comprehensive feature value corresponding to the target first weight data group. For example, if N is 2, the feature information is (2,2), the weight data in the target first weight data group is 0.1, 0.3, and the first bias data is 1, then the comprehensive feature value can be determined as x = 2*0.1 + 2*0.3 + 1 = 1.8.

[0107] The transformed eigenvalues ​​of the target first-weighted data group are obtained by performing a nonlinear transformation on the comprehensive eigenvalues. This transformation can be achieved using an activation function (such as the Rectified Linear Unit (ReLU) function or the Sigmoid function). For example, using the ReLU activation function, the transformed eigenvalue y = ReLU(x), where x is the comprehensive eigenvalue.

[0108] After determining the transformation feature values ​​corresponding to the M first weighted data groups in the weighted data set, feature induction processing is performed on the M weighted data in the second weighted data group, the second bias data, and the transformation feature values ​​corresponding to the M first weighted data groups in the weighted data set to obtain the summary feature value corresponding to the second weighted data group. After determining the transformation feature values ​​corresponding to each first weighted data group, feature induction processing can be performed using the M transformation feature values, the M weighted data in the second weighted data group, and the second bias data to obtain the summary feature value. For example, if M is 3, the M transformation feature values ​​are 1, 2, and 3, the M weighted data are 0.1, 0.2, and 0.3, and the second bias data is 0.5, then the summary feature value is f = 1*0.1 + 2*0.2 + 3*0.3 + 0.5 = 1.9.

[0109] The summarized feature values ​​are subjected to nonlinear transformation to obtain the object type prediction probability of the logged-in object. Alternatively, an activation function (Sigmoid function) can be used to perform a nonlinear transformation on the summarized feature values ​​to obtain the object type prediction probability of the logged-in object. For example, the object type prediction probability p = Sigmoid(f), where f is the summarized feature value. The method provided in this application embodiment can process the feature information of the logged-in object to obtain the object type prediction probability, which helps to improve the accuracy of the object type prediction probability.

[0110] In one embodiment, the object type prediction model can be a Multilayer Perceptron (MLP). MLPs possess nonlinear modeling capabilities and good feature learning abilities. An MLP can include nonlinear activation functions and multiple hidden layers, enabling it to model complex nonlinear relationships and automatically learn feature representations from the input data. MLPs perform well when processing data with complex structures and nonlinear features. Specifically, an MLP can include an input layer, hidden layers, and an output layer. The input layer is the first layer of the MLP, used to receive raw data or features as input. The input layer can include one or more input layer neurons, each corresponding to a feature of the input data. The input layer does not perform any calculations or transformations; it simply passes the input data to the next layer.

[0111] Hidden layers are located between the input and output layers. An MLP can contain one or more hidden layers. Each hidden layer can contain one or more hidden neurons, and each hidden neuron is connected to all neurons in the previous layer (either the input or hidden layer). The main function of hidden layers is to perform non-linear transformations and feature extraction on the input data. Hidden layers compute and transmit information through weights and activation functions.

[0112] The output layer is the last layer of an MLP, used to produce the prediction result or output. In the object type prediction model, the output layer is a single neuron with the sigmoid function as the activation function, and the output is the predicted probability of the object type of the logged-in object. It should be noted that the model structure of the object type prediction model can also be similar to that of a Gradient Boosting Decision Tree (GBDT) model or a Support Vector Machine (SVM) model.

[0113] Please see Figure 4This figure is a schematic diagram of an object type prediction model provided in an embodiment of this application. The object type prediction model may include an input layer, a hidden layer, and an output layer. The feature information of the logged-in object can be a K-dimensional feature vector (K is a positive integer). The input layer may include K input layer neurons, the hidden layer may include K hidden layer neurons (each hidden layer neuron is connected to one of the K input layer neurons), and the output layer may include one output layer neuron. When determining the object type prediction probability of the logged-in object, the feature information can be input into the object type prediction model. The input layer of the object type prediction model transmits the feature values ​​(i.e., X1, X2, X3, ..., XK) of the K feature dimensions to the hidden layer through the input layer neurons. Taking the first hidden layer neuron (H1) in the hidden layer as an example, hidden layer neuron H1 is connected to K input layer neurons, and the weight data (i.e., w1, w2, w3, ..., wK), bias data (b1), and hidden layer activation function (i.e., ReLU function) between hidden layer neuron H1 and K input layer neurons can be determined. Then, the formula for calculating the output data of hidden layer neuron H1 can be shown in the following formula (1):

[0114]

[0115] In equation (1) above, Xi represents the output data of the i-th input layer neuron, wi represents the weight between the i-th input layer neuron and the hidden layer neuron H1, and a1 represents the output data of the hidden layer neuron H1. The output data of K hidden layer neurons (i.e., a1, a2, a3, ..., aK) can be determined using the same method.

[0116] The weight data, bias data, and activation function (Sigmoid function) between the output layer neurons and the K hidden layer neurons can be determined. A method similar to equation (1) above can be used to determine the output data of the output layer neurons, which is the predicted probability of the object type of the logged-in object. Through the method provided in this application embodiment, the predicted probability of the object type can be quickly and accurately determined using the object type prediction model, which is beneficial for subsequent personalized content push processing based on the predicted probability of the object type.

[0117] S205. Based on the predicted probability of the object type, perform content push processing on the logged-in object.

[0118] In this embodiment, the game server can perform content push processing on logged-in objects based on the predicted probability of object type. Since the object data of logged-in objects are different, and object information can reflect the preferences, behaviors, and operational characteristics of logged-in objects to a certain extent, the predicted probability of object type determined based on object data is also different. Performing content push processing on logged-in objects based on the predicted probability of object type can effectively achieve personalized content push processing, making the content push more in line with the needs of the business objects corresponding to the logged-in objects, thereby improving the game user experience.

[0119] In one embodiment, the implementation method for pushing content to a logged-in object based on the predicted probability of the object type can be as follows: comparing the predicted probability of the object type with a second probability threshold corresponding to the logged-in object; if the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then determining the target push content associated with the game and pushing the target push content to the logged-in object; if the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then not pushing content to the logged-in object.

[0120] The second probability threshold can be different for different login objects. The game server can compare the predicted probability of the object type with the second probability threshold corresponding to the login object. If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then the object type of the login object can be determined to be the target object type. The login object has a high probability of performing a conversion operation on the pushed content, and the target pushed content associated with the game can be determined and pushed to the login object.

[0121] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, it can be determined that the object type of the logged-in object is not the target object type. The probability that the logged-in object will perform a conversion operation on the pushed content is low, and content push processing can be omitted. The method provided in this application embodiment can perform content push processing based on the predicted probability of the logged-in object's object type, which helps improve the conversion rate of pushed content, reduce invalid content pushes, and save resources. Furthermore, using different content push processing for objects of different types (non-target and target) can effectively improve the user experience for both types of objects in the game.

[0122] It should be noted that both the first and second probability thresholds can be adaptively adjusted according to different application scenarios, and the second probability threshold can also be adaptively adjusted based on the relevant data of its corresponding login object. For example, if the probability of login object 1 performing a conversion operation on the pushed content is low for a period of time, then using the method provided in this application, the game server will not process content pushes for login object 1. However, this situation may result in login object 1 not receiving push content in a timely manner. Therefore, the second probability threshold corresponding to login object 1 can be appropriately reduced to ensure that login object 1 can receive push content in a timely manner. As another example, if the probability of login object 2 performing a conversion operation on the pushed content is high for a period of time, then using the method provided in this application, the game server can process content pushes for login object 2 more frequently. However, this situation may result in a poor user experience for login object 2 in the game. Therefore, the second probability threshold corresponding to login object 2 can be appropriately increased to appropriately reduce the amount of push content received by login object 2 and improve the user experience.

[0123] Please see Figure 5 This figure is a schematic diagram of a content push method provided in an embodiment of this application. The game server can correspond to the above... Figure 1 The game server 102 in the game server. When implementing the content push method, the game server can perform the following steps: S501, Data Acquisition: In response to a content push request, obtain the object data of the logged-in object; the game server can respond to a content push request carrying the logged-in object sent by the game client and obtain the object data of the logged-in object, which may include basic data and game data. S502, Feature Engineering: Perform feature extraction processing on the object data to obtain the feature information of the logged-in object; the game server can perform feature extraction processing on the object data to obtain the feature information of the logged-in object, which may be a feature vector including one or more feature dimensions. S503, Model Prediction: Input the feature information into an object type prediction model for processing to obtain the object type prediction probability; the game server can input the feature information into an object type prediction model for object type prediction processing to obtain the object type prediction probability of the logged-in object. This object type prediction probability is the probability that the object type of the logged-in object is the target object type, and objects of the target object type have a higher probability of performing a conversion operation in response to the pushed content.

[0124] S504. Threshold Judgment: The predicted probability of the object type is compared with a second probability threshold to obtain a comparison result. The game server can compare the predicted probability of the object type of the logged-in object with the second probability threshold corresponding to the logged-in object to obtain a comparison result. The second probability threshold corresponding to different logged-in objects can be different. S505. Content Push: Content push processing is performed on the logged-in object based on the comparison result. If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold corresponding to the logged-in object, the game server can determine the target push content associated with the game and push the determined target push content to the logged-in object. If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold corresponding to the logged-in object, the game server can choose not to perform content push processing on the logged-in object. The method provided by this application embodiment can achieve personalized content push, which is beneficial to improving the game user experience.

[0125] In one embodiment, the target push content can be game-scene promotional content, which can be used to facilitate resource transfer operations within the game, increase the game participation of logged-in users, and promote social interaction among logged-in users. Specifically, game-scene promotional content can serve as a resource transfer promotion strategy to increase the frequency of resource transfers by logged-in users; game-scene promotional content can enable logged-in users to acquire more game content, scenes, or items at a lower cost, thereby increasing their game participation and gaming experience. Game-scene promotional content can also enable logged-in users to share push content with other logged-in users, thereby promoting social interaction among them.

[0126] In one embodiment, if the comparison result indicates that the predicted probability of the object type of the logged-in object is greater than the second probability threshold corresponding to the logged-in object, the game server can also obtain the push content acquisition configuration information of the logged-in object. This push content acquisition configuration information is used to instruct the logged-in object on relevant configuration information for acquiring push content. If the push content acquisition configuration information indicates that the logged-in object has a push content acquisition requirement, the game server can determine the target push content associated with the game and push the target push content to the logged-in object. If the push content acquisition configuration information indicates that the logged-in object does not have a push content acquisition requirement, the game server can choose not to perform content push processing on the logged-in object. Through the method provided in this application embodiment, content push processing can be performed on the logged-in object when the logged-in object has a push content acquisition requirement, which can prevent logged-in objects without a push content requirement from receiving push content, thus improving the user experience of the logged-in object and increasing the conversion rate of the logged-in object for push content.

[0127] In one embodiment, the content push method provided in this application may further include the following steps: if the comparison result indicates that the predicted probability of the object type is less than or equal to a second probability threshold, then determine the push frequency of the login object based on the number of times content is pushed to the login object within the target time; compare the push frequency of the login object with the push frequency threshold to obtain a push frequency comparison result; if the push frequency comparison result indicates that the push frequency of the login object is less than or equal to the push frequency threshold, then reduce the second probability threshold to obtain a reduced second probability threshold.

[0128] When the predicted probability of the login object's object type is less than or equal to the second probability threshold, the game server can determine the push frequency of the login object within a target time (e.g., one day, one week, etc.), and compare the push frequency of the login object with the push frequency threshold. If the push frequency of the login object is less than or equal to the push frequency threshold, it means that the number of times content is pushed to the login object within the target time is relatively small. The second probability threshold can be reduced to obtain a reduced second probability threshold, thereby increasing the number of times content is pushed to the login object in the future. Through the method provided in this application embodiment, the second probability threshold corresponding to the login object can be adjusted, ensuring the rationality of content push and also helping to improve the game user experience.

[0129] It should be noted that the method provided in this application can determine the object type prediction probability of the login object based on the object data of the login object, and thus roughly determine the object type of the login object. This allows the determination of whether to respond to the content push request from the game client and push content to the login object. The specific content to be pushed to the login object can be determined based on the relevant data of the login object.

[0130] Please see Figure 6 This figure is a schematic diagram of another content push method provided in an embodiment of this application. Figure 6 This includes the game client (which can correspond to the above). Figure 1 The game client 101 and the game server (which can correspond to the above) Figure 1 (Game server 102). A game client can log in to a specific user object. The game client can load and run a game scene (e.g., a game match). After a game scene completes (e.g., after a game match ends), the game client can generate a content push request carrying the logged-in user object and send this request to the game server. The game server can respond to the content push request and obtain the logged-in user object data. This object data can include basic data and game data, including one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0131] The game server can perform feature extraction on object data to determine the characteristic information of the logged-in object; and based on the characteristic information, perform object type prediction to obtain the object type prediction probability of the logged-in object. The object type prediction probability of the logged-in object is the probability that the object type of the logged-in object is the target object type; the probability that an object of the target object type will perform a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0132] The game server can compare the predicted probability of an object type with a second probability threshold corresponding to the logged-in object. If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then target push content associated with the game is determined and pushed to the logged-in object, which the game client can receive and display. If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then no content push processing is performed on the logged-in object. The method provided in this application embodiment can perform content push processing at specific times (e.g., after the game scene finishes running, after the login operation is completed), making content push more timely. It can also perform content push operations based on the object data of the logged-in object, realizing personalized content push processing for the logged-in object, making the content push more in line with the needs of the logged-in object, and improving the user experience of the game.

[0133] The content push method provided in this application can respond to game client requests and perform content push processing at specific times (e.g., after the game scene finishes running, after the login operation is completed), making the content push more timely; it can perform content push operations based on the object data of the logged-in object, realizing personalized content push processing for the logged-in object, making the content push more in line with the needs of the logged-in object, which is conducive to improving the user experience of the game; it can push content to the logged-in object, thereby increasing the frequency of in-game resource transfer operations, increasing the conversion rate of pushed content, enhancing the game participation of the logged-in object, and increasing social interaction among multiple logged-in objects.

[0134] Please see Figure 7 , Figure 7 This is a flowchart illustrating a model training method provided in an embodiment of this application. This model training method can be a model training method for an object type prediction model. This model training method can be implemented by the aforementioned game server 102, or by other devices capable of executing this model training method. The following description uses the game server 102 as an example to illustrate the model training method. The flowchart of the model training method provided in this embodiment includes, but is not limited to:

[0135] S701. Obtain a sample dataset, which includes multiple sample data. The sample data includes sample feature information and sample labels of the sample feature information. The sample feature information with positive sample labels is determined based on the object data of positive sample objects. The sample feature information with negative sample labels is determined based on the object data of negative sample objects. The positive sample objects are those for which no conversion operation was performed on the pushed content in the first time period and which performed a conversion operation on the pushed content in the second time period. The negative sample objects are those for which a conversion operation was performed on the pushed content in both the first and second time periods, with the first time period being earlier than the second time period.

[0136] In this embodiment, the game server can obtain a sample dataset, which may include multiple sample data. Each sample data may include sample feature data and sample labels for the sample feature data. The sample labels may be positive or negative. The sample feature information with positive sample labels is determined based on the object data of the positive sample object, and the sample feature information with negative sample labels is determined based on the object data of the negative sample object.

[0137] The sample data is represented as {(n1, n2, ..., nj), label}, where (n1, n2, ..., nj) represents the sample feature information, which can be an i-dimensional feature vector (i is an integer greater than 2), and label represents the sample label of the sample feature information.

[0138] The model training method provided in this application is used to train a model that can accurately predict the object type prediction probability of a logged-in object. This object type prediction probability indicates the probability that the logged-in object's object type is the target object type, and objects of the target object type have a higher probability of performing a conversion operation on the pushed content. For example, if logged-in object 3 does not perform a conversion operation on the pushed content on the first day, but performs a conversion operation on the pushed content on the second day, then for the first day, the probability that logged-in object 3 will subsequently perform a conversion operation is higher (or logged-in object 3 is said to be the object that will perform a potential conversion operation). In this case, the object type of logged-in object 3 is the target object type. If logged-in object 3 performs a conversion operation on the pushed content on both the first and second days, then for the first day, logged-in object 3 is not the object that will perform a potential conversion operation. In this case, the object type of logged-in object 3 is not the target object type.

[0139] Based on the above logic, positive sample objects can be identified as those that did not perform a conversion operation on the pushed content in the first time period but performed a conversion operation on the pushed content in the second time period. Negative sample objects are those that performed a conversion operation on the pushed content in both the first and second time periods, with the first time period being earlier than the second time period. The first and second time periods can have the same length; for example, the first time period could be January 1st, and the second time period could be January 2nd. The method provided in this application embodiment can reasonably determine sample data, which is beneficial for subsequently using the sample data to train a model that meets application requirements.

[0140] In one embodiment, the model training method provided in this application may further include the following steps: acquiring object data of a sample object, wherein the type of the sample object is a positive sample object or a negative sample object; performing feature extraction processing on the object data of the sample object to obtain sample feature information; and generating sample data based on the sample feature information and the type of the sample object. The game server can acquire the object data of the sample object and perform feature extraction processing on the object data of the sample object to obtain sample feature information. The specific implementation process for determining the sample feature information is similar to the implementation method in step S203 above. Sample data can be generated based on the sample feature information and the type of the sample object.

[0141] In one embodiment, after determining the sample data based on the object data of the sample object, feature derivation processing can be performed on the sample feature information in the determined sample data to obtain new sample feature information, thereby increasing the richness of the sample feature information. Specifically, the sample dataset may include multiple first sample data and second sample data. The sample labels included in the first sample data are positive sample labels. The model training method provided in this application embodiment may further include the following steps: obtaining the target feature value under the target feature dimension from the sample feature information included in the multiple first sample data. The target feature dimension is any one of the N feature dimensions included in the sample feature information, where N is a positive integer. For any feature dimension (i.e., the target feature dimension) in the sample feature information, the feature value under the target feature dimension in the sample feature information of the multiple first sample data can be obtained.

[0142] Based on the target feature values ​​obtained under the target feature dimension, determine the feature value set corresponding to the target feature dimension; generate the feature value set corresponding to the target feature dimension based on the multiple target feature values ​​obtained, and the feature value set may not include duplicate feature values.

[0143] Feature value extraction is performed on the feature value set corresponding to the target feature dimension to obtain the extracted feature value under the target feature dimension; feature value extraction can also be performed on the feature value set (e.g., random extraction) to obtain the extracted feature value.

[0144] The second sample data is determined based on the extracted feature values ​​under the target feature dimension. The sample labels of the second sample data are positive sample labels.

[0145] For example, the sample dataset includes three first sample data points, all of which have positive labels. The feature information for these three first sample data points are (0.1, 0.2, 0.5, 0.14), (0.2, 0.36, 0.28, 0.07), and (0.87, 0.52, 0.71, 0.39), respectively. Assuming the target feature dimension is the first feature dimension, then the feature value under the first feature dimension of the three sample feature information can be obtained, which is 0.1. The feature set {0.1, 0.2, 0.87} is determined based on these three feature values, and the feature set is extracted to obtain the extracted feature value (0.87). This extracted feature value is then combined with the sample feature information (0.1, 0.2, 0.5, 0.14) from the first sample data to obtain the second sample data. This second sample data includes positive sample labels and sample feature information (0.87, 0.2, 0.5, 0.14). As the example shows, the feature derivation processing based on the sample feature information (hereinafter referred to as the first sample feature information) results in a second sample data with high similarity to the first sample feature information, but also some differences. This effectively increases the richness of the sample feature data, which is beneficial for avoiding overfitting when training the model using the sample data, and helps improve the model's prediction accuracy. In some cases, the process of obtaining second sample data by performing feature derivation processing on the first sample data can also be to perform feature multiplication or division processing on the sample feature information of the two sample data with the same sample label to obtain the sample feature information of the second sample data.

[0146] It should be noted that the second sample feature information in the above embodiments is obtained by adjusting the feature values ​​under a certain feature dimension of the first sample feature information. In practical applications, the second sample feature information can be obtained by adjusting the feature values ​​under one or more feature dimensions of the first sample feature information. Furthermore, the above embodiments only show the implementation method for determining the second sample data including positive sample labels; the implementation method for determining the second sample data including negative sample labels is similar to the above embodiments.

[0147] Please see Figure 8 This figure is a schematic diagram of a feature derivation method provided in an embodiment of this application. Assume the sample dataset includes four first sample data points (i.e....) Figure 8The sample data (1-4) in the example consist of four first sample data sets, all of which contain positive sample labels. The sample feature information for these four first sample data sets is (0.10, 0.20, 0.50, 0.14), (0.20, 0.36, 0.35, 0.07), (0.87, 0.52, 0.71, 0.39), and (0.51, 0.07, 0.35, 0.98), respectively. Assuming feature derivation processing is performed on the second and third feature dimensions, the game server can obtain the feature values ​​under the second feature dimension from the four sample feature information sets, resulting in the feature value set {0.20, 0.36, 0.52, 0.07}; it can also obtain the feature values ​​under the third feature dimension from the four sample feature information sets, resulting in the feature value set {0.50, 0.35, 0.71}. The feature value set corresponding to the second feature dimension can be randomly sampled to obtain the sampled feature value (0.52) under the second feature dimension. The feature value set corresponding to the third feature dimension can be randomly sampled to obtain the sampled feature value (0.35) under the third feature dimension.

[0148] The extracted feature values ​​under the second feature dimension, the extracted feature values ​​under the third feature dimension, and the sample feature information (0.10, 0.20, 0.50, 0.14) in the first sample data can be used (i.e. Figure 8 The sample feature information of sample data 1 is combined and processed to obtain the sample feature information (0.10, 0.52, 0.35, 0.14) in the second sample data. The sample labels included in the second sample data are positive sample labels. The method provided in this application embodiment can obtain more diverse sample feature information, which is beneficial for increasing the richness of sample feature information, avoiding overfitting when using sample feature information for model training, and improving the prediction accuracy of the model.

[0149] S702. Input the sample feature information included in each sample data in the sample dataset into the initial type prediction model for object type prediction processing to obtain the sample probability corresponding to each sample feature information.

[0150] In this embodiment, the game server can input the sample feature information of each sample data in the sample dataset into an initial type prediction model for object type prediction processing, thereby obtaining the sample probability corresponding to each sample feature information. This initial type prediction model can be an MLP (Multi-Level Processing), which may include an input layer, one or more hidden layers, and an output layer. The model structure of the initial type prediction model can be as described above. Figure 4 As shown.

[0151] S703. Determine the difference data based on the sample probability corresponding to each of the sample feature information and the sample label of each of the sample feature information.

[0152] In this embodiment, the game server can determine the difference data based on the sample probability corresponding to each sample feature information and the sample label corresponding to each sample feature information. This difference data can reflect, to some extent, the difference between the predicted probability and the actual probability of the initial type prediction model.

[0153] In one embodiment, the method for determining the difference data based on the sample probability corresponding to each sample feature information and the sample label of each sample feature information can be as follows: determining the reference probability of each sample feature information based on the sample label of each sample feature information; performing difference processing based on the sample probability and reference probability corresponding to each sample feature information to obtain the first difference data; performing regularization processing based on the model parameters of the initial probability model to obtain the second difference data; and determining the difference data based on the first difference data and the second difference data.

[0154] For any sample feature in the overall sample feature information, if the sample label of that sample feature is positive, the reference probability of that sample feature is determined to be 1; if the sample label of that sample feature is negative, the reference probability of that sample feature is determined to be 0. Based on the sample probabilities and reference probabilities corresponding to each sample feature, difference processing is performed to obtain the first difference data. This difference processing can be performed using the cross-entropy loss function. The cross-entropy loss function directly measures the difference between the model's predicted probability (i.e., sample probability) and the sample label, effectively solving the gradient vanishing problem, and performing well in multi-class classification problems.

[0155] The model parameters in the initial type prediction model can be obtained, and regularization processing can be performed on the obtained model parameters to obtain the second difference data. The regularization process can be performed using L1 regularization, which involves adding an L1 regularization term (i.e., the second difference data) during the calculation of loss data. This achieves regularization optimization when updating parameters based on the loss data, thereby improving the model's prediction accuracy, controlling model complexity, and preventing overfitting. The first and second difference data can be summed to obtain the difference data. The method provided in this application embodiment can determine the difference data, which is beneficial for subsequent adjustment of model parameters based on the difference data, thereby improving the model's prediction accuracy and avoiding overfitting.

[0156] S704. Adjust the model parameters of the initial type prediction model according to the difference data, and determine the object type prediction model according to the adjusted initial type prediction model.

[0157] In this embodiment, the game server can adjust the model parameters of the initial type prediction model based on the difference data to obtain an adjusted initial type prediction model. The game server can use the model training method provided in this embodiment to adjust the model parameters of the adjusted initial type prediction model multiple times. When the adjusted initial type prediction model meets the requirements (e.g., the number of model parameter adjustments reaches a preset number or the loss function converges), the adjusted initial type prediction model can be determined as the object type prediction model. The method provided in this embodiment can obtain an object type prediction model with high prediction accuracy, which is beneficial for determining the accurate object type prediction probability of logged-in objects, thereby facilitating personalized content push processing.

[0158] In one embodiment, the model parameters of the initial type prediction model include weight parameters and bias parameters; the method for adjusting the model parameters of the initial type prediction model based on the difference data can be: determining the adjustment gradient data of the weight parameters and the adjustment gradient data of the bias parameters in the initial type prediction model based on the difference data; as described above. Figure 4 Taking the model shown as an example, the game server can determine the weight data of each neuron in the initial type prediction model based on the difference data (e.g., Figure 4 The adjusted gradient data of the bias parameters (including w1, w2, w3, ..., wK, and bias data).

[0159] The gradient mean data is obtained by averaging the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters, and the gradient variance data is obtained by averaging the adjusted gradient data of the weight parameters and the bias parameters. The gradient mean data can also be called the first moment estimate data, which is used to indicate the moving average of the gradient. The gradient variance data can be determined, which can be called the second moment estimate data, and is used to indicate the moving average of the squared gradient.

[0160] The gradient mean and gradient variance data are corrected to obtain corrected gradient mean and gradient variance data. The gradient mean and gradient variance data may contain errors, so they can be corrected to obtain corrected gradient mean and gradient variance data.

[0161] Based on the corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters, the weight parameters and bias parameters are adjusted. The corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters are used to adjust the weight parameters, and the corrected gradient mean and gradient variance data, the adjusted gradient data of the bias parameters are used to adjust the bias parameters.

[0162] In some cases, the adjustment of model parameters can be achieved using the Adaptive Moment Estimation (Adam) algorithm. Adam is a commonly used optimization algorithm that combines the characteristics of momentum estimation and adaptive learning rate, and can be used to train machine learning models. Relevant parameters of the Adam algorithm can be initialized, such as the learning rate, momentum, and decay rate. Specifically, when adjusting model parameters, the game server can determine the gradient of the model parameters based on the difference data, and determine the first-order moment estimation data and second-order moment estimation data based on the gradient of the model parameters. After correcting the first-order moment estimation data and second-order moment estimation data, the model parameters are updated using the corrected first-order moment estimation data and second-order moment estimation data, learning rate, momentum, and other parameters. During model training, the Adam algorithm can adaptively adjust the learning rate, making it suitable for different parameters and data; it can combine the smoothness of momentum estimation and the advantages of adaptive learning rate to accelerate the convergence speed of the model, thereby effectively improving the training efficiency of the model.

[0163] The model training method provided in this application can use the first sample data in the sample dataset to derive features to obtain the second sample data, which increases the richness of sample feature information and effectively avoids overfitting. The cross-entropy function and L1 regularization technique can be used to determine the difference data, thereby controlling the complexity of the model and further preventing model overfitting. The difference data can be used to adjust the model parameters, which can improve the training efficiency of the model and the prediction accuracy of the model. This is beneficial for using the trained object type prediction model to achieve personalized content push processing.

[0164] Please see Figure 9 , Figure 9 This is a structural block diagram of a content push device provided in an embodiment of this application. The content push device can be disposed in the computer device provided in this embodiment of the application, and the computer device can be as described above. Figure 1 The game server 102 in the content push system shown. Figure 9The content delivery device shown can be a computer program running on a computer device, and the content delivery device can be used to execute... Figure 2 or Figure 7 Some or all of the steps in the method embodiments shown. Please refer to [link / reference]. Figure 9 The content push device may include the following units:

[0165] The receiving unit 901 is used to receive a content push request sent by the game client, wherein the content push request carries a login object, and the login object is the object currently logged into the game client.

[0166] The acquisition unit 902 is used to acquire the object data of the logged-in object in response to the content push request. The object data includes basic data and game data. The game data includes one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0167] Processing unit 903 is used to perform feature extraction processing on the object data to obtain feature information of the logged-in object;

[0168] The processing unit 903 is further configured to perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, wherein the object type prediction probability is the probability that the object type of the login object is the target object type; the probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0169] The processing unit 903 is also used to perform content push processing on the logged-in object based on the probability predicted by the object type.

[0170] In one embodiment, when the processing unit 903 performs content push processing on the logged-in object based on the predicted probability of the object type, it specifically performs the following steps:

[0171] The predicted probability of the object type is compared with the second probability threshold corresponding to the login object;

[0172] If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then the target push content associated with the game is determined, and the target push content is pushed to the logged-in object;

[0173] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then no content push processing will be performed on the logged-in object.

[0174] In one embodiment, the processing unit 903 is further configured to perform the following steps:

[0175] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then the push frequency of the login object is determined based on the number of times content is pushed to the login object within the target time.

[0176] The push frequency of the logged-in object is compared with the push frequency threshold to obtain the push frequency comparison result;

[0177] If the push frequency comparison result indicates that the push frequency of the logged-in object is less than or equal to the push frequency threshold, then the second probability threshold is reduced to obtain a reduced second probability threshold.

[0178] In one embodiment, the basic data includes target location data, and the processing unit 903 is used to perform feature extraction processing on the object data to obtain the feature information of the logged-in object. Specifically, it performs the following steps:

[0179] Obtain location mapping information, which indicates the linear mapping relationship between location data and location feature values ​​under the location feature dimension;

[0180] Based on the target location data, query the location mapping information to determine the location feature value under the location feature dimension corresponding to the target location data.

[0181] In one embodiment, the game data includes game scene description data, which includes the target game duration; when the processing unit 903 performs feature extraction processing on the object data to obtain the feature information of the logged-in object, it specifically performs the following steps:

[0182] Obtain the duration mapping polynomial, which is used to indicate the nonlinear mapping relationship between game duration and duration feature values ​​under the game duration feature dimension;

[0183] Using the duration mapping polynomial, the target game duration is nonlinearly mapped to obtain the duration feature value under the game duration feature dimension corresponding to the target game duration.

[0184] In one embodiment, when the processing unit 903 performs object type prediction processing based on the feature information to obtain the object type prediction probability of the logged-in object, it specifically performs the following steps:

[0185] The feature information is input into the object type prediction model for object type prediction processing to obtain the object type prediction probability of the login object.

[0186] The object type prediction model is obtained by adjusting the initial type prediction model using difference data. The difference data is determined based on the sample label and sample probability of the sample feature information. The sample label is used to indicate the probability that the object type of the sample object is the target object type. The sample object is the object corresponding to the sample feature information. The sample probability is obtained by performing object type prediction processing on the sample feature information using the initial type prediction model. The sample feature information is determined based on the object data of the sample object.

[0187] In one embodiment, the processing unit 903 is further configured to perform the following steps:

[0188] Obtain a sample dataset, which includes multiple sample data. The sample data includes sample feature information and sample labels for the sample feature information. The sample feature information with positive sample labels is determined based on the object data of positive sample objects. The sample feature information with negative sample labels is determined based on the object data of negative sample objects. The positive sample objects are those for which no conversion operation was performed on the push content in the first time period and which was performed on the push content in the second time period. The negative sample objects are those for which a conversion operation was performed on the push content in both the first and second time periods, with the first time period being earlier than the second time period.

[0189] The sample feature information included in each sample data in the sample dataset is input into the initial type prediction model for object type prediction processing to obtain the sample probability corresponding to each sample feature information;

[0190] The difference data is determined based on the sample probability corresponding to each of the sample feature information and the sample label of each of the sample feature information;

[0191] The model parameters of the initial type prediction model are adjusted based on the difference data, and the object type prediction model is determined based on the adjusted initial type prediction model.

[0192] In one embodiment, the sample dataset includes multiple first sample data and second sample data, wherein the sample labels included in the first sample data are positive sample labels; the processing unit 903 is further configured to perform the following steps:

[0193] From the sample feature information included in the plurality of first sample data, the target feature value under the target feature dimension is obtained, wherein the target feature dimension is any one of the N feature dimensions included in the sample feature information, and N is a positive integer;

[0194] Based on the target feature values ​​obtained under the target feature dimension, determine the feature value set corresponding to the target feature dimension;

[0195] The feature value set corresponding to the target feature dimension is subjected to feature value extraction processing to obtain the extracted feature value under the target feature dimension;

[0196] The second sample data is determined based on the extracted feature values ​​under the target feature dimension, and the sample labels included in the second sample data are positive sample labels.

[0197] In one embodiment, the model parameters of the initial type prediction model include weight parameters and bias parameters; when the processing unit 903 adjusts the model parameters of the initial type prediction model based on the difference data, it specifically performs the following steps:

[0198] Based on the difference data, determine the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters in the initial type prediction model;

[0199] The gradient data of the weight parameters and the gradient data of the bias parameters are averaged to obtain gradient mean data, and the gradient data of the weight parameters and the gradient data of the bias parameters are varianced to obtain gradient variance data.

[0200] The gradient mean data and gradient variance data are corrected to obtain corrected gradient mean data and gradient variance data.

[0201] Based on the corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters, and the adjusted gradient data of the bias parameters, the weight parameters and the bias parameters are adjusted.

[0202] Figure 9 The various units in the content push device shown can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of this application. The above units are based on logical function division. In practical applications, the function of one unit can also be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of this application, the blockchain-based data processing device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0203] In one embodiment, the ability to perform such operations can be achieved by running on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). Figure 2 or Figure 7 Computer programs for the steps involved in some or all of the methods shown, to construct, for example... Figure 9 The content push device shown herein, and the content push method for implementing the embodiments of this application, are described. The computer program may be recorded on, for example, a computer-readable storage medium, loaded onto the aforementioned computing device via the computer-readable storage medium, and run therein.

[0204] The content push device provided in this application embodiment can respond to game client requests and perform content push processing at specific times (e.g., after the game scene finishes running, after the login operation is completed, etc.), making the content push more timely; it can perform content push operations based on the object data of the logged-in object, realizing personalized content push processing for the logged-in object, making the content push more in line with the needs of the logged-in object, which is conducive to improving the user experience of the game; it can push content to the logged-in object, thereby increasing the frequency of in-game resource transfer operations, increasing the conversion rate of pushed content, increasing the game participation of the logged-in object, and increasing social interaction between multiple logged-in objects; it can realize the training process of the object type prediction model, effectively improving the model training efficiency and ensuring the prediction accuracy of the object type prediction model.

[0205] Based on the above methods and apparatus embodiments, this application provides a computer device. Please refer to... Figure 10 , Figure 10 This is a structural block diagram of a computer device provided in an embodiment of this application. Figure 10 The computer device shown can be the one described above. Figure 1 Game server 102. Figure 10 The computer device shown includes at least a processor 1001, an input interface 1002, an output interface 1003, and a computer-readable storage medium 1004. The processor 1001, input interface 1002, output interface 1003, and computer-readable storage medium 1004 can be connected via a bus or other means.

[0206] The computer-readable storage medium 1004 can be stored in the memory of a computer device. The computer-readable storage medium 1004 is used to store computer programs, which include computer instructions. The processor 1001 is used to execute the computer program stored in the computer-readable storage medium 1004. The processor 1001 (or CPU (Central Processing Unit)) is the computing and control core of the computer device. It is suitable for implementing computer programs, specifically for loading and executing computer programs to achieve corresponding methods or functions.

[0207] This application also provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space for storing the operating system of the computer device. Furthermore, the storage space also stores computer programs suitable for loading and execution by a processor. It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0208] In a specific implementation, the processor 1001 can load and execute the computer program stored in the computer-readable storage medium 1004 to achieve the aforementioned related... Figure 2 The corresponding steps in the content push method are shown. Specifically, the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 as follows:

[0209] Receive a content push request sent by the game client, the content push request carrying a login object, the login object being the object currently logged into the game client;

[0210] In response to the content push request, the object data of the logged-in object is obtained. The object data includes basic data and game data. The game data includes one or more of the following: game scene description data, game scene statistics data, and game object description data.

[0211] The object data is subjected to feature extraction processing to obtain the feature information of the login object;

[0212] Based on the feature information, object type prediction processing is performed to obtain the object type prediction probability of the login object. The object type prediction probability is the probability that the object type of the login object is the target object type. The probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold.

[0213] Content is pushed to the logged-in object based on the predicted probability of the object type.

[0214] In one embodiment, when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 to perform content push processing on the logged-in object based on the predicted probability of the object type, it is specifically used to perform the following steps:

[0215] The predicted probability of the object type is compared with the second probability threshold corresponding to the login object;

[0216] If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then the target push content associated with the game is determined, and the target push content is pushed to the logged-in object;

[0217] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then no content push processing will be performed on the logged-in object.

[0218] In one embodiment, when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001, it is further configured to perform the following steps:

[0219] If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then the push frequency of the login object is determined based on the number of times content is pushed to the login object within the target time.

[0220] The push frequency of the logged-in object is compared with the push frequency threshold to obtain the push frequency comparison result;

[0221] If the push frequency comparison result indicates that the push frequency of the logged-in object is less than or equal to the push frequency threshold, then the second probability threshold is reduced to obtain a reduced second probability threshold.

[0222] In one embodiment, the basic data includes target location data. When the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 to perform feature extraction processing on the object data to obtain the feature information of the logged-in object, it is specifically used to perform the following steps:

[0223] Obtain location mapping information, which indicates the linear mapping relationship between location data and location feature values ​​under the location feature dimension;

[0224] Based on the target location data, query the location mapping information to determine the location feature value under the location feature dimension corresponding to the target location data.

[0225] In one embodiment, the game data includes game scene description data, which includes a target game duration; when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 to perform feature extraction processing on the object data to obtain the feature information of the logged-in object, it is specifically used to perform the following steps:

[0226] Obtain the duration mapping polynomial, which is used to indicate the nonlinear mapping relationship between game duration and duration feature values ​​under the game duration feature dimension;

[0227] Using the duration mapping polynomial, the target game duration is nonlinearly mapped to obtain the duration feature value under the game duration feature dimension corresponding to the target game duration.

[0228] In one embodiment, when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 to perform object type prediction processing based on the feature information and obtain the object type prediction probability of the logged-in object, it is specifically used to perform the following steps:

[0229] The feature information is input into the object type prediction model for object type prediction processing to obtain the object type prediction probability of the login object.

[0230] The object type prediction model is obtained by adjusting the initial type prediction model using difference data. The difference data is determined based on the sample label and sample probability of the sample feature information. The sample label is used to indicate the probability that the object type of the sample object is the target object type. The sample object is the object corresponding to the sample feature information. The sample probability is obtained by performing object type prediction processing on the sample feature information using the initial type prediction model. The sample feature information is determined based on the object data of the sample object.

[0231] In one embodiment, when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001, it is further configured to perform the following steps:

[0232] Obtain a sample dataset, which includes multiple sample data. The sample data includes sample feature information and sample labels for the sample feature information. The sample feature information with positive sample labels is determined based on the object data of positive sample objects. The sample feature information with negative sample labels is determined based on the object data of negative sample objects. The positive sample objects are those for which no conversion operation was performed on the push content in the first time period and which was performed on the push content in the second time period. The negative sample objects are those for which a conversion operation was performed on the push content in both the first and second time periods, with the first time period being earlier than the second time period.

[0233] The sample feature information included in each sample data in the sample dataset is input into the initial type prediction model for object type prediction processing to obtain the sample probability corresponding to each sample feature information;

[0234] The difference data is determined based on the sample probability corresponding to each of the sample feature information and the sample label of each of the sample feature information;

[0235] The model parameters of the initial type prediction model are adjusted based on the difference data, and the object type prediction model is determined based on the adjusted initial type prediction model.

[0236] In one embodiment, the sample dataset includes multiple first sample data and second sample data, wherein the sample labels included in the first sample data are positive sample labels; when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001, it is further configured to perform the following steps:

[0237] From the sample feature information included in the plurality of first sample data, the target feature value under the target feature dimension is obtained, wherein the target feature dimension is any one of the N feature dimensions included in the sample feature information, and N is a positive integer;

[0238] Based on the target feature values ​​obtained under the target feature dimension, determine the feature value set corresponding to the target feature dimension;

[0239] The feature value set corresponding to the target feature dimension is subjected to feature value extraction processing to obtain the extracted feature value under the target feature dimension;

[0240] The second sample data is determined based on the extracted feature values ​​under the target feature dimension, and the sample labels included in the second sample data are positive sample labels.

[0241] In one embodiment, the model parameters of the initial type prediction model include weight parameters and bias parameters; when the computer program in the computer-readable storage medium 1004 is loaded and executed by the processor 1001 to adjust the model parameters of the initial type prediction model according to the difference data, it is specifically used to perform the following steps:

[0242] Based on the difference data, determine the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters in the initial type prediction model;

[0243] The gradient data of the weight parameters and the gradient data of the bias parameters are averaged to obtain gradient mean data, and the gradient data of the weight parameters and the gradient data of the bias parameters are varianced to obtain gradient variance data.

[0244] The gradient mean data and gradient variance data are corrected to obtain corrected gradient mean data and gradient variance data.

[0245] Based on the corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters, and the adjusted gradient data of the bias parameters, the weight parameters and the bias parameters are adjusted.

[0246] The computer device provided in this application embodiment can respond to requests from game clients and perform content push processing at specific times (e.g., after the game scene finishes running, after the login operation is completed, etc.), making the content push more timely; it can perform content push operations based on the object data of the logged-in object, realizing personalized content push processing for the logged-in object, making the content push more in line with the needs of the logged-in object, which is conducive to improving the user experience of the game; it can push content to the logged-in object, thereby increasing the frequency of in-game resource transfer operations, increasing the conversion rate of pushed content, increasing the game participation of the logged-in object, and increasing social interaction between multiple logged-in objects; it can realize the training process of the object type prediction model, effectively improving the model training efficiency and ensuring the prediction accuracy of the object type prediction model.

[0247] This application also provides a computer-readable storage medium storing computer instructions. When executed on a computer device, the instructions cause the computer device to perform the steps in the various method embodiments of this application to implement the content push method provided in this application. The specific implementation method can be found in the foregoing description and will not be repeated here.

[0248] This application also provides a computer program product, which includes a computer program or computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer program or computer instructions from the computer-readable storage medium and executes the computer program or computer instructions, causing the computer device to perform the steps in the various method embodiments of this application to implement the content push method provided in this application. The specific implementation method can be found in the foregoing description and will not be repeated here.

[0249] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program with a predetermined function, which works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0250] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0251] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A content push method, characterized in that, The method includes: Receive a content push request sent by the game client, the content push request carrying a login object, the login object being the object currently logged into the game client; In response to the content push request, the object data of the logged-in object is obtained. The object data includes basic data and game data. The game data includes one or more of the following: game scene description data, game scene statistics data, and game object description data. The object data is subjected to feature extraction processing to obtain the feature information of the login object; Based on the feature information, object type prediction processing is performed to obtain the object type prediction probability of the login object. The object type prediction probability is the probability that the object type of the login object is the target object type. The probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold. Content is pushed to the logged-in object based on the predicted probability of the object type.

2. The method as described in claim 1, characterized in that, The step of predicting the probability based on the object type and then pushing content to the logged-in object includes: The predicted probability of the object type is compared with the second probability threshold corresponding to the login object; If the comparison result indicates that the predicted probability of the object type is greater than the second probability threshold, then the target push content associated with the game is determined, and the target push content is pushed to the logged-in object; If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then no content push processing will be performed on the logged-in object.

3. The method as described in claim 2, characterized in that, The method further includes: If the comparison result indicates that the predicted probability of the object type is less than or equal to the second probability threshold, then the push frequency of the login object is determined based on the number of times content is pushed to the login object within the target time. The push frequency of the logged-in object is compared with the push frequency threshold to obtain the push frequency comparison result; If the push frequency comparison result indicates that the push frequency of the logged-in object is less than or equal to the push frequency threshold, then the second probability threshold is reduced to obtain a reduced second probability threshold.

4. The method according to any one of claims 1-3, characterized in that, The basic data includes target location data. The feature extraction process performed on the object data to obtain the feature information of the logged-in object includes: Obtain location mapping information, which indicates the linear mapping relationship between location data and location feature values ​​under the location feature dimension; Based on the target location data, query the location mapping information to determine the location feature value under the location feature dimension corresponding to the target location data.

5. The method according to any one of claims 1-3, characterized in that, The game data includes game scene description data, which includes the target game duration; the feature extraction process on the object data to obtain the feature information of the logged-in object includes: Obtain the duration mapping polynomial, which is used to indicate the nonlinear mapping relationship between game duration and duration feature values ​​under the game duration feature dimension; Using the duration mapping polynomial, the target game duration is nonlinearly mapped to obtain the duration feature value under the game duration feature dimension corresponding to the target game duration.

6. The method according to any one of claims 1-3, characterized in that, The step of performing object type prediction processing based on the feature information to obtain the object type prediction probability of the logged-in object includes: The feature information is input into the object type prediction model for object type prediction processing to obtain the object type prediction probability of the login object. The object type prediction model is obtained by adjusting the initial type prediction model using difference data. The difference data is determined based on the sample label and sample probability of the sample feature information. The sample label is used to indicate the probability that the object type of the sample object is the target object type. The sample object is the object corresponding to the sample feature information. The sample probability is obtained by performing object type prediction processing on the sample feature information using the initial type prediction model. The sample feature information is determined based on the object data of the sample object.

7. The method as described in claim 6, characterized in that, The method further includes: Obtain a sample dataset, which includes multiple sample data. The sample data includes sample feature information and sample labels for the sample feature information. The sample feature information with positive sample labels is determined based on the object data of positive sample objects. The sample feature information with negative sample labels is determined based on the object data of negative sample objects. The positive sample objects are those for which no conversion operation was performed on the push content in the first time period and which was performed on the push content in the second time period. The negative sample objects are those for which a conversion operation was performed on the push content in both the first and second time periods, with the first time period being earlier than the second time period. The sample feature information included in each sample data in the sample dataset is input into the initial type prediction model for object type prediction processing to obtain the sample probability corresponding to each sample feature information; The difference data is determined based on the sample probability corresponding to each of the sample feature information and the sample label of each of the sample feature information; The model parameters of the initial type prediction model are adjusted based on the difference data, and the object type prediction model is determined based on the adjusted initial type prediction model.

8. The method as described in claim 7, characterized in that, The sample dataset includes multiple first sample data and second sample data, wherein the sample labels included in the first sample data are positive sample labels; the method further includes: From the sample feature information included in the plurality of first sample data, the target feature value under the target feature dimension is obtained, wherein the target feature dimension is any one of the N feature dimensions included in the sample feature information, and N is a positive integer; Based on the target feature values ​​obtained under the target feature dimension, determine the feature value set corresponding to the target feature dimension; The feature value set corresponding to the target feature dimension is subjected to feature value extraction processing to obtain the extracted feature value under the target feature dimension; The second sample data is determined based on the extracted feature values ​​under the target feature dimension, and the sample labels included in the second sample data are positive sample labels.

9. The method as described in claim 7, characterized in that, The model parameters of the initial type prediction model include weight parameters and bias parameters; adjusting the model parameters of the initial type prediction model based on the difference data includes: Based on the difference data, determine the adjusted gradient data of the weight parameters and the adjusted gradient data of the bias parameters in the initial type prediction model; The gradient data of the weight parameters and the gradient data of the bias parameters are averaged to obtain gradient mean data, and the gradient data of the weight parameters and the gradient data of the bias parameters are varianced to obtain gradient variance data. The gradient mean data and gradient variance data are corrected to obtain corrected gradient mean data and gradient variance data. Based on the corrected gradient mean and gradient variance data, the adjusted gradient data of the weight parameters, and the adjusted gradient data of the bias parameters, the weight parameters and the bias parameters are adjusted.

10. A content push device, characterized in that, include: A receiving unit is used to receive a content push request sent by a game client, wherein the content push request carries a login object, and the login object is the object currently logged into the game client. The acquisition unit is configured to, in response to the content push request, acquire object data of the logged-in object, wherein the object data includes basic data and game data, and the game data includes one or more of game scene description data, game scene statistics data, and game object description data; The processing unit is used to perform feature extraction processing on the object data to obtain the feature information of the login object; The acquisition unit is further configured to perform object type prediction processing based on the feature information to obtain the object type prediction probability of the login object, wherein the object type prediction probability is the probability that the object type of the login object is the target object type; the probability that the target object type object performs a conversion operation on the pushed content is greater than or equal to a first probability threshold. The processing unit is also used to perform content push processing on the logged-in object based on the probability predicted by the object type.

11. A computer device, characterized in that, The computer device includes: A processor is a tool for implementing computer programs. A computer-readable storage medium storing a computer program adapted to be loaded by the processor and to implement the content push method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to implement the content push method as described in any one of claims 1-9.

13. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the content push method as described in any one of claims 1-9.

Images