Message push method and device, computer storage medium, electronic device

Abstract

This application discloses a message push method and apparatus, a computer storage medium, and an electronic device. The method mainly includes: acquiring a first type of data to be processed and a second type of data to be processed from a user; performing feature processing on the first type of data to be processed and the second type of feature data to obtain a joint feature vector corresponding to the two types of data to be processed; obtaining target information based on the joint feature vector, wherein the target information can reflect the difference between the user and other users in a first specified aspect and a second specified aspect; determining whether the user is a target user based on the target information; and pushing a message to the target user if the user is a target user. The target user is a user who meets the corresponding preset requirements in both the first and second specified aspects, so as to improve the accuracy of the push by considering both specified aspects.

Description

Technical Field

[0001] This application relates to the field of push notification technology, and in particular to a push notification method and apparatus, a computer storage medium, and an electronic device. Background Technology

[0002] In game message push, game messages can be pushed to users with high game interest. Users with high game interest are those who are more interested in the game than other users. Game messages such as game news, game gift packs, and game events can be pushed to these users.

[0003] Currently, game message push is based on users' historical game data, which determines the user's interest in multiple games, identifies the target game the user is interested in during the current or future time period, obtains game messages for the target game, and pushes game messages for the target game to the user. However, this method of pushing game messages based on users' historical game data may result in a low degree of matching between game messages and users, reducing the accuracy of the push. Summary of the Invention

[0004] In view of the shortcomings of the prior art, this application provides a message push method and apparatus, a computer storage medium, and an electronic device to solve the problem of low push accuracy.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] On the one hand, this application provides a message push method, including:

[0007] Obtain user information of the user to be processed; wherein, the user information includes basic user information and a game list;

[0008] The user's basic information and the game list are processed to obtain a joint feature vector of the user to be processed. The joint feature vector is used to characterize the user's interest in the target game and the user's payment status in the target game.

[0009] Based on the joint feature vector, the user's interest in the target game and the user's willingness to spend money in the target game are obtained.

[0010] Based on the game interest level and the game payment ability, target users are identified from the pending users, and game messages of the target game are pushed to the target users.

[0011] Optionally, the process of processing the user's basic information and the game list to obtain the joint feature vector of the user to be processed includes:

[0012] Obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list;

[0013] Based on the feature vector of each game in the game list, an associated feature vector of the game list is obtained, which is used to indicate the association relationship between the games in the game list;

[0014] Select a game from the game list as the target game, and obtain the interest feature vector of the target game based on the feature vector of the target game and the associated feature vector of the game list. The interest feature vector of the target game is used to characterize the user's interest in the target game.

[0015] Based on the interest feature vector of the target game and the basic information feature vector, the joint feature vector of the user to be processed is obtained.

[0016] Optionally, the process of processing the user's basic information and the game list to obtain the joint feature vector of the user to be processed includes:

[0017] By utilizing the embedding layer in the shallow network of the prediction model, the basic information feature vector of the user's basic information and the feature vector of each game in the game list are obtained;

[0018] The feature vector of each game in the game list is learned by using the feature transformation structure layer in the shallow network to obtain the associated feature vector of the game list;

[0019] By utilizing the game attention layer in the shallow network, attention learning is performed on the associated feature vectors of the game list and the feature vectors of the target game to obtain the interest feature vector of the target game.

[0020] The interest feature vector and the basic information feature vector are subjected to attention learning using the joint layer in the first top-level network of the prediction model to obtain a first joint feature vector; and the interest feature vector and the basic information feature vector are subjected to attention learning using the joint layer in the second top-level network of the prediction model to obtain a second joint feature vector.

[0021] The joint feature vector includes a first joint feature vector and a second joint feature vector. The first joint feature vector primarily represents the user's interest in the target game. The second joint feature vector primarily represents the user's spending on the target game. The prediction model is trained using the user's basic information and game list from historical users, enabling the shallow network to obtain the interest feature vector and basic information feature vector of any user. This allows the first top-level network to generate and utilize the first joint feature vector of any user to obtain the user's game interest level, and enables the second top-level network to generate and utilize the second joint feature vector of any user to obtain the user's game spending ability.

[0022] Optionally, obtaining the user's interest in the target game and their willingness to spend money in the target game based on the joint feature vector includes:

[0023] The first joint feature vector is input into the fully connected layer in the first top-level network to obtain the game interest degree of the user to be processed in the target game output by the fully connected layer of the first top-level network.

[0024] The second joint feature vector is input into the fully connected layer in the second top-level network to obtain the game payment ability of the user to be processed in the target game output by the fully connected layer of the second top-level network.

[0025] The first top-level network is trained based on the user base information of historical users with first sample labels and the game list of historical users with first sample labels; the second top-level network is trained based on the user base information of historical users with second sample labels and the game list of historical users with second sample labels; the first sample label is used to indicate whether a historical user has registered for a game; the second sample label is the historical game payment amount of a historical user.

[0026] Optionally, the training method for the prediction model includes:

[0027] User registration samples and user payment samples are constructed separately. The user registration sample includes user information and a first sample tag. The user payment sample includes user information and a second sample tag. The user information includes the user's basic information and the user's game list. The first sample tag is used to indicate whether the user has registered for a game. The second sample tag is the user's historical game payment amount.

[0028] The user information from the user registration sample is input into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the first top-level network to be trained to obtain the game interest degree corresponding to the user registration sample. The user information from the user payment sample is input into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the second top-level network to be trained to obtain the game payment amount corresponding to the user payment sample.

[0029] Based on the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, the parameters of the shallow network and the first top-level network are adjusted. Based on the error between the game payment amount corresponding to the user payment sample and the second sample label, the parameters of the shallow network and the second top-level network are also adjusted. Then, the process of inputting the user information from the user registration sample into the shallow network to be trained is repeated until the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and the error between the game payment amount corresponding to the user payment sample and the second sample label, both satisfy the corresponding preset conditions.

[0030] Optional, also includes:

[0031] Run the target game on the electronic device;

[0032] If the target user is detected to have logged into the target game, the target game is redirected to the game interface corresponding to the game message, and the game message is displayed in the game interface.

[0033] On the other hand, this application provides a message push device, including:

[0034] The acquisition unit is used to acquire user information of the user to be processed; wherein, the user information includes basic user information and a game list;

[0035] The processing unit is used to process the user's basic information and the game list to obtain a joint feature vector of the user to be processed. The joint feature vector is used to characterize the user's interest in the target game and the user's payment status in the target game.

[0036] The prediction unit is used to obtain the user's interest in the target game and the user's willingness to spend money in the target game based on the joint feature vector.

[0037] The push unit is used to determine the target user from the users to be processed based on the game interest and the game payment ability, and push game messages of the target game to the target user.

[0038] Optionally, the processing unit includes:

[0039] The feature acquisition unit is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list;

[0040] The extraction unit is used to obtain the associated feature vector of the game list based on the feature vector of each game in the game list. The associated feature vector is used to indicate the association relationship between the games in the game list.

[0041] The selection unit is used to select a game as the target game from the game list, and to obtain the interest feature vector of the target game based on the feature vector of the target game and the associated feature vector of the game list. The interest feature vector of the target game is used to characterize the user's interest in the target game.

[0042] The joint unit is used to obtain the joint feature vector of the user to be processed based on the interest feature vector of the target game and the basic information feature vector.

[0043] Optionally, the processing unit includes:

[0044] An embedding unit is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list by utilizing the embedding layer in the shallow network of the prediction model.

[0045] The first learning unit is used to learn the feature vector of each game in the game list by utilizing the feature transformation structure layer in the shallow network, so as to obtain the associated feature vector of the game list;

[0046] The second learning unit is used to learn attention from the associated feature vectors of the game list and the feature vectors of the target game using the game attention layer in the shallow network, so as to obtain the interest feature vector of the target game.

[0047] The third learning unit is used to perform attention learning on the interest feature vector and the basic information feature vector using the joint layer in the first top-level network of the prediction model to obtain a first joint feature vector, and to perform attention learning on the interest feature vector and the basic information feature vector using the joint layer in the second top-level network of the prediction model to obtain a second joint feature vector.

[0048] The joint feature vector includes a first joint feature vector and a second joint feature vector. The first joint feature vector primarily represents the user's interest in the target game. The second joint feature vector primarily represents the user's spending on the target game. The prediction model is trained using the user's basic information and game list from historical users, enabling the shallow network to obtain the interest feature vector and basic information feature vector of any user. This allows the first top-level network to generate and utilize the first joint feature vector of any user to obtain the user's game interest level, and enables the second top-level network to generate and utilize the second joint feature vector of any user to obtain the user's game spending ability.

[0049] Optionally, the prediction unit includes:

[0050] The first prediction unit is used to input the first joint feature vector into the fully connected layer in the first top-level network to obtain the game interest degree of the user to be processed in the target game output by the fully connected layer in the first top-level network.

[0051] The second prediction unit is used to input the second joint feature vector into the fully connected layer in the second top-level network to obtain the game payment ability of the user to be processed in the target game output by the fully connected layer of the second top-level network.

[0052] The first top-level network is trained based on the user base information of historical users with first sample labels and the game list of historical users with first sample labels; the second top-level network is trained based on the user base information of historical users with second sample labels and the game list of historical users with second sample labels; the first sample label is used to indicate whether a historical user has registered for a game; the second sample label is the historical game payment amount of a historical user.

[0053] Optional, also includes:

[0054] A construction unit is used to construct user registration samples and user payment samples respectively; wherein, the user registration sample includes user information and a first sample tag; the user payment sample includes user information and a second sample tag, the user information includes the user's basic information and the user's game list; the first sample tag is used to indicate whether the user has registered for a game; the second sample tag is the user's historical game payment amount;

[0055] The input unit is used to input user information from the user registration sample into a shallow network to be trained, to obtain the interest feature vector and basic information feature vector output by the shallow network, input the interest feature vector and basic information feature vector output by the shallow network into a first top-level network to be trained, to obtain the game interest degree corresponding to the user registration sample, and input user information from the user payment sample into a shallow network to be trained, to obtain the interest feature vector and basic information feature vector output by the shallow network, and input the interest feature vector and basic information feature vector output by the shallow network into a second top-level network to be trained, to obtain the game payment amount corresponding to the user payment sample;

[0056] The parameter tuning unit is used to adjust the parameters of the shallow network and the first top-level network based on the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and to adjust the parameters of the shallow network and the second top-level network based on the error between the game payment amount corresponding to the user payment sample and the second sample label, and then return to the input unit until the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label and the error between the game payment amount corresponding to the user payment sample and the second sample label both meet the corresponding preset conditions.

[0057] Optional, also includes:

[0058] A running unit for running the target game in an electronic device;

[0059] The jump unit is used to control the target game to jump to the game interface corresponding to the game message if it is detected that the target user has logged into the target game, and to display the game message in the game interface.

[0060] In another aspect, this application provides a computer storage medium for storing a computer program, which, when executed, is used for the aforementioned message push method.

[0061] Furthermore, this application provides an electronic device, comprising:

[0062] Memory and processor;

[0063] The memory is used to store programs;

[0064] The processor is used to execute the program, which, when executed, is specifically used to implement the aforementioned message push method.

[0065] This application provides a message push method and apparatus, a computer storage medium, and an electronic device. The method involves acquiring user information of a user to be processed, including basic user information and a game list. The basic user information and game list are processed to obtain a joint feature vector of the user to be processed. This joint feature vector indicates the user's interest in a target game and their spending status in that game. Based on the joint feature vector, the user's game interest level and spending ability in the target game are obtained. Target users are identified from the user pool based on their game interest level and spending ability, and game messages for the target game are pushed to them. This method considers the target user's game interest level and spending ability when pushing game messages, ensuring that users interested in and willing to spend money on the target game receive the game messages, thus achieving precise message push and improving the accuracy of the push notifications. Attached Figure Description

[0066] 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0067] Figure 1 A flowchart illustrating a message push method provided in an embodiment of this application;

[0068] Figure 2 A flowchart illustrating a method for obtaining joint feature vectors provided in this application embodiment;

[0069] Figure 3 A schematic diagram of the architecture of the prediction model provided in the embodiments of this application;

[0070] Figure 4 This is a schematic diagram of the feature transformation structure layer in the prediction model provided in the embodiments of this application;

[0071] Figure 5 A flowchart illustrating a training method for a prediction model provided in an embodiment of this application;

[0072] Figure 6 A schematic diagram of a sampling space for sample data provided in an embodiment of this application;

[0073] Figure 7 A flowchart illustrating another method for obtaining joint feature vectors provided in this application embodiment;

[0074] Figure 8A schematic diagram of the structure of a message push device provided in another embodiment of this application;

[0075] Figure 9 This is a schematic diagram of the structure of a processing unit provided in another embodiment of this application;

[0076] Figure 10 A schematic diagram of the structure of an electronic device provided in another embodiment of this application;

[0077] Figure 11 This is a schematic diagram illustrating the structure of a distributed system applied to a blockchain system, as provided in an embodiment of this application.

[0078] Figure 12 This is a schematic diagram of the block structure provided in an embodiment of this application. Detailed Implementation

[0079] 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 skilled in the art without creative effort are within the scope of protection of this application.

[0080] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0081] Currently, game message push notifications are based on a user's historical game data, determining their interest in multiple games and pushing game messages accordingly. Other game message pushes could incorporate additional information beyond user interest, such as determining a user's asset level based on their basic information and using that as a basis for game spending power, pushing messages to users with high game interest and high spending power. However, a user's spending power and game interest are closely related; a user's asset level does not necessarily represent their actual game spending power. For example, some users may have high assets but low game interest, while others may have low assets but high interest. Using asset level as the sole indicator of spending power ignores the correlation between spending power and game interest. Therefore, using asset level as the basis for spending power can lead to discrepancies between actual game spending and user behavior. This can result in game messages being pushed to users with high assets but low game interest, while users with low assets but high game interest may not receive the messages, thus reducing the accuracy of the push notifications.

[0082] The message push method provided in this application obtains a joint feature vector representing a user's interest in a target game and their spending status in that target game through basic user information and a game list. Then, based on the joint feature vector, it obtains the user's game interest level and game spending ability in the target game. The method fully considers the correlation between the user's interest in the target game and their spending status when generating game interest level and game spending ability, thereby accurately identifying users who can receive game messages and improving the accuracy of push notifications.

[0083] The message push method provided in this application will be described below with reference to the accompanying drawings. Figure 1 As shown, it illustrates an optional flow of a message push method provided in an embodiment of this application, including the following steps:

[0084] S101. Obtain user information of the user to be processed; whereby, user information includes basic user information and a list of games.

[0085] Basic user information is used to characterize the identity and gaming spending capacity of the user to be processed. For example, basic user information may include the user's identity information and their spending on games. User identity information may include gender, age, education level, etc., while spending information may include the user's asset information and other data reflecting their gaming spending ability. The game list records the games installed on the user's electronic devices, such as game name, game type, and the user's usage patterns, including the number of times the user has used the game. Games with higher usage can receive more game notifications.

[0086] The user information of the user to be processed can be collected from the game server. When the electronic device of the user to be processed is running the game, and when the user to be processed purchases virtual products such as items in the game, the game server will record these behaviors of the user to be processed. Therefore, this embodiment can obtain the user information of the user to be processed from the game server.

[0087] S102. Process the user's basic information and game list to obtain the joint feature vector of the user to be processed. The joint feature vector is used to represent the user's interest in the target game and the user's spending on the target game.

[0088] User basic information can reflect a user's level of interest in a game to some extent. For example, male users generally show higher game interest than female users, and users around 20 years old typically show higher game interest than users of other age groups. Furthermore, user basic information can also reflect a user's spending power in the game; generally, users with higher assets have higher spending power, and users aged 18 to 30 have higher spending power than users of other age groups. Correspondingly, this embodiment can obtain information that characterizes a user's interest in a game and their spending habits based on user basic information.

[0089] Similarly, a user's game list not only reflects their gaming interest but also their spending power. A higher proportion of games in the list that require payment or have multiple paid activities indicates a higher spending power. Conversely, a user's game list primarily consisting of single-player games that don't require in-app purchases suggests a relatively lower spending power. Accordingly, this embodiment can obtain information representing a user's interest in and spending habits regarding a particular game based on their game list.

[0090] This embodiment utilizes user basic information and a game list to obtain a joint feature vector, enabling the joint feature vector to be obtained based on more comprehensive data and improving its accuracy. Furthermore, the joint feature vector represents both the user's interest in the target game and their spending behavior within that game. By using a single feature vector to represent a user's interest and spending behavior in a target game, the relationship between these two aspects can be comprehensively considered to complete game message pushes, thus improving accuracy.

[0091] To facilitate the acquisition of the joint feature vector, vectorization of user basic information and the game list is used to obtain a basic information feature vector and a game list vector. The basic information feature vector includes vectors of each piece of information in the user basic information, and the game list vector includes vectors of each game. Then, using the basic information feature vector and the game list vector, a joint feature vector representing the user's interest in the target game and the user's spending on the target game is obtained.

[0092] For example, embedding (a method of converting discrete variables into continuous vector representations) is used to vectorize user basic information and the game list, resulting in a basic information feature vector and a game list vector. Then, attention mechanisms or self-attention mechanisms are used to learn attention between the basic information feature vector and the game list vector, yielding a joint feature vector. When using attention mechanisms or self-attention mechanisms for attention learning, the feature vector of the target game is also introduced to determine the user's interest in the target game and their spending on it. The target game can be a game in the game list or a game related to a game in the game list, such as a game of the same type as a game in the game list. When determining the target game, the game selected from the game list can be a game with high usage frequency; there is no restriction on the target game here.

[0093] S103. Based on the joint feature vector, obtain the user's interest level in the target game and the user's willingness to spend money in the target game. In this embodiment, the user's interest level in the target game is obtained based on the joint feature vector; the user's willingness to spend money in the target game is obtained based on the joint feature vector.

[0094] For example, the joint feature vector can be input into an activation function. This activation function has two parts: one part focuses on processing the user's interest in the target game, as represented by the joint feature vector, to obtain the game interest level; the other part focuses on processing the user's spending behavior in the target game, as represented by the joint feature vector, to obtain the user's spending ability in the game. The form of the activation function will not be described further in this embodiment. Alternatively, a mapping relationship can be constructed between the joint feature vector and the game interest level and spending ability in the game. Based on this mapping relationship, the joint feature vector can be processed to obtain the user's game interest level and spending ability in the target game.

[0095] S104. Based on game interest and willingness to pay, identify target users from the pending users and push game messages of the target game to them.

[0096] Target users are those who are interested in the target game and willing to pay for it. Optionally, a user is identified as a target user when both their interest in the target game and their willingness to pay for it meet the corresponding first preset conditions. For example, the first preset conditions can be set as minimum values ​​for game interest and willingness to pay. If a user's interest in the target game exceeds the set minimum value, and their willingness to pay also exceeds the set minimum value, then the user is identified as a target user. After a target user receives a game message from the target game, the likelihood of that message being used increases, improving both the accuracy of the push notification and the probability of the message being used.

[0097] If a user is identified as a target user based on their game interest and spending ability, it means that the user is interested in the target game and will make purchases on it (such as buying game items). This type of user can be identified as a recipient of game messages for the target game. If a user is identified as not a target user based on their game interest and spending ability, it means that the user may not be interested in the target game or, although interested, will not make purchases on it. This type of user can be identified as a non-recipient of game messages for the target game.

[0098] Alternatively, target users can be those who have some interest in the target game but spend little money on it. For example, if both the user's interest in the target game and their willingness to spend money on it meet the corresponding second preset conditions, the user is identified as a target user. For instance, the second preset conditions could be set as maximum values ​​for game interest and spending ability. If the user's interest in the target game is less than the set maximum value, and their spending ability is also less than the set maximum value, the user is identified as a target user, and game information about the target game is pushed to them. Even if the target user's interest in the target game is only moderate, the electronic device can still send them promotional messages, discount information, and game screenshots, allowing them to better understand the game and improving its promotion.

[0099] In this embodiment, the target user can configure message settings for the target game. If the target user sets the settings to disable receiving game messages, the electronic device will no longer push game messages for the target game to the target user. If the target user sets the settings to receive game messages, the electronic device will treat the target user as a pending user and use the message push method provided in this embodiment to determine whether the target user is the target user. If the target user is determined to be the target user, the electronic device will continue to push game messages for the target game to the target user.

[0100] Optionally, after performing step S104, the process may further include: running the target game on the electronic device; if it is detected that the target user has logged into the target game, controlling the target game to jump to the game interface corresponding to the game message, and displaying the game message in the game interface.

[0101] For example, if the game message is related to a game gift pack, specifically, it pushes promotional messages about game items to the target user. If it detects that the target user has logged into the target game, it controls the target game to directly jump from the login screen to the game store screen, where the promotional messages about game items are displayed.

[0102] The aforementioned message push method obtains user information of the users to be processed, including basic user information and a game list. The basic user information and game list are processed to obtain a joint feature vector of the users to be processed. This joint feature vector indicates the user's interest in the target game and their spending status in the target game. Based on the joint feature vector, the user's game interest level and spending ability in the target game are obtained. Target users are identified from the users to be processed based on their game interest level and spending ability, and game messages for the target game are pushed to them. This method considers the target user's game interest level and spending ability when pushing game messages, ensuring that users who are interested in and will spend money on the target game receive the game messages, achieving precise push notifications and improving accuracy.

[0103] Optionally, embodiments of this application provide an implementation method for obtaining joint feature vectors, such as... Figure 2 As shown, it includes the following steps:

[0104] S201. Obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list. For example, by using vectorization processing techniques such as Embedding, the user's basic information and the game list are vectorized to obtain the basic information feature vector and the feature vector of each game. The user's basic information and the attribute information of each game are represented by vectors, which facilitates subsequent processing.

[0105] S202. Based on the feature vector of each game in the game list, obtain the associated feature vector of the game list. The associated feature vector is used to indicate the relationship between the games in the game list.

[0106] S203. Select a game from the game list as the target game. Based on the feature vector of the target game and the associated feature vector of the game list, obtain the interest feature vector of the target game. The interest feature vector of the target game is used to characterize the user's interest in the target game.

[0107] S204. Based on the interest feature vector and basic information feature vector of the target game, obtain the joint feature vector of the user to be processed.

[0108] Steps S202 to S204 above can be processed by an attention mechanism. However, the difference lies in the input of the attention mechanism. For example, when obtaining the associated feature vector of the game list, the feature vector of each game in the game list is used as the input, while when obtaining the interest feature vector of the target game, the feature vector of the target game and the associated feature vector of the game list are used as the input. This embodiment will not elaborate on how to process these features.

[0109] Apart from Figure 2 In addition to the method shown for obtaining the joint feature vector of the user to be processed, this embodiment can also obtain the joint feature vector of the user to be processed using a neural network model. The neural network model used in this embodiment is a prediction model, and the structure of the prediction model is as follows: Figure 3 As shown, prediction models can be divided into different categories based on the task: interest prediction networks ( Figure 3 (as shown on the left) and paid prediction network ( Figure 3 (As shown on the right). The interest prediction network includes a shallow network and a first top-level network, while the payment prediction network includes a shallow network and a second top-level network. The two shallow networks are shared networks, meaning they have the same structure and parameters. This is just one possible approach; only one shallow network can be used. The first and second top-level networks have the same structure, based on an attention mechanism, but because they are responsible for different tasks, they are trained separately for those tasks, therefore their parameters are different.

[0110] Optional, see also Figure 3 As shown, the shallow network can be further divided into an embedding layer, a first processing network, and a second processing network. The embedding layer converts each piece of information from the input user basic information into a vector. This vector is then connected by subsequent connection networks of the first processing network to obtain a basic information feature vector, which is then passed to the top-level network, such as the first top-level network and / or the second top-level network.

[0111] The embedding layer transforms the game list into a game list vector. This vector includes the feature vectors of each game in the list. Then, the feature transformation layer (also known as the translation layer) in the second processing network learns the associated feature vectors (or implicit feature vectors) of the game list. Finally, the game attention layer in the second processing network learns attention between the associated feature vectors of the game list and the feature vector of the target game passed from the embedding layer, resulting in the user's interest feature vector for the target game.

[0112] Among them, the feature transformation structure layer provided in the embodiments of this application, such as Figure 4 As shown, it is the same as the existing feature transformation structure layer, so it will not be described again here.

[0113] See also Figure 3As shown, both the first and second top-level networks consist of a joint layer (also called a user attention layer) and a fully connected layer. The user attention layer learns attention from the interest feature vector and basic information feature vector input from the shallow network to obtain the user's joint feature vector. The fully connected layer processes the obtained user joint feature vector to obtain the corresponding predicted values; that is, the first top-level network obtains the user's game interest level, and the second top-level network obtains the user's game spending ability. The fully connected layer includes a multilayer perceptron (MLP) and a corresponding activation function (Sigmoid).

[0114] against Figure 3 The prediction model shown can be trained using historical user information and game lists. This allows the shallow network to obtain the interest feature vector and basic information feature vector for any user. The first top-level network generates and uses the first joint feature vector of any user to determine the user's game interest level. The second top-level network generates and uses the second joint feature vector of any user to determine the user's game spending ability. The first joint feature vector primarily represents the user's interest in the target game, enabling the prediction of the user's game interest level. The second joint feature vector primarily represents the user's spending behavior in the target game, enabling the prediction of the user's game spending ability. The training process of the prediction model is as follows: Figure 5 As shown, the following steps may be included:

[0115] S501. Construct user registration samples and user payment samples respectively.

[0116] The user registration sample includes user information and a first sample tag. The user payment sample includes user information and a second sample tag; the user information includes the basic user information and the list of games played by historical users.

[0117] In other words, the user registration sample is a sample of historical users with the first sample label. For example, the user registration sample includes the user information of historical users with the first sample label, as well as the first sample label itself. The user information specifically includes the historical user's basic user information and a list of games. Basic user information may include gender, age, education level, asset information, etc. The first sample label is used to indicate whether a historical user has registered for a game; one first sample label corresponds to one game in the game list, indicating whether the user has registered for the corresponding game.

[0118] The user payment sample is a sample of historical users with a second sample label. For example, the user payment sample includes user information of historical users with the second sample label, as well as the second sample label itself. The type of user information included in the user payment sample can be the same as the user information included in the user registration sample. The second sample label can be the historical game payment amount of a historical user, such as the historical average daily game payment amount. The average daily game payment amount can be obtained by: calculating the total payment amount of all games in the user's game list over a period of time, and then averaging the total payment amount to obtain the user's average daily game payment amount.

[0119] It should be noted that, as Figure 6 As shown, the user registration sample targets registered users, so it is selected from all users. The user payment sample targets paying users, who have already registered for the game, so it can be selected from registered users.

[0120] S502. Input the user information from the user registration samples into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. Input the interest feature vector and basic information feature vector output by the shallow network into the first top-level network to be trained to obtain the game interest degree corresponding to the user registration sample.

[0121] The shallow network is trained using user information from user registration samples, enabling it to convert user information into interest feature vectors and basic information feature vectors. Then, the first top-level network is trained using the interest feature vectors and basic information feature vectors output by the shallow network, enabling it to obtain game interest level. In the process of obtaining game interest level, the first top-level network pays more attention to the interest feature vectors in the input, making the obtained game interest level more accurate.

[0122] When the interest feature vector and basic information feature vector output by the shallow network are input into the first top-level network to be trained, the interest feature vector and basic information feature vector are input into the joint layer of the first top-level network. The joint layer outputs the first joint feature vector, and then the first joint feature vector is processed by the fully connected layer of the first top-level network to obtain the game interest degree.

[0123] S503. Input the user information from the user payment sample into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. Input the interest feature vector and basic information feature vector output by the shallow network into the second top-level network to be trained to obtain the game payment amount corresponding to the user payment sample.

[0124] In this embodiment, the amount of money a user spends on the game represents their ability to pay for the game. The higher the amount of money spent on the game, the higher their ability to pay; the lower the amount of money spent on the game, the lower their ability to pay.

[0125] The shallow network is trained using user information from user payment samples, enabling it to convert user information into interest feature vectors and basic information feature vectors. Then, the second top-level network is trained using the interest feature vectors and basic information feature vectors output by the shallow network, enabling it to obtain the game payment amount. In the process of obtaining the game payment amount, the second top-level network pays more attention to the basic information feature vectors in the input, making the obtained game payment amount more accurate.

[0126] When the interest feature vector and basic information feature vector output by the shallow network are input into the second top-level network to be trained, the interest feature vector and basic information feature vector are input into the joint layer of the second top-level network. The joint layer outputs the second joint feature vector, and then the fully connected layer of the second top-level network is used to process the second joint feature vector to obtain the game payment amount.

[0127] In this embodiment, the training of the shallow network utilizes user information from both user payment samples and user registration samples. User payment samples can use the user's historical game payment amount as the second sample label, while user registration samples can use the user's game registration as the first sample label. In this way, when the shallow network is trained using user information from these two types of samples, it can comprehensively consider the user's interest in the game and the user's payment in the game, that is, consider the correlation between interest and payment. Compared with considering interest alone or using the user's assets as the ability to pay in the game, this improves accuracy.

[0128] S504. Determine whether both the first error and the second error satisfy the corresponding preset conditions.

[0129] The first error refers to the difference between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label. The second error refers to whether the difference between the game payment amount corresponding to the user payment sample and the second sample label both meet the corresponding preset conditions.

[0130] If either the first error or the second error is determined not to meet the corresponding preset conditions—that is, if the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label does not meet the corresponding preset conditions, or the error between the game payment amount corresponding to the user payment sample and the second sample label does not meet the corresponding preset conditions—then step S505 is executed. If both the first error and the second error are determined to meet the corresponding preset conditions—that is, if the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and the error between the game payment amount corresponding to the user payment sample and the second sample label, both meet the corresponding preset conditions—then step S506 is executed. The preset conditions corresponding to game interest level and game payment amount will not be illustrated in this embodiment.

[0131] S505. Based on the error between the game interest corresponding to the user registration sample and the game interest represented by the first sample label, adjust the parameters of the shallow network and the first top-level network. Based on the error between the game payment amount corresponding to the user payment sample and the second sample label, adjust the parameters of the shallow network and the second top-level network.

[0132] If two shallow networks are set up, in order to ensure the consistency of the parameters of the two shallow networks, parameter adjustments need to be made based on the principle of parameter consistency between the two shallow networks. Furthermore, after executing step S505, the process returns to step S502 to continue adjusting the prediction model using other user registration samples and user payment samples.

[0133] The first top-level network is used to obtain game interest level. Since the first sample label indicates whether a user has registered for a game, the more games a user has registered for, the higher their game interest level. Therefore, the first top-level network can be used to obtain the game interest level of users with the first sample label, and then compared with the game interest level corresponding to the user registration sample. The parameters of the shallow network and the first top-level network are adjusted based on the error obtained from the comparison. The second top-level network is used to obtain game spending ability, such as game spending amount. Since the second sample label indicates the user's historical game spending amount, the higher the user's historical game spending amount, the higher their game spending ability. Therefore, the second top-level network can be used to obtain the game spending amount of users with the second sample label, and then compared with the game spending amount corresponding to the user's payment sample. The parameters of the shallow network and the second top-level network are adjusted based on the error obtained from the comparison.

[0134] S506. End training of the prediction model. After ending training of the prediction model, it can be optimized at intervals to allow the prediction model to change with changes in user registration samples and user payment samples, making the prediction model more in line with the current user interest and spending ability of the game.

[0135] After training and obtaining the prediction model, this embodiment uses the prediction model to obtain the joint feature vector of the user to be processed. An optional process is as follows: Figure 7 As shown, the following steps may be included:

[0136] S701. Use the embedding layer in the shallow network to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list.

[0137] The embedding layers of shallow networks transform the input user information and game list into vector form to facilitate the use and learning optimization of the prediction model. The vector transformation differs depending on the type of feature in the input user information and game list. For example, discrete features are transformed into discrete vectors, while continuous features are transformed into vectors multiplied by continuous values. For instance, gender is transformed into a discrete vector, while spending level is transformed into a vector multiplied by continuous values.

[0138] S702. The feature vector of each game in the game list is learned by using the feature transformation structure layer in the shallow network to obtain the associated feature vector of the game list.

[0139] If the sequence of feature vectors from each game is input into the feature transformation layer of a shallow network, the associated feature vector of the game list can be obtained. This associated feature vector includes the associated feature vectors of each individual game, and the associated feature vector of any game represents the relationship between that game and other games. For example, the relationship can indicate whether game types are related, or whether a user is interested in different types of games, etc.

[0140] S703. Using the game attention layer in the shallow network, attention is learned on the associated feature vectors of the game list and the feature vector of the target game to obtain the interest feature vector of the target game.

[0141] The game attention layer can learn attention from the associated feature vectors of the game list and the feature vector of the target game based on the attention mechanism.

[0142] The attention mechanism is represented as follows:

[0143]

[0144] When learning using the attention mechanism in the game attention layer, Q in the above expression is the feature vector of the target game, and K and V are both associated feature vectors of the game list; d kThis indicates the dimension of the vectors mentioned above. In addition to learning attention between the associated feature vectors and the target game's feature vectors based on an attention mechanism, the game attention layer can also learn attention between the associated feature vectors and the target game's feature vectors based on a self-attention mechanism. Of course, other mechanisms can also be used, such as forward mechanisms, but these will not be elaborated upon in this embodiment.

[0145] S704. Using the joint layer in the first top-level network, attention learning is performed on the interest feature vector and the basic information feature vector to obtain the first joint feature vector. And using the joint layer in the second top-level network, attention learning is performed on the interest feature vector and the basic information feature vector to obtain the second joint feature vector.

[0146] The joint layers in the first and second top-level networks can also learn interest feature vectors and basic information feature vectors based on attention mechanisms. For example:

[0147]

[0148] X represents the interest feature vector of the target game, while Y and Z are both user basic information vectors; p k This represents the dimension of the vectors mentioned above. Although the joint layer in the first and second top-level networks can be processed based on an attention mechanism, the different parameters of the two joint layers result in different outputs. Besides attention mechanisms, processing can also be performed using self-attention or forward mechanisms, which will not be elaborated upon in this embodiment.

[0149] Correspondingly, see also Figure 7 In this embodiment of the application, the prediction model is used to obtain the user's interest in the target game and the user's willingness to spend money in the target game, including:

[0150] S705. Input the first joint feature vector into the fully connected layer in the first top-level network to obtain the user's game interest in the target game output by the fully connected layer in the first top-level network.

[0151] S706. Input the second joint feature vector into the fully connected layer in the second top-level network to obtain the user's game payment ability in the target game output by the fully connected layer in the second top-level network.

[0152] The first top-level network is trained based on the user base information of historical users with the first sample label and the game list of historical users with the first sample label. The first sample label is used to indicate whether the historical user has registered for a game. Therefore, the joint layer of the first top-level network can output the first joint feature vector that emphasizes game interest, and the fully connected layer in the first top-level network can use the first joint feature vector to obtain the user's game interest in the target game.

[0153] Similarly, the second top-level network is trained based on the user base information of historical users with the second sample label and the game list of historical users with the second sample label. The second sample label is the historical game payment amount of the historical user. Therefore, the joint layer of the second top-level network can output a second joint feature vector that focuses on representing the game payment ability. Furthermore, the fully connected layer in the second top-level network can use the second joint feature vector to obtain the user's game payment ability in the target game.

[0154] In this embodiment, the prediction model can be trained using user information of users who register for the game and user information of users who pay for the game. The trained prediction model can comprehensively consider users' interest in the game and users' ability to pay for the game. When using this prediction model to obtain the user's interest in the target game and their ability to pay for the target game, it can also comprehensively consider their interest in the game and their ability to pay for the target game, thereby improving accuracy.

[0155] This application provides a message push device, such as... Figure 8 As shown, it includes the following units:

[0156] The acquisition unit 801 is used to acquire user information of the user to be processed.

[0157] The user information includes basic user information and a list of games.

[0158] The processing unit 802 is used to process the user's basic information and the game list to obtain the joint feature vector of the user to be processed. The joint feature vector is used to characterize the user's interest in the target game and the user's payment status in the target game.

[0159] The prediction unit 803 is used to obtain the user's interest in the target game and the user's willingness to spend money in the target game based on the joint feature vector.

[0160] Push unit 804 is used to identify target users from the pending users based on their game interest and willingness to pay, and to push game messages of the target game to the target users.

[0161] Optionally, in another embodiment of the message push device provided in this application, the processing unit, such as Figure 9 As shown, it includes:

[0162] The feature acquisition unit 901 is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list.

[0163] Extraction unit 902 is used to obtain the associated feature vector of the game list based on the feature vector of each game in the game list. The associated feature vector is used to indicate the relationship between the games in the game list.

[0164] The selection unit 903 is used to select a game from the game list as the target game. Based on the feature vector of the target game and the associated feature vector of the game list, the interest feature vector of the target game is obtained. The interest feature vector of the target game is used to represent the user's interest in the target game.

[0165] The joint unit 904 is used to obtain the joint feature vector of the user to be processed based on the interest feature vector and basic information feature vector of the target game.

[0166] Optionally, in another embodiment of the message push device provided in this application, the processing unit includes:

[0167] An embedding unit is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list by utilizing the embedding layer in the shallow network of the prediction model.

[0168] The first learning unit is used to learn the feature vector of each game in the game list by utilizing the feature transformation structure layer in the shallow network, so as to obtain the associated feature vector of the game list.

[0169] The second learning unit is used to learn the attention of the associated feature vectors of the game list and the feature vector of the target game by utilizing the game attention layer in the shallow network, so as to obtain the interest feature vector of the target game.

[0170] The third learning unit is used to perform attention learning on the interest feature vector and the basic information feature vector using the joint layer in the first top-level network of the prediction model to obtain the first joint feature vector, and the joint layer in the second top-level network of the prediction model to perform attention learning on the interest feature vector and the basic information feature vector to obtain the second joint feature vector.

[0171] The joint feature vector includes a first joint feature vector and a second joint feature vector. The first joint feature vector primarily represents the user's interest in the target game. The second joint feature vector represents the user's spending behavior in the target game. The prediction model is trained using historical users' basic information and game lists. This allows the shallow network to obtain the interest feature vector and basic information feature vector for any user, enables the first top-level network to generate and utilize the first joint feature vector to determine the user's game interest level, and enables the second top-level network to generate and utilize the second joint feature vector to determine the user's game spending ability.

[0172] Optionally, in another embodiment of the message push device provided in this application, the prediction unit includes:

[0173] The first prediction unit is used to input the first joint feature vector into the fully connected layer in the first top-level network to obtain the user's game interest in the target game output by the fully connected layer in the first top-level network.

[0174] The second prediction unit is used to input the second joint feature vector into the fully connected layer in the second top-level network to obtain the user's game payment ability in the target game output by the fully connected layer in the second top-level network.

[0175] Specifically, when training the first top-level network, it is based on the user base information of historical users with first sample labels and their game lists. When training the second top-level network, it is based on the user base information of historical users with second sample labels and their game lists; the first sample label indicates whether a historical user has registered for a game; the second sample label represents the historical user's historical game spending amount.

[0176] Optionally, in another embodiment of the message push device provided in this application, the device further includes:

[0177] The building unit is used to build user registration samples and user payment samples respectively.

[0178] The user registration sample includes user information and a first sample label; the user payment sample includes user information and a second sample label. The user information includes the user's basic information and the user's game list. The first sample label is used to indicate whether the user has registered for a game. The second sample label is the user's historical game payment amount.

[0179] The input unit is used to input user information from user registration samples into the shallow network to be trained, obtaining the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the first top-level network to be trained to obtain the game interest level corresponding to the user registration sample. Similarly, the input unit is used to input user information from user payment samples into the shallow network to be trained, obtaining the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the second top-level network to be trained to obtain the game payment amount corresponding to the user payment sample.

[0180] The parameter tuning unit is used to adjust the parameters of the shallow network and the first top-level network based on the error between the game interest corresponding to the user registration sample and the game interest represented by the first sample label, and to adjust the parameters of the shallow network and the second top-level network based on the error between the game payment amount corresponding to the user payment sample and the second sample label, and then return to the input unit until the error between the game interest corresponding to the user registration sample and the game interest represented by the first sample label, and the error between the game payment amount corresponding to the user payment sample and the second sample label, both meet the corresponding preset conditions.

[0181] Optionally, in another embodiment of the message push device provided in this application, the device further includes:

[0182] A runtime unit used to run a target game on an electronic device.

[0183] The jump unit is used to control the target game to jump to the game interface corresponding to the game message if the target user is detected to have logged into the target game, and to display the game message in the game interface.

[0184] It should be noted that the specific working process of each unit provided in the above embodiments of this application can be referred to the corresponding steps in the above method embodiments, and will not be repeated here.

[0185] Another embodiment of this application provides a computer storage medium for storing a computer program, which, when executed, implements the message push method provided in any of the above embodiments.

[0186] Computer storage media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0187] Another embodiment of this application provides an electronic device, such as... Figure 10 As shown, it includes:

[0188] Memory 1001 and processor 1002.

[0189] The memory 1001 is used to store a program, and the processor 1001 is used to execute the program stored in the memory 1001. When the program is executed, it is specifically used to implement any message push method provided in any of the above embodiments.

[0190] The system involved in the above message push method can be a distributed system formed by connecting clients and multiple nodes (any form of computing device in the network, such as servers and user terminals) through network communication.

[0191] Taking a distributed system as an example, see blockchain system. Figure 11 , Figure 11 This is an optional structural diagram of the distributed system 100 provided in this application embodiment applied to a blockchain system. It consists of multiple nodes (any form of computing device accessing the network, such as servers or user terminals) and clients, forming a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol running on top of the Transmission Control Protocol (TCP). In the distributed system, any machine, such as a server or user terminal, can join and become a node. A node includes a hardware layer, a middleware layer, an operating system layer, and an application layer. Servers, user terminals, and clients can execute the message push method provided in this embodiment.

[0192] See Figure 11 The functions of each node in the blockchain system shown include:

[0193] 1) Routing: A basic function of nodes used to support communication between nodes.

[0194] In addition to routing capabilities, nodes can also have the following functions:

[0195] 2) Applications are deployed in the blockchain to implement specific business needs. They record data related to the implementation of functions to form record data, carry digital signatures in the record data to indicate the source of the task data, and send the record data to other nodes in the blockchain system. When other nodes successfully verify the source and integrity of the record data, they add the record data to a temporary block.

[0196] 3) A blockchain consists of a series of blocks that are sequentially generated. Once a new block is added to the blockchain, it will not be removed. The blocks contain the data submitted by the nodes in the blockchain system.

[0197] See Figure 12 , Figure 12 This is an optional schematic diagram of the block structure provided in this application embodiment. Each block includes the hash value of the transaction records stored in this block (the hash value of this block) and the hash value of the previous block. The blocks are connected through their hash values ​​to form a blockchain. Additionally, the block may include information such as a timestamp when it was generated. A blockchain is essentially a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains relevant information used to verify the validity of the information (anti-counterfeiting) and to generate the next block.

[0198] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. 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.

[0199] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0200] It should be noted that the user information (including but not limited to basic user information, user identity information, user asset information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

Claims

1. A message push method, characterized in that, include: Obtain user information of the user to be processed; wherein, the user information includes basic user information and a game list; The user's basic information and the game list are processed using a neural network model to obtain a joint feature vector of the user to be processed. The joint feature vector is used to characterize the user's interest in the target game, the user's payment status in the target game, and the correlation between the interest and the payment status. Based on the joint feature vector and the correlation, the user's interest in the target game and the user's willingness to spend money on the target game are obtained. Based on the game interest level and the game payment ability, target users are identified from the users to be processed, and game messages of the target game are pushed to the target users. The neural network model is a prediction model; the prediction model includes an interest prediction network and a payment prediction network; the interest prediction network and the payment prediction network share a shallow network, or the shallow network of the interest prediction network and the shallow network of the payment prediction network are shared networks. The joint layer in the first top-level network of the interest prediction network is used to perform attention learning on the interest feature vector of the user to be processed for the target game and the basic information feature vector of the user's basic information, so as to obtain a first joint feature vector that emphasizes the interest of the user to be processed for the target game. The joint layer in the second top-level network of the payment prediction network is used to perform attention learning on the interest feature vector and the basic information feature vector to obtain a second joint feature vector that focuses on characterizing the payment status of the user to be processed in the target game; the joint feature vector includes the first joint feature vector and the second joint feature vector.

2. The method according to claim 1, characterized in that, The process of using a neural network model to process the user's basic information and the game list to obtain the joint feature vector of the user to be processed includes: Obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list; Based on the feature vector of each game in the game list, an associated feature vector of the game list is obtained, which is used to indicate the association relationship between the games in the game list; Select a game from the game list as the target game, and obtain the interest feature vector of the target game based on the feature vector of the target game and the associated feature vector of the game list. The interest feature vector of the target game is used to characterize the user's interest in the target game. Based on the interest feature vector of the target game and the basic information feature vector, the joint feature vector of the user to be processed is obtained.

3. The method according to claim 1 or 2, characterized in that, The process of using a neural network model to process the user's basic information and the game list to obtain the joint feature vector of the user to be processed includes: By utilizing the embedding layer in the shallow network of the prediction model, the basic information feature vector of the user's basic information and the feature vector of each game in the game list are obtained; The feature vector of each game in the game list is learned by using the feature transformation structure layer in the shallow network to obtain the associated feature vector of the game list; By utilizing the game attention layer in the shallow network, attention learning is performed on the associated feature vectors of the game list and the feature vectors of the target game to obtain the interest feature vector of the target game. The interest feature vector and the basic information feature vector are subjected to attention learning using the joint layer in the first top-level network of the prediction model to obtain a first joint feature vector; and the interest feature vector and the basic information feature vector are subjected to attention learning using the joint layer in the second top-level network of the prediction model to obtain a second joint feature vector. The prediction model is trained using the user base information and game list of historical users, enabling the shallow network to obtain the interest feature vector and basic information feature vector of any user, enabling the first top-level network to generate and use the first joint feature vector of any user to obtain the user's game interest level, and enabling the second top-level network to generate and use the second joint feature vector of any user to obtain the user's game payment ability.

4. The method according to claim 3, characterized in that, The step of obtaining the user's interest in the target game and their willingness to spend money in the target game based on the joint feature vector and the correlation includes: The first joint feature vector is input into the fully connected layer in the first top-level network to obtain the game interest degree of the user to be processed in the target game output by the fully connected layer of the first top-level network. The second joint feature vector is input into the fully connected layer in the second top-level network to obtain the game payment ability of the user to be processed in the target game output by the fully connected layer of the second top-level network. The first top-level network is trained based on the user base information of historical users with first sample labels and the game list of historical users with first sample labels; the second top-level network is trained based on the user base information of historical users with second sample labels and the game list of historical users with second sample labels; the first sample label is used to indicate whether a historical user has registered for a game; the second sample label is the historical game payment amount of a historical user.

5. The method according to claim 3, characterized in that, The training method for the prediction model includes: User registration samples and user payment samples are constructed separately. The user registration sample includes user information and a first sample tag. The user payment sample includes user information and a second sample tag. The user information includes the user's basic information and the user's game list. The first sample tag is used to indicate whether the user has registered for a game. The second sample tag is the user's historical game payment amount. The user information from the user registration sample is input into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the first top-level network to be trained to obtain the game interest degree corresponding to the user registration sample. The user information from the user payment sample is input into the shallow network to be trained to obtain the interest feature vector and basic information feature vector output by the shallow network. The interest feature vector and basic information feature vector output by the shallow network are then input into the second top-level network to be trained to obtain the game payment amount corresponding to the user payment sample. Based on the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, the parameters of the shallow network and the first top-level network are adjusted. Based on the error between the game payment amount corresponding to the user payment sample and the second sample label, the parameters of the shallow network and the second top-level network are also adjusted. Then, the process of inputting the user information from the user registration sample into the shallow network to be trained is repeated until the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and the error between the game payment amount corresponding to the user payment sample and the second sample label, both satisfy the corresponding preset conditions.

6. The method according to claim 1, characterized in that, Also includes: Run the target game on the electronic device; If the target user is detected to have logged into the target game, the target game is redirected to the game interface corresponding to the game message, and the game message is displayed in the game interface.

7. A message push device, characterized in that, include: The acquisition unit is used to acquire user information of the user to be processed; wherein, the user information includes basic user information and a game list; The processing unit is used to process the user's basic information and the game list using a neural network model to obtain a joint feature vector of the user to be processed. The joint feature vector is used to characterize the user's interest in the target game, the user's payment status in the target game, and the correlation between the interest and the payment status. The prediction unit is used to obtain the user's interest in the target game and the user's willingness to spend money in the target game based on the joint feature vector and the correlation relationship. The push unit is used to determine the target user from the users to be processed based on the game interest and the game payment ability, and push game messages of the target game to the target user. The neural network model is a prediction model; the prediction model includes an interest prediction network and a payment prediction network; the interest prediction network and the payment prediction network share a shallow network, or the shallow network of the interest prediction network and the shallow network of the payment prediction network are shared networks. The joint layer in the first top-level network of the interest prediction network is used to perform attention learning on the interest feature vector of the user to be processed for the target game and the basic information feature vector of the user's basic information, so as to obtain a first joint feature vector that emphasizes the interest of the user to be processed for the target game. The joint layer in the second top-level network of the payment prediction network is used to perform attention learning on the interest feature vector and the basic information feature vector to obtain a second joint feature vector that focuses on characterizing the payment status of the user to be processed in the target game; the joint feature vector includes the first joint feature vector and the second joint feature vector.

8. The apparatus according to claim 7, characterized in that, The processing unit includes: The feature acquisition unit is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list; The extraction unit is used to obtain the associated feature vector of the game list based on the feature vector of each game in the game list. The associated feature vector is used to indicate the association relationship between the games in the game list. The selection unit is used to select a game as the target game from the game list, and to obtain the interest feature vector of the target game based on the feature vector of the target game and the associated feature vector of the game list. The interest feature vector of the target game is used to characterize the user's interest in the target game. The joint unit is used to obtain the joint feature vector of the user to be processed based on the interest feature vector of the target game and the basic information feature vector.

9. The apparatus according to claim 7 or 8, characterized in that, The processing unit includes: An embedding unit is used to obtain the basic information feature vector of the user's basic information and the feature vector of each game in the game list by utilizing the embedding layer in the shallow network of the prediction model. The first learning unit is used to learn the feature vector of each game in the game list by utilizing the feature transformation structure layer in the shallow network, so as to obtain the associated feature vector of the game list; The second learning unit is used to learn attention from the associated feature vectors of the game list and the feature vectors of the target game using the game attention layer in the shallow network, so as to obtain the interest feature vector of the target game. The third learning unit is used to perform attention learning on the interest feature vector and the basic information feature vector using the joint layer in the first top-level network of the prediction model to obtain a first joint feature vector, and to perform attention learning on the interest feature vector and the basic information feature vector using the joint layer in the second top-level network of the prediction model to obtain a second joint feature vector. The prediction model is trained using the user base information and game list of historical users, enabling the shallow network to obtain the interest feature vector and basic information feature vector of any user, enabling the first top-level network to generate and use the first joint feature vector of any user to obtain the user's game interest level, and enabling the second top-level network to generate and use the second joint feature vector of any user to obtain the user's game payment ability.

10. The apparatus according to claim 9, characterized in that, The prediction unit includes: The first prediction unit is used to input the first joint feature vector into the fully connected layer in the first top-level network to obtain the game interest degree of the user to be processed in the target game output by the fully connected layer in the first top-level network. The second prediction unit is used to input the second joint feature vector into the fully connected layer in the second top-level network to obtain the game payment ability of the user to be processed in the target game output by the fully connected layer of the second top-level network. The first top-level network is trained based on the user base information of historical users with first sample labels and the game list of historical users with first sample labels; the second top-level network is trained based on the user base information of historical users with second sample labels and the game list of historical users with second sample labels; the first sample label is used to indicate whether a historical user has registered for a game; the second sample label is the historical game payment amount of a historical user.

11. The apparatus according to claim 9, characterized in that, Also includes: A construction unit is used to construct user registration samples and user payment samples respectively; wherein, the user registration sample includes user information and a first sample tag; the user payment sample includes user information and a second sample tag, the user information includes the user's basic information and the user's game list; the first sample tag is used to indicate whether the user has registered for a game; the second sample tag is the user's historical game payment amount; The input unit is used to input user information from the user registration sample into a shallow network to be trained, to obtain the interest feature vector and basic information feature vector output by the shallow network, input the interest feature vector and basic information feature vector output by the shallow network into a first top-level network to be trained, to obtain the game interest degree corresponding to the user registration sample, and input user information from the user payment sample into a shallow network to be trained, to obtain the interest feature vector and basic information feature vector output by the shallow network, and input the interest feature vector and basic information feature vector output by the shallow network into a second top-level network to be trained, to obtain the game payment amount corresponding to the user payment sample; The parameter tuning unit is used to adjust the parameters of the shallow network and the first top-level network based on the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and to adjust the parameters of the shallow network and the second top-level network based on the error between the game payment amount corresponding to the user payment sample and the second sample label. It then returns to the step of inputting user information from the user registration sample into the shallow network to be trained, until the error between the game interest level corresponding to the user registration sample and the game interest level represented by the first sample label, and the error between the game payment amount corresponding to the user payment sample and the second sample label, both satisfy the corresponding preset conditions.

12. The apparatus according to claim 7, characterized in that, Also includes: A running unit for running the target game in an electronic device; The jump unit is used to control the target game to jump to the game interface corresponding to the game message if it is detected that the target user has logged into the target game, and to display the game message in the game interface.

13. A computer storage medium, characterized in that, Used to store a computer program, which, when executed, is used to implement the message push method as described in any one of claims 1 to 6.

14. An electronic device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is used to execute the program, which, when executed, is specifically used to implement the message push method as described in any one of claims 1 to 6.

Images