WO2020156389A1

WO2020156389A1 - Information pushing method and device

Info

Publication number: WO2020156389A1
Application number: PCT/CN2020/073593
Authority: WO
Inventors: 何强; 杨欣豫; 杜思良; 项亮; 王灿
Original assignee: 北京字节跳动网络技术有限公司
Priority date: 2019-01-30
Filing date: 2020-01-21
Publication date: 2020-08-06
Also published as: CN109872242B; CN109872242A

Abstract

An information pushing method and device, wherein the method comprises: acquiring a social network of a target user (201), wherein the social network comprises behavior information of users in the social network and indication information indicating whether an association relationship is established among the users; on the basis of the behavior information of the target user and the behavior information of the user in the social network who has not established the association relationship with the target users, determining the probability of establishing an association relationship between the target user and the user who has not established an association relationship (202); pushing user identifiers of the users corresponding to the probability exceeding a preset threshold to the target user (203). The method can more accurately predict the degree of interest of the target user to other users, and therefore improve the information pushing efficiency.

Description

Information pushing method and device

Cross references to related applications

This application is filed based on the Chinese patent application with the application number 201910092332.X, the filing date of January 30, 2019, and the name "Information Push Method and Device", and claims the priority of the Chinese patent application. The entire content is hereby incorporated into this application as a reference.

Technical field

The embodiments of the present disclosure relate to the field of computer technology, in particular to an information push method and device.

Background technique

With the popularization of terminal equipment and the rapid development of mobile terminal development technology, various client applications have emerged. These client applications can usually include social applications. Through social applications, users can interact with other users.

In related technologies, other users who have not established an associated relationship with the user are usually pushed to the user through the above-mentioned social applications based on the user's own attribute characteristics (such as age, gender, hobbies, and habitual residence address).

Summary of the invention

The embodiments of the present disclosure propose information pushing methods and devices.

In the first aspect, the embodiments of the present disclosure provide an information push method. The method includes: obtaining a social network including a target user. The social network includes behavior information of users located in the social network and is used to indicate whether or not between users Instruction information for establishing an association relationship; based on the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user in the social network, determine the probability of establishing an association relationship between the target user and users who have not established an association relationship; The user ID of the user corresponding to the probability exceeding the preset threshold is pushed to the target user.

In some embodiments, based on the behavior information of the target user and the behavior information of other users who have not established an association relationship with the target user in the social network, the probability of establishing an association relationship between the target user and a user who has not established an association relationship is determined, including : The behavior information of the target user and the behavior information of the user who has not established an association relationship with the target user are respectively input into the pre-trained feature value determination model to obtain the target user behavior information feature value corresponding to the target user's behavior information, and the target user Other user behavior information feature values corresponding to the behavior information of users that have not established an association relationship; based on the target user behavior information feature value and other user behavior information feature values, determine the target user and the user who has not established an association relationship to establish an association relationship Probability.

In some embodiments, the feature value determination model is obtained by training in the following steps: Obtain a training sample set, wherein each training sample in the training sample set includes behavior information of a specified user, behavior information of other users, and instructions for other users Information indicating whether to establish an association relationship with a specified user; perform the following training steps: For the training samples in the training sample set, input the behavior information of the specified user and the behavior information of other users in the training sample into the neural network to be trained, and obtain Behavior information feature values corresponding to the specified user, behavior information feature values corresponding to other users; based on the behavior information feature values corresponding to the specified user and the behavior information feature values corresponding to other users, determine the probability of the specified user establishing an association relationship with other users Value; based on the probability value corresponding to the training sample in the determined training sample set, determine whether the preset loss function converges; in response to determining that the preset loss function converges, determine the feature value to determine that the model training is completed.

In some embodiments, the method further includes: in response to determining that the preset loss function does not converge, using a back propagation algorithm to update the parameters of the neural network to be trained, and continue to perform the training step.

In some embodiments, the behavior information of other users in the training sample set includes one of the following: behavior information of a positive sample user that has an association relationship with a specified user, and behavior information of a negative sample user that has not established an association relationship with the specified user. , Negative sample users are users who are randomly selected from the preset user set, have not established an association relationship with the specified user, and have established an association relationship with other users.

In some embodiments, the training samples in the training sample set also include time stamps corresponding to the behavior information of other users; wherein the time stamps corresponding to the behavior information of the positive sample user are based on the association between the specified user and the positive sample user The time is determined; the time stamp corresponding to the behavior information of the negative sample user is randomly selected from the time stamp corresponding to the positive sample user.

In some embodiments, the neural network includes an n-layer cascaded single-layer feedforward neural network and a fully connected layer. For the i-th layer feedforward neural network, the output parameters of the feedforward neural network of the previous layer are used as the feedforward of the next layer The input parameters of the neural network, n is a positive integer greater than 1, and i is a positive integer greater than 2 and less than n; and each layer of feedforward neural network performs feature extraction through the following steps: based on the first preset activation function, specify the user The behavior information characteristics of other users, the behavior information characteristics of other users, and the timestamps corresponding to the behavior information of other users, determine the correlation index between the behavior information of the specified user and the behavior information of other users; based on the determined correlation index and the second The preset activation function determines the parameters used to indicate the behavior information characteristics of the specified user; the obtained parameters used to indicate the behavior information characteristics of the specified user are used as the output of the feedforward neural network of the current layer.

In some embodiments, the behavior information includes at least one of the following: browsing information and publishing information. The browsing information is generated based on browsing information published by other users who have established an association relationship with the user.

In a second aspect, an embodiment of the present disclosure provides an information push device, the device includes: an acquiring unit configured to acquire a social network including a target user, the social network including behavior information of users located in the social network and Indication information indicating whether an association relationship is established between users; the determining unit is configured to determine the target user and the non-association relationship based on the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user in the social network The probability of establishing an association relationship between users; the pushing unit is configured to push the user identification of the user corresponding to the probability exceeding the preset threshold to the target user.

In some embodiments, the determining unit includes: an input sub-unit configured to input the behavior information of the target user and the behavior information of users that have not established an association relationship with the target user into the pre-trained feature value determination model to obtain The target user behavior information feature value corresponding to the user's behavior information, and other user behavior information feature values corresponding to the behavior information of users who have not established an association relationship with the target user; the determining sub-unit is configured to be based on the target user behavior information feature value, other The characteristic value of user behavior information determines the probability of establishing an association relationship between the target user and the user who has not established an association relationship.

In some embodiments, the step of training to obtain the feature value determination model further includes: in response to determining that the preset loss function does not converge, using a back propagation algorithm to update the parameters of the neural network to be trained, and continuing to perform the training step.

In a third aspect, embodiments of the present disclosure provide a terminal device, the terminal device includes: one or more processors; a storage device for storing one or more programs; when one or more programs are used by one or more Execution by two processors, so that one or more processors implement the method described in any implementation manner of the first aspect.

In a fourth aspect, embodiments of the present disclosure provide a computer-readable medium on which a computer program is stored, and the computer program, when executed by a processor, implements the method described in any implementation manner in the first aspect.

The information pushing method and device provided by the embodiments of the present disclosure determine the target user and the unassociated relationship based on the degree of association between the target user's behavior information and the behavior information of other users who have not established an association relationship with the target user Establish the probability of the association relationship between the users, and then push the user ID of the user corresponding to the probability exceeding the preset threshold to the target user, so as to more accurately predict the target user’s interest in other users, thereby improving the information Push efficiency.

Description of the drawings

By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, purposes and advantages of the present disclosure will become more apparent:

Fig. 1 is an exemplary system architecture diagram in which an embodiment of the present disclosure can be applied;

Fig. 2 is a flowchart of an embodiment of an information push method according to the present disclosure;

FIG. 3 is a flowchart of another embodiment of the information pushing method according to the present disclosure;

Fig. 4 is a flow of an optional implementation method of the training method for determining a model according to the feature value of the present disclosure;

Fig. 5 is a flow chart of an alternative implementation of the method for feature extraction according to each layer of feedforward neural network of the present disclosure;

Fig. 6 is a schematic diagram of an application scenario of an information push method according to an embodiment of the present disclosure;

Fig. 7 is a schematic structural diagram of an embodiment of an information pushing device according to the present disclosure;

Fig. 8 is a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present disclosure.

detailed description

The present disclosure will be further described in detail below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described here are only used to explain the related invention, but not to limit the invention. In addition, it should be noted that, for ease of description, only the parts related to the relevant invention are shown in the drawings.

It should be noted that the embodiments in the present disclosure and the features in the embodiments can be combined with each other if there is no conflict. Hereinafter, the present disclosure will be described in detail with reference to the drawings and in conjunction with embodiments.

FIG. 1 shows an exemplary architecture 100 to which an embodiment of the information pushing method or information pushing device of the present disclosure can be applied.

As shown in FIG. 1, the system architecture 100 may include

terminal devices

101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the

terminal devices

101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

Various client applications may be installed on the

terminal devices

101, 102, 103. For example, web browser applications, video applications, content sharing applications, social applications, etc. The

terminal devices

101, 102, and 103 can interact with the server 105 through the network 104 to receive or send messages and so on.

The

terminal devices

101, 102, and 103 may be hardware or software. When the

terminal devices

101, 102, 103 are hardware, they may be various electronic devices that can receive user operations, including but not limited to smart phones, tablet computers, e-book readers, laptop computers, desktop computers, and so on. When the

terminal devices

101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, multiple software or software modules used to provide distributed services), or as a single software or software module. No specific restrictions are made here.

The server 105 may be a background server that supports client applications installed on the

terminal devices

101, 102, 103. The server 105 can analyze the acquired browsing information and published information of the target user, and determine the characteristic value of the user's browsing information and the characteristic value of the published information, and then based on the comparison between the characteristic value of the browsing information and the characteristic value of the published information among users Similarity, the user ID of one user is pushed to another user.

It should be noted that the server 105 may be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules (for example, multiple software or software modules for providing distributed services), or as a single software or software module. No specific restrictions are made here.

It should be noted that the information pushing method provided by the embodiments of the present disclosure is generally executed by the

terminal devices

101, 102, 103, and correspondingly, the information pushing device is generally set in the

terminal devices

101, 102, 103.

It should be understood that the numbers of terminal devices, networks, and servers in FIG. 1 are merely illustrative. According to implementation needs, there can be any number of terminal devices, networks and servers.

Continuing to refer to FIG. 2, it shows a process 200 of an embodiment of the information pushing method according to the present disclosure. The information push method includes the following steps:

Step 201: Obtain the social network of the target user.

In this embodiment, the social network is usually a network structure composed of users and an association relationship between users. The association relationship between users may be, for example, blood relationship, friend relationship, colleague relationship, relationship with common interests and hobbies. A user located in a social network can serve as a node of the social network. In other words, a social network can be formed by many user nodes. The social network may include users who have established an association relationship with the user, and may also include other users who have not established an association relationship with the user. For example, user A, user B, and user C all belong to network nodes in the same social network. An association relationship is established between user A and user B, an association relationship is established between user B and user C, and an association relationship is not established between user A and user C. As an example, a communication application can be used as a social network. When users use the terminal application, they usually need to register their account in the terminal application. That is to say, if the user joins the social network, the account registration has been performed. Of users can be seen as nodes of social networks. As another example, a certain communication application can also form multiple social networks according to different categories. For example, a social network based on geographic information, a social network based on interest information, etc.

In this embodiment, the aforementioned communication applications may include, but are not limited to, social applications, blog applications, and the like. Through this communication application, users can communicate with other users, such as voice communication, video communication, short message communication, etc. Through this communication application, users can also post information such as articles, micro-messages, and comments, and browse information such as articles, micro-messages, and comments published by other users. The above information such as communication information, publishing information, and browsing information is the user's behavior information.

In this embodiment, the execution subject of the information push method (the server 105 shown in FIG. 1) can obtain the social network to which the target user belongs. Here, the social network to which the user belongs may include one social network or multiple social networks. Specifically, the above-mentioned execution subject may first determine a social network associated with the user tag based on a preset user tag of the target user, and determine the social network as a social network including the target user. The preset user tag may be determined based on user attribute information of the user, or may be selected by the user based on multiple user tags provided by the foregoing execution subject. Here, the acquired social network includes indication information for indicating whether an association relationship is established between users and behavior information of each user. Here, the behavior information includes information posted by other users that are browsed, that is, browsing information, and the posted information, that is, posted information.

Step 202: Determine the probability of establishing an association relationship between the target user and the user who has not established an association relationship based on the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user in the social network.

In this embodiment, the above-mentioned association relationship may be a mutual friend relationship, and may be a following relationship and a followed relationship. By using the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user in the social network, the probability of establishing an association relationship between the target user and the user who has not established an association relationship can be determined.

Specifically, the probability of establishing an association relationship between the target user may be determined based on the browsing information of the target user and the posted information of users who have not established an association relationship with the target user in the social network. Here, the feature value of the browsing information corresponding to the browsing information of the target user and the feature value of the posted information corresponding to the posted information of the user who has not established an association relationship with the target user can be determined.

Generally, users can browse multiple pieces of information or post multiple pieces of information. Each piece of information browsed or published may include a preset tag word, and the preset tag word may be a keyword extracted from the information content, or a tag word set by the information publisher when publishing the information. The preset tag word can represent the category to which the piece of information belongs. The above-mentioned execution subject may determine the multiple categories to which the browsing information of the target user belongs based on the preset tag words corresponding to the browsing information of the target user. Then, each category is regarded as a one-dimensional feature, so as to finally form a feature value of browsing information including multi-dimensional features. In the same way, the above-mentioned executive body may determine the category to which the posted information belongs based on the preset tag words corresponding to the posted information of the users who have not established an association relationship with the target user. Then, each category is regarded as a one-dimensional feature, thereby finally forming a feature value of the posted information including a multi-dimensional feature vector.

Then, the above-mentioned execution subject may determine the probability of establishing an association relationship between two users based on the degree of association between the browsing information feature value corresponding to the target user and the published information feature value corresponding to the user who has not established an association relationship. Specifically, the sigmoid function can be used to calculate the correlation between the two. Take the probability that the target user pays attention to other users who are not followed as an example to elaborate. See Formula 1 and Formula 2 for details.

Among them, p _ij is the probability of target user i following user j. σ(x) is the sigmoid function, and x is a random variable. Here, the random variable is

The matrix of eigenvalues,

Is the eigenvalue matrix of the browsing information of the target user,

Is the matrix of the eigenvalues of other users' published information, and b is a random constant. The random variable x is

Eigenvalue matrix for browsing information

Column vector and eigenvalue matrix of published information

The column vectors of are merged into a new column vector.

In this embodiment, the greater the value of the probability p _ij , the greater the probability that the target user pays attention to other users, that is, the greater the probability of establishing an association relationship between two users.

Step 203: Push the user identifier of the user corresponding to the probability exceeding the preset threshold to the target user.

In this embodiment, after each user joins the aforementioned social network, the aforementioned executive subject usually records a user identifier for uniquely identifying the user's identity. The user identification may be a terminal device number, a registration number used for account registration, or a screen name, avatar, etc. used by the user for account registration. It is worth noting here that the aforementioned user identification may also be a public reading platform such as an official account. Generally, the articles and other information published by each official account can be regarded as a user's release based on platforms such as social applications. Therefore, public reading platforms such as official accounts are used to uniquely instruct the public to read The identity of the platform can also be regarded as a user identity.

In this embodiment, the above-mentioned execution subject may determine the probability of establishing an association relationship between the target user determined in step 202 and the user who has not established an association relationship, and then calculate the probability of the probability value exceeding the preset threshold of the user corresponding to the relationship. The user ID is pushed to the target user. Therefore, the target user can select the operation of adding the user corresponding to the user ID as a friend, follow the user operation corresponding to the user ID, and so on based on the user ID pushed by the received server.

This embodiment uses the feature values of the browsing information of the target user and the feature values of the published information of other users to determine whether the target user is interested in the content published by other users to establish an association relationship between the two, which can make the push more targeted , To avoid pushing many other users who are not strongly related or users are not interested to the target user, which improves the probability of establishing an association relationship between the target user and the pushed user, thereby improving the push efficiency.

With further reference to FIG. 3, it shows a flow 300 of another embodiment of the information pushing method. The process 300 of the information push method includes the following steps:

Step 301: Obtain a social network including the target user.

In this embodiment, the execution subject of the information push method (the server 105 shown in FIG. 1) can obtain the social network to which the target user belongs. Here, the social network to which the user belongs may include one social network or multiple social networks. Specifically, the above-mentioned execution subject may first determine a social network associated with the user tag based on a preset user tag of the target user, and determine the social network as a social network including the target user.

Here, the acquired social network includes indication information for indicating whether an association relationship is established between users, and behavior information of each user. Here, the behavior information includes information posted by other users that are browsed, that is, browsing information, and the posted information, that is, posted information.

Step 302: Input the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user into the pre-trained feature value determination model to obtain the target user behavior information feature value corresponding to the target user's behavior information, and Other user behavior information feature values corresponding to the behavior information of users whose target users have not established an association relationship.

In this embodiment, the aforementioned behavior information may be browsing information or publishing information. Among them, the behavior information of the target user is browsing information, and the behavior information of users who have not established an associated relationship with the target user is publishing information. The browsing information may be the title of the browsed content, may be a keyword extracted from the browsed content, or may be a tag word preset when the user publishes the content. The published information may be the title of the published content, may be a keyword extracted from the published content, or may be a tag word preset when the user publishes the content.

In this embodiment, the feature value determination model is used to predict the user's point of interest based on the user's behavior information, so as to obtain the behavior information feature value based on the behavior information. The above-mentioned feature value determination model is obtained based on neural network training. The neural network may be a deep neural network, and the deep neural network may include multiple feature extraction layers and one fully connected layer. The feature extraction layer can be used to extract features of behavior information. Finally, the features extracted from each layer are fully connected in the fully connected layer to obtain the final behavior information feature value.

In this embodiment, the target user behavior information feature value may be a browsing information feature value obtained based on feature extraction of the browsing information of the target user. Specifically, the browsing information of the target user may be input to the aforementioned pre-trained feature value determination model, so as to obtain the browsing information feature value indicating the browsing interest of the target user.

In this embodiment, other user information feature values corresponding to users who have not established an association relationship with the target user may be posted information feature values obtained based on feature extraction of other users' posted information. Specifically, the posted information of other users may be input to the aforementioned pre-trained feature value determination model, so as to obtain the feature values of other user posted information used to indicate the posting interests of other users.

Step 303: Based on the target user behavior information characteristic value and other user behavior information characteristic values, determine the probability of establishing an association relationship between the target user and a user who has not established an association relationship.

In this embodiment, according to the feature value of the browsing information of the target user determined in step 302 and the feature value of the posted information of the user who has not established an association relationship with the target user, it can be determined that the relationship between the target user and the user who has not established an association relationship is established. The probability of association. For the specific implementation of this step, reference may be made to the related description in step 202 of the embodiment shown in FIG. 2, which will not be repeated here.

Step 304: Push the user identifier of the user corresponding to the probability exceeding the preset threshold to the target user.

For the specific implementation and beneficial effects of step 301 and step 304 shown in this embodiment, reference may be made to the relevant description of step 201 and step 203 shown in FIG. 2, and details are not repeated here.

It can be seen from the embodiment shown in FIG. 3 that, unlike the embodiment shown in FIG. 2, this embodiment discloses a step of using a pre-trained neural network to determine the characteristic value of behavior information. Since the neural network can process a large amount of data and learn the characteristics of various behavior information of users, it can make the determined characteristic values of behavior information more accurate.

In some optional implementation manners of the foregoing embodiment, the foregoing feature value determination model may be obtained by training based on a sample set. Continuing to refer to FIG. 4, it shows a flow 400 of an alternative implementation of the training method of the feature value determination model provided by the present disclosure. The process 400 includes the following steps:

Step 401: Obtain a training sample set.

Here, each training sample in the training sample set includes user information of a designated user, behavior information of other users, and indication information for indicating whether other users have established an association relationship with the designated user.

Here, the behavior information can include browsing information and publishing information. Wherein, the browsing information is generated based on browsing information published by other users who have established an association relationship with the user.

In some optional implementations, the behavior information of other users in the training sample set includes one of the following: behavior information of a positive sample user that has established an association relationship with the specified user, and behavior information of a negative sample user that has not established an association relationship with the specified user information. Among them, the negative sample user is a user who is randomly selected from the preset user set, has not established an association relationship with the specified user, and has established an association relationship with other users.

Specifically, after the above-mentioned executive body detects that a certain user has established an association relationship with other users, the user identifier corresponding to the user may be added to the above-mentioned preset user set. Add it every time it is detected. That is to say, after the association relationship between the user A and the user B is established, the user identification corresponding to the user A and the user identification corresponding to the user B can be added to the aforementioned preset user set. When it is detected that the association relationship between the user A and the user C is established, the user identification corresponding to the user A and the user identification corresponding to the user C are added to the aforementioned preset user set. In other words, the preset user set contains two user IDs corresponding to user A. When the above-mentioned executive body selects a negative sample user, it is selected from the above-mentioned preset user information set. Therefore, the more user identities of a certain user included in the aforementioned preset user set, the greater the probability of being selected as a negative sample user. By determining the negative samples in the training sample set in the above manner, the negative samples can be made more universal, so that the trained model has better generalization ability.

In some optional implementation manners, the behavior information includes at least one of the following: browsing information and publishing information. The browsing information is generated based on browsing information published by other users who have established an association relationship with the user.

In some optional implementations, the training samples in the training sample set also include a time stamp corresponding to the behavior information; the time stamp corresponding to the behavior information of the positive sample user is determined based on the time when the specified user establishes an association relationship with the positive sample user The time stamp corresponding to the behavior information of the negative sample user is randomly selected from the time stamp corresponding to the positive sample user.

Specifically, the above-mentioned executive body may select the positive sample users who have established an association relationship with the designated user before the designated time t, and then set a time stamp for each positive sample user. Specifically, the timestamp may be a multi-dimensional time feature vector, where the one-dimensional time feature vector is used to indicate the time feature of the specified user and the positive sample user, and the one-dimensional time feature vector is used to instruct the specified user to browse the positive sample user The time characteristics of the published information. The closer the time for establishing the association relationship is to the specified time t, the greater the influence of the information released by the positive sample users on the interest of the specified user in the current time period. Then, the above-mentioned set timestamp can be randomly assigned to the negative sample users. In other words, the behavior information corresponding to each positive sample user or the behavior information corresponding to the negative sample user in the sample training set is set with a time stamp.

Here, the browsing information and publication information of the specified user are obtained based on the time stamp of the positive sample user corresponding to the specified user, that is, the positive sample user who browses before the timestamp and corresponds to the time stamp is published information. The above-mentioned browsing information and published information of other users are based on the timestamp corresponding to the negative sample user, and the information browsed and published by the other users before the selected timestamp. By adding time stamp information, the training samples can be time-sensitive. In this way, the weight of the browsing information can be determined according to the length of the browsing time corresponding to the browsing information from the current time, and the weight of the published information can be determined according to the length of the publishing time corresponding to the published information from the current time, so that the trained The feature value determination model has better timeliness, that is, it can better reflect the user's current interest features, so that the determined feature values of browsing information and published information are more accurate.

In some optional implementation manners, the execution subject may preprocess the acquired behavior information of the specified user and the behavior information of other users. Here, the pre-processing can be standardized processing of behavior information. The normalization processing may be, for example, norm normalization processing, maximum and minimum value normalization processing, or the like. Through standardized processing of behavior information, the learning speed of machine learning can be quickly promoted and the learning efficiency of machine learning can be improved.

Step 402, based on the training sample set obtained in step 401, the following training steps may be performed:

Step 4021: For the training samples in the training sample set, input the behavior information of the specified user and the behavior information of other users in the training sample into the neural network to be trained to obtain the characteristic values of the behavior information corresponding to the specified user and correspond to other users The characteristic value of behavior information.

Here, the behavior information characteristic value may include the browsing information characteristic value and the publishing information characteristic value.

Step 4022: Based on the behavior information feature value corresponding to the specified user and the behavior information feature value corresponding to other users, determine the probability value of the specified user establishing an association relationship with other users.

Specifically, the association relationship may be a following relationship and a followed relationship. Here, the probability value of the designated user's attention to other users can be determined based on the characteristic value of the browsing information of the designated user and the characteristic value of the published information corresponding to other users. Or, based on the feature value of the posted information of the specified user and the feature value of the browsing information of other users, the probability value of the specified user being followed by other users is determined.

Here, the specific realization of the above probability value can refer to the specific realization of determining the probability of establishing an association relationship between the target user and the user who has not established an association relationship shown in step 203 of the embodiment in FIG. 2, which will not be repeated here.

Step 4023: Determine whether the preset loss function converges based on the probability value corresponding to the training sample in the determined training sample set.

In this embodiment, the aforementioned preset loss function may be a logarithmic loss function, for example. Here, determining whether the preset loss function converges is to determine whether the loss value of the loss function reaches the preset threshold, or whether the change in the loss value is absolutely less than the preset threshold. In response to the loss value reaching the preset threshold, or the absolute value of the change of the loss value is less than the preset threshold, it may be determined that the preset loss function converges. It is worth noting here that the absolute value of the aforementioned loss value change is based on the absolute value of the difference between the loss value calculated using the loss function in the current training and the loss value obtained in the previous training.

In step 403, in response to determining that the preset loss function converges, it is determined that the eigenvalues determine that the model training is completed.

In this embodiment, according to whether the preset loss function determined in step 4023 converges, when the preset loss function converges, it can be determined that the above-mentioned feature value determination model training is completed.

Step 404, in response to determining that the preset loss function does not converge, use the back propagation algorithm to update the parameters of the neural network to be trained, and continue to perform the training step shown in step 402.

In this embodiment, the above-mentioned updating the parameters of the neural network to be trained can be, for example, updating the value of the filter of each layer of the neural network in the neural network to be trained, the size of the filter, the step size, etc., and the neural network can also be updated. The number of layers. In response to determining that the preset loss function has not converged, the above-mentioned execution subject may use the direction propagation algorithm to update the parameters of the neural network to be trained, and then continue to perform the training steps shown in step 4021-step 4023.

The feature value determination model trained by the above training method can make the determined behavior information feature value more accurate, and can also improve the robustness of the trained feature value determination model.

In some embodiments, the aforementioned neural network for training the feature value determination model may include n-layer cascaded single-layer feedforward neural network and a fully connected layer. n is a positive integer greater than 1. Among them, for the i-th layer of feedforward neural network, the output parameter of the previous layer of feedforward neural network is used as the input parameter of the next layer of feedforward neural network, and i is a positive integer greater than 2 and less than n. Here, each layer of feedforward neural network is used for feature extraction. In other words, the output of the last layer of feedforward neural network includes the feature extraction results of the previous layers of feedforward neural network. Specifically, as shown in FIG. 5, it shows a process 500 of an optional implementation manner of the method for feature extraction of each layer of the feedforward neural network provided by the present disclosure. The process 500 includes the following steps:

Step 501: Determine the specified user based on the first preset activation function, the behavior information feature of the specified user, the behavior information feature of other users, the time stamp corresponding to the behavior information of other users, and the time stamp corresponding to the behavior information of other users. The correlation index between the behavior information of other users and the behavior information of other users.

Specifically, the above-mentioned association relationship is the relationship between following and being followed. The behavior information feature of the specified user may be the browsing information feature of the specified user, and the behavior information feature of other users may be the published information feature of other users.

Based on the first preset activation function, the timestamp corresponding to the behavior information of other users, the browsing information characteristics of the specified user, and the published information characteristics of other users concerned by the specified user, the browsing information corresponding to the specified user and the specified user's information can be determined. The correlation index between the posted information corresponding to other users concerned is taken as the first correlation index.

The first relevance index is used to indicate the degree of influence of the published information of other users followed by the specified user on the interest characteristics of the specified user. The first preset activation function may be, for example, a ReLU activation function. Because the information browsed by the user in the preset time period may include multiple pieces of information. Therefore, the browsing information can be converted into a feature vector with multi-dimensional features, and each dimensional feature component is used to represent a piece of browsing information. This layer of feedforward neural network can perform feature extraction on each dimensional feature component of the feature vector used to characterize browsing information to obtain a feature matrix corresponding to the browsing information. In the same way, published information can also be converted into a feature vector with multi-dimensional features, and each dimensional feature component is used to represent a piece of published information. This layer of feedforward neural network can perform feature extraction on each dimensional feature component of the feature vector used to characterize the posted information, and obtain the feature matrix corresponding to the posted information. Here, each piece of browsing information is set with a timestamp. Therefore, based on the time stamps corresponding to each browsing information, a time stamp matrix can be formed. For the setting of the time stamp, reference may be made to the related description of setting the time stamp shown in step 4201. Then, the time influence features are divided into a preset number of levels according to the order of distance from the current time. That is to say, the closer the time stamp is to the current time, the higher the corresponding level, that is, the greater the time impact. In this way, a time characteristic matrix for indicating the degree of time influence of each piece of browsing information is formed. Finally, this layer of feedforward neural network can substitute the feature matrix corresponding to the browsing information of the specified user, the feature matrix and the time feature matrix corresponding to the posted information of other users that the specified user is concerned with, into the ReLU activation function, and finally get the specified user The correlation index between the corresponding browsing information and the posted information corresponding to other users that the user is concerned about is designated as the first correlation index.

Here, the behavior information feature of the specified user may also be the published information feature of the specified user, and the behavior information feature of other users may also be the browsing information feature of other users browsing the information published by the specified user.

It can also be based on the first preset activation function, the time stamp corresponding to the behavior information of other users, the characteristics of the published information of the specified user, and the characteristics of the browsing information of other users who follow the specified user. The correlation index between the browsing information corresponding to other users concerned is used as the second correlation index.

Here, the second correlation index is used to indicate the degree of influence of the information posted by the specified user on the interest characteristics of other users who follow the specified user. This layer of feedforward neural network can generate a feature matrix corresponding to the posted information of a specified user, a feature matrix and a time feature matrix corresponding to the browsing information of other users who pay attention to the specified user. Then, this layer of feedforward neural network can substitute the generated feature matrix corresponding to the posted information of the specified user, feature matrix and time feature matrix corresponding to the browsing information of other users who follow the specified user into the ReLU activation function, and finally get The correlation index between the posted information corresponding to the designated user and the browsing information corresponding to other users that the designated user pays attention to is used as the second correlation index, and the correlation index is used as the second correlation index.

It is worth noting here that the behavior information characteristics of the first-layer feedforward neural network are the acquired behavior information of the specified user and the behavior information of other users. The behavior information characteristics of the second layer to the nth layer of the feedforward neural network are the parameters output by the previous layer of the feedforward neural network to indicate the behavior information characteristics of the specified user. When the behavior information is browsing information and publishing information, the behavior information features include browsing information features and publishing information features.

Step 502, based on the determined correlation index and the second preset activation function, determine a parameter for indicating the behavior information feature of the specified user.

Specifically, based on the obtained first association index and the second preset activation function, a parameter used to indicate the characteristics of the browsing information of the specified user may be determined. Based on the obtained second correlation index and the second preset activation function, the parameters used to indicate the characteristics of the published information of the specified user may be determined.

Specifically, the second preset activation function may be a Sigmoid function.

Here, the obtained first correlation index can be normalized. Then, the normalized first association index is used as the coefficient of the feature matrix corresponding to the browsing information of the specified user and the coefficient of the feature matrix corresponding to the posted information of other users, respectively, to obtain the transformed feature corresponding to the browsing information of the specified user The matrix and the transformed feature matrix corresponding to the posted information of other users. The sum of the obtained feature matrix corresponding to the browsing information of the specified user after the transformation and the feature matrix corresponding to the post information of other users after the transformation is used as the variable of the second preset activation function, so as to obtain the browsing information used to indicate the specified user Information feature parameters.

Here, the obtained second correlation index can be normalized. Then, the normalized second association index is used as the coefficients of the feature matrix corresponding to the posted information of the specified user and the coefficients of the feature matrix corresponding to the browsing information of other users to obtain the transformed features corresponding to the posted information of the specified user. The matrix, the feature matrix corresponding to the transformed browsing information of other users. The sum of the obtained feature matrix corresponding to the post information of the specified user after the transformation and the feature matrix corresponding to the browsing information of other users after the transformation is used as the variable of the second preset activation function, thereby obtaining the post for indicating the specified user Information feature parameters.

Step 503: Use the obtained parameter indicating the behavior information feature of the specified user as the output of the feedforward neural network of the current layer.

Specifically, the obtained parameter indicating the characteristics of the posted information of the specified user and the parameter indicating the characteristics of the browsing information of the specified user may be used as the output of the current layer feedforward neural network.

Here, the obtained parameters used to indicate the characteristics of the published information of the specified user and the obtained parameters used to indicate the characteristics of the browsing information of the specified user may be used as the output of the current layer feedforward neural network. In other words, the parameters obtained in the current layer are used as the input of the feedforward neural network of the next layer connected to this layer.

The feedforward neural network of each layer uses the parameters extracted by the above-mentioned feature extraction method to determine the user’s final browsing information feature parameters and published information feature parameters based on the degree of influence of other users’ published information or browsing information on the specified user, so that each The features extracted by a layer of feedforward neural network are more accurate, thereby improving the accuracy of the feature values extracted by the trained feature value determination model.

Further refer to FIG. 6, which is a schematic diagram of an application scenario of the information pushing method according to this embodiment.

In the application scenario of FIG. 6, the social network 601 is a social network including user A, where user A is a target user. The social network 601 also includes user B, user C, and user D. Among them, user A and user B follow each other, and user B follows user C and user D. There is no association relationship between user A, user C, and user D. User A has viewed “How the Wormhole is Formed” and “Football World Cup” published by User B. The information released by user C includes "European Cup Highlights" and "Einstein's Theory of Relativity", and the information released by user D includes "travel diary" and so on. Then, the electronic device 602 on which the information push method runs can determine the target user A and user C, based on the information published by user B viewed by user A, based on the information published by user C, and based on the information published by user D. The probability of user D establishing an association relationship. Among them, the probability that the user A and the user C establish an association relationship is 0.8, and the probability that the user A and the user D establish an association relationship is 0.2. When the preset probability threshold is 0.5, the electronic device 602 can push the user identification of the user C to the user A.

With further reference to FIG. 7, as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of an information push device. The device embodiment corresponds to the method embodiment shown in FIG. 2, and the device can be specifically applied to Various electronic devices.

As shown in FIG. 7, the information pushing device 700 provided in this embodiment includes an acquiring unit 701, a determining unit 702, and a pushing unit 703. Wherein, the obtaining unit 701 is configured to obtain a social network including the target user, the social network including behavior information of users located in the social network and indication information for indicating whether an association relationship is established between the users; the determining unit 702 is configured to Based on the behavior information of the target user and the behavior information of users who have not established an association relationship with the target user in the social network, the probability of establishing an association relationship between the target user and the user who has not established an association relationship is determined 703. The push unit is configured to exceed The user ID of the user corresponding to the probability of the preset threshold is pushed to the target user.

In this embodiment, in the information pushing device 700: the specific processing of the acquiring unit 701, the determining unit 702, and the pushing unit 703 and the technical effects brought by them can be referred to step 201, step 202, and step 202 in the corresponding embodiment in FIG. 2 respectively. The description of step 203 will not be repeated here.

In some optional implementations of this embodiment, the determining unit 702 includes: an input subunit (not shown in the figure) configured to combine the behavior information of the target user and the behavior of users that have not established an association relationship with the target user. The information is respectively input to the pre-trained feature value determination model to obtain the target user behavior information feature value corresponding to the target user's behavior information, and other user behavior information feature values corresponding to the behavior information of users who have not established an association relationship with the target user; determine; The subunit (not shown in the figure) is configured to determine the probability of establishing an association relationship between the target user and a user who has not established an association relationship based on the target user behavior information characteristic value and other user behavior information characteristic values.

In some optional implementations of this embodiment, the feature value determination model is obtained by training in the following steps: obtaining a training sample set, where each training sample in the training sample set includes behavior information of a specified user and behaviors of other users Information, instruction information used to indicate whether other users have established an association relationship with a specified user; perform the following training steps: for the training samples in the training sample set, input the behavior information of the specified user and the behavior information of other users in the training sample into The neural network to be trained obtains the behavior information feature value corresponding to the specified user and the behavior information feature value corresponding to other users; based on the behavior information feature value corresponding to the specified user and the behavior information feature value corresponding to other users, it is determined that the specified user and The probability value of the association relationship established by other users; based on the determined probability value corresponding to the training sample in the training sample set, determine whether the preset loss function converges; in response to determining that the preset loss function converges, the feature value is determined to determine that the model training is completed.

In some optional implementations of this embodiment, the step of training to obtain the feature value determination model further includes: in response to determining that the preset loss function does not converge, using a backpropagation algorithm to update the parameters of the neural network to be trained, and continue execution Training steps.

In some optional implementations of this embodiment, the behavior information of other users in the training sample set includes one of the following: the behavior information of the positive sample user that has an association relationship with the specified user, and the negative behavior information that has not established an association relationship with the specified user. The behavior information of the sample user, where the negative sample user is a user who is randomly selected from the preset user set and has not established an association relationship with the specified user and has established an association relationship with other users.

In some optional implementations of this embodiment, the training samples in the training sample set also include time stamps corresponding to the behavior information of other users; wherein, the time stamps corresponding to the behavior information of the positive sample user are based on the specified user and The time when the positive sample user establishes the association relationship is determined; the time stamp corresponding to the behavior information of the negative sample user is randomly selected from the time stamp corresponding to the positive sample user.

In some optional implementations of this embodiment, the neural network includes n-layer cascaded single-layer feedforward neural network and a fully connected layer. For the i-th layer of feedforward neural network, the output of the previous layer of feedforward neural network The parameters are used as the input parameters of the feedforward neural network of the next layer, n is a positive integer greater than 1, and i is a positive integer greater than 2 and less than n; and each layer of feedforward neural network performs feature extraction through the following steps: Preset the activation function, the behavior information characteristics of the specified user, the behavior information characteristics of other users, and the time stamp corresponding to the behavior information of other users to determine the correlation index between the behavior information of the specified user and the behavior information of other users; The determined correlation index and the second preset activation function determine the parameters used to indicate the behavior information characteristics of the specified user; the obtained parameters used to indicate the behavior information characteristics of the specified user are used as the output of the current layer feedforward neural network.

In some optional implementations of this embodiment, the behavior information includes at least one of the following: browsing information and publishing information. The browsing information is generated based on browsing information published by other users who have an association relationship with the user.

The information push device provided by the embodiment of the present disclosure determines the target user and the user who has not established an association relationship based on the association degree between the target user's behavior information and the behavior information of other users who have not established an association relationship with the target user. The probability of an association relationship is established, and then the user ID of the user corresponding to the probability exceeding the preset threshold is pushed to the target user, so that the degree of interest of the target user to other users can be predicted more accurately, thereby improving the efficiency of information push .

Referring now to Fig. 8, which shows a schematic structural diagram of an electronic device (for example, the terminal device in Fig. 1) 800 suitable for implementing the embodiments of the present disclosure. Terminal devices in the embodiments of the present disclosure may include, but are not limited to, mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablets), PMPs (portable multimedia players), vehicle-mounted terminals ( For example, mobile terminals such as car navigation terminals and fixed terminals such as digital TVs and desktop computers. The terminal device shown in FIG. 8 is only an example, and should not bring any limitation to the function and scope of use of the embodiments of the present disclosure.

As shown in FIG. 8, the electronic device 800 may include a processing device (such as a central processing unit, a graphics processor, etc.) 801, which may be loaded into a random access device according to a program stored in a read-only memory (ROM) 802 or loaded from a storage device 808 The program in the memory (RAM) 803 executes various appropriate actions and processes. In the RAM 803, various programs and data required for the operation of the electronic device 800 are also stored. The processing device 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input/output (I/O) interface 805 is also connected to the bus 804.

Generally, the following devices can be connected to the I/O interface 805: including input devices 806 such as touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, liquid crystal display (LCD), speakers, vibration An output device 807 such as a device; a storage device 808 such as a magnetic tape and a hard disk; and a communication device 809. The communication device 809 may allow the electronic device 800 to perform wireless or wired communication with other devices to exchange data. Although FIG. 8 shows an electronic device 800 having various devices, it should be understood that it is not required to implement or have all the illustrated devices. It may alternatively be implemented or provided with more or fewer devices. Each block shown in FIG. 8 may represent one device, or may represent multiple devices as needed.

In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network through the communication device 809, or installed from the storage device 808, or installed from the ROM 802. When the computer program is executed by the processing device 801, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

It should be noted that the computer-readable medium described in the embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In the embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device. In the embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as a part of a carrier wave, and a computer-readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium. The computer-readable signal medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device . The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to: wire, optical cable, RF (Radio Frequency), etc., or any suitable combination of the above.

The above-mentioned computer-readable medium may be included in the above-mentioned terminal device; or it may exist alone without being assembled into the terminal device. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: obtains the social network including the target user, and the social network includes the user's information in the social network. Behavior information and indication information used to indicate whether an association relationship between users is established; based on the behavior information of the target user, and the behavior information of users who have not established an association relationship with the target user in the social network, determine the target user and those who have not established an association relationship. The probability of establishing an association relationship between users; the user ID of the user corresponding to the probability exceeding the preset threshold is pushed to the target user.

The computer program code for performing the operations of the embodiments of the present disclosure can be written in one or more programming languages or a combination thereof. The programming languages include object-oriented programming languages—such as Java, Smalltalk, C++, and Conventional procedural programming language-such as "C" language or similar programming language. The program code can be executed entirely on the user's computer, partly on the user's computer, executed as an independent software package, partly on the user's computer and partly executed on a remote computer, or entirely executed on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user’s computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to Connect via the Internet).

The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions, and operations of the system, method, and computer program product according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram can represent a module, program segment, or part of code, and the module, program segment, or part of code contains one or more for realizing the specified logic function Executable instructions. It should also be noted that, in some alternative implementations, the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two blocks shown in succession can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations Or it can be realized by a combination of dedicated hardware and computer instructions.

The units involved in the embodiments described in the present disclosure may be implemented in a software manner, and may also be implemented in a hardware manner. The described unit may also be provided in the processor, for example, it may be described as: a processor includes a processor, including an acquiring unit, a determining unit, and a pushing unit. Wherein, the names of these units do not constitute a limitation on the unit itself under certain circumstances. For example, the obtaining unit can also be described as "a unit for obtaining a social network including a target user".

The above description is only a preferred embodiment of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover the above-mentioned inventive concept without departing from the above-mentioned inventive concept. Other technical solutions formed by any combination of technical features or equivalent features. For example, the above-mentioned features and the technical features disclosed in the embodiments of the present disclosure (but not limited to) having similar functions are replaced with each other to form a technical solution.

Claims

An information push method, including:

Acquiring a social network including target users, the social network including behavior information of users located in the social network and indication information used to indicate whether an association relationship is established between users;

Determine the probability of establishing an association relationship between the target user and a user who has not established an association relationship based on the behavior information of the target user and the behavior information of users in the social network who have not established an association relationship with the target user;

Push the user identification of the user corresponding to the probability exceeding the preset threshold to the target user.
The method according to claim 1, wherein the determining whether the target user is different from the target user based on the behavior information of the target user and the behavior information of other users in the social network that has not established an association relationship with the target user The probability of establishing an association relationship between users who have established an association relationship includes:

The behavior information of the target user and the behavior information of users who have not established an association relationship with the target user are respectively input into a pre-trained feature value determination model to obtain the target user behavior information characteristics corresponding to the target user's behavior information Value, and other user behavior information characteristic values corresponding to the behavior information of users who have not established an association relationship with the target user;

Based on the target user behavior information characteristic value and other user behavior information characteristic values, the probability of establishing an association relationship between the target user and a user who has not established an association relationship is determined.
The method according to claim 2, wherein the characteristic value determination model is obtained through training in the following steps:

Acquiring a training sample set, where each training sample in the training sample set includes behavior information of a specified user, behavior information of other users, and indication information for indicating whether other users have established an association relationship with the specified user;

Perform the following training steps: for the training samples in the training sample set, input the behavior information of the specified user and the behavior information of other users in the training sample into the neural network to be trained, and obtain the behavior information characteristic value corresponding to the specified user , Behavior information feature values corresponding to other users; based on the behavior information feature values corresponding to the specified user and the behavior information feature values corresponding to other users, determine the probability value of the specified user establishing an association relationship with other users; based on the determined training sample set The probability value corresponding to the training sample to determine whether the preset loss function converges;

In response to determining that the preset loss function converges, it is determined that the eigenvalue determines that the model training is completed.
The method according to claim 3, wherein the method further comprises:

In response to determining that the preset loss function has not converged, the parameters of the neural network to be trained are updated using the back propagation algorithm, and the training step is continued.
The method according to claim 3, wherein the behavior information of other users in the training sample set includes one of the following: behavior information of a positive sample user that has established an association relationship with a specified user, and negative behavior information that has not established an association relationship with a specified user. The behavior information of the sample user, where the negative sample user is a user who is randomly selected from the preset user set and has not established an association relationship with the specified user and has established an association relationship with other users.
The method according to claim 5, wherein the training samples in the training sample set further include time stamps corresponding to the behavior information of other users; wherein,

The timestamp corresponding to the behavior information of the positive sample user is determined based on the time when the designated user establishes an association relationship with the positive sample user;

The time stamp corresponding to the behavior information of the negative sample user is randomly selected from the time stamp corresponding to the positive sample user.
The method according to claim 6, wherein the neural network comprises n-layer cascaded single-layer feedforward neural network and a fully connected layer, for the i-th layer of feedforward neural network, the output of the previous layer of feedforward neural network The parameters are used as the input parameters of the feedforward neural network of the latter layer, where n is a positive integer greater than 1, and i is a positive integer greater than 2 and less than n; and

Each layer of feedforward neural network performs feature extraction through the following steps:

Based on the first preset activation function, the behavior information feature of the specified user, the behavior information feature of other users, and the time stamp corresponding to the behavior information of other users, determine the correlation index between the behavior information of the specified user and the behavior information of other users ；

Based on the determined correlation index and the second preset activation function, determine the parameters used to indicate the behavior information characteristics of the specified user;

The obtained parameters used to indicate the characteristics of the behavior information of the specified user are used as the output of the feedforward neural network of the current layer.
The method according to any one of claims 1-7, wherein the behavior information includes at least one of the following: browsing information and publishing information, and the browsing information is generated based on browsing information published by other users who have established an association relationship with the user.
An information push device includes:

An obtaining unit configured to obtain a social network including a target user, the social network including behavior information of users located in the social network and indication information for indicating whether an association relationship is established between the users;

The determining unit is configured to determine the relationship between the target user and the user who has not established an association relationship based on the behavior information of the target user and the behavior information of users in the social network that has not established an association relationship with the target user, The probability of establishing an association relationship;

The pushing unit is configured to push the user identification of the user corresponding to the probability exceeding the preset threshold to the target user.
A terminal device, including:

One or more processors;

A storage device on which one or more programs are stored;

When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1-8.
A computer-readable medium with a computer program stored thereon, wherein the program is executed by a processor to implement the method according to any one of claims 1-8.