CN115577171A

CN115577171A - Information pushing method and device, electronic equipment and storage medium

Info

Publication number: CN115577171A
Application number: CN202211177896.1A
Authority: CN
Inventors: 潘迪生
Original assignee: Beijing IQIYI Science and Technology Co Ltd
Current assignee: Beijing IQIYI Science and Technology Co Ltd
Priority date: 2022-09-23
Filing date: 2022-09-23
Publication date: 2023-01-06

Abstract

The embodiment of the disclosure relates to an information pushing method, an information pushing device, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring first operation information of a first operation executed by a first user through a target application, wherein the execution frequency of the first user for the first operation is less than or equal to a preset frequency threshold; determining target push information used for pushing to a user terminal of a first user from a predetermined set of information to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is greater than the preset frequency threshold; and pushing the target pushing information to the user terminal. By the method, sparsity of user behaviors is relieved, and accuracy of pushing information to the users with sparse behaviors is improved.

Description

Information pushing method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to an information pushing method and apparatus, an electronic device, and a storage medium.

Background

Information push is a technology for reducing information overload by transmitting information required by a user on the internet through a certain technical standard or protocol and a certain method.

In the prior art, different information (such as videos, news, images, products and the like) is pushed to a user generally depending on user behaviors (such as searching, watching, commenting, popping up a bullet screen and the like). However, for a user with sparse behavior, the accuracy of information pushing for the user tends to be low.

Disclosure of Invention

In view of this, in order to solve some or all of the above technical problems, embodiments of the present disclosure provide an information pushing method and apparatus, an electronic device, and a storage medium.

In a first aspect, an embodiment of the present disclosure provides an information pushing method, where the method includes:

acquiring first operation information of a first operation executed by a first user through a target application, wherein the execution frequency of the first user for the first operation is less than or equal to a preset frequency threshold;

determining target push information used for pushing to a user terminal of a first user from a predetermined set of information to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is greater than the preset frequency threshold;

and pushing the target pushing information to the user terminal.

In one possible implementation, the determining, from a predetermined set of information to be pushed, target push information for pushing to a user terminal of a first user based on the first operation information and second operation information of a second operation performed by a second user through the target application includes:

generating a first operation characteristic of the first user based on the first operation information and second operation information of a second operation executed by a second user through the target application;

and determining target push information for pushing to the user terminal of the first user from a predetermined set of information to be pushed based on the first operation characteristic.

In one possible implementation, the generating a first operation characteristic of the first user based on the first operation information and second operation information of a second operation performed by a second user through the target application includes:

and inputting the first operation information into a pre-trained first model to obtain a first operation characteristic of the first user, wherein the first model represents a corresponding relation between the first operation information and the first operation characteristic, and the first model is obtained by training based on second operation information of a second operation executed by a second user through the target application.

In one possible embodiment, the first model is trained as follows:

inputting the second operation information into a pre-trained second model to obtain a second operation characteristic of the second user, wherein the second model represents a corresponding relation between the second operation information and the second operation characteristic;

adjusting model parameters of a first initial model of the first model based on the first operational information and the second operational characteristics to train the first initial model;

and determining the first initial model after training as the first model.

In one possible embodiment, the second model is trained as follows:

acquiring a training sample set, wherein training samples in the training sample set comprise second operation information and corresponding user category information, and the user category information represents a user category of an executing user of an operation indicated by the second operation information;

training a second initial model of the second model by adopting a machine learning algorithm based on the second operation information and the user category information;

and determining the second initial model after training as the second model.

In one possible embodiment, the training, by using a machine learning algorithm, a second initial model of the second model based on the second operation information and the user category information includes:

acquiring a second initial model of the second model;

inputting the second operation information into the second initial model to obtain first output data of the second initial model;

inputting the first output data into a full connection layer to obtain actual output data of the full connection layer;

and adjusting model parameters of the second initial model based on the user category information and the actual output data by adopting a machine learning algorithm so as to train the second initial model.

In one possible embodiment, the adjusting model parameters of a first initial model of the first model based on the first operational information and the second operational characteristics to train the first initial model includes:

obtaining a first initial model of the first model;

inputting the first operation information into the first initial model to obtain second output data of the first initial model;

adjusting model parameters of a first initial model of the first model to train the first initial model, using:

when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to the same user category, adjusting the model parameters of the first initial model based on the second operation characteristic and the second output data by using a first loss function, wherein a function value of the first loss function is positively correlated with a target similarity, and the target similarity is a similarity between the second output data and the second operation characteristic;

and when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to different user categories, adjusting the model parameters of the first initial model based on the second operation characteristics and the second output data by adopting a second loss function, wherein the function value of the second loss function is in negative correlation with a target similarity, and the target similarity is the similarity between the second output data and the second operation characteristics.

In one possible embodiment of the method according to the invention,

the first loss function is: l = d (H) _ua ，H _ub )；

The second loss function is: l = max (0,m-d (H) _ua ，H _ub ))；

Wherein H _ua Characterizing said second output data, H _ub Characterizing the second operating characteristic, d (H) _ua ，H _ub ) And characterizing the target similarity, wherein m is a preset threshold value.

In one possible embodiment, the initial model parameters of the second initial model are the same as the initial model parameters of the first initial model, and the model structure of the second initial model is the same as the model structure of the first initial model.

In a second aspect, an embodiment of the present disclosure provides an information pushing apparatus, where the apparatus includes:

the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring first operation information of a first operation executed by a first user through a target application, and the execution frequency of the first user aiming at the first operation is less than or equal to a preset frequency threshold;

a determining unit, configured to determine, based on the first operation information and second operation information of a second operation performed by a second user through the target application, target push information to be pushed to a user terminal of the first user from a predetermined set of information to be pushed, where an execution frequency of the second user for the second operation is greater than the preset frequency threshold;

and the pushing unit is used for pushing the target pushing information to the user terminal.

In one possible implementation, the determining, from a predetermined set of information to be pushed, target push information for pushing to a user terminal of the first user based on the first operation information and second operation information of a second operation performed by a second user through the target application includes:

generating a first operation characteristic of the first user based on the first operation information and second operation information of a second operation performed by a second user through the target application;

In one possible embodiment, the generating a first operation feature of the first user based on the first operation information and second operation information of a second operation performed by a second user through the target application includes:

and inputting the first operation information into a pre-trained first model to obtain a first operation characteristic of the first user, wherein the first model represents the corresponding relation between the first operation information and the first operation characteristic, and the first model is obtained by training on the basis of second operation information of a second operation executed by a second user through the target application.

In one possible embodiment, the first model is trained as follows:

and determining the first initial model after training as the first model.

In one possible embodiment, the second model is trained as follows:

acquiring a training sample set, wherein training samples in the training sample set comprise second operation information and corresponding user category information, and the user category information represents a user category of an execution user of an operation indicated by the second operation information;

and determining the trained second initial model as the second model.

acquiring a second initial model of the second model;

In one possible embodiment, the adjusting model parameters of a first initial model of the first model based on the first operation information and the second operation characteristics to train the first initial model includes:

obtaining a first initial model of the first model;

when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to the same user category, adjusting the model parameters of the first initial model based on the second operation characteristic and the second output data by adopting a first loss function, wherein a function value of the first loss function is positively correlated with a target similarity, and the target similarity is the similarity between the second output data and the second operation characteristic;

and if the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to different user categories, adopting a second loss function, and adjusting the model parameters of the first initial model based on the second operation characteristics and the second output data, wherein the function value of the second loss function is in negative correlation with a target similarity, and the target similarity is the similarity between the second output data and the second operation characteristics.

In one possible embodiment of the method according to the invention,

the first loss function is: l = d (H) _ua ，H _ub )；

The second loss function is: l = max (0,m-d (H) _ua ，H _ub ))；

In a third aspect, an embodiment of the present disclosure provides an electronic device, including:

a memory for storing a computer program;

a processor, configured to execute the computer program stored in the memory, and when the computer program is executed, implement the method of any embodiment of the information pushing method of the first aspect of the present disclosure.

In a fourth aspect, the disclosed embodiments provide a computer-readable storage medium, and when being executed by a processor, the computer program implements the method of any embodiment of the information pushing method of the first aspect.

In a fifth aspect, the disclosed embodiments provide a computer program, which includes computer readable code, when the computer readable code is run on a device, causes a processor in the device to execute instructions for implementing the steps in the method according to any one of the embodiments of the information pushing method in the first aspect.

The information pushing method provided by the embodiment of the present disclosure includes acquiring first operation information of a first operation executed by a first user through a target application, where an execution frequency of the first operation by the first user is less than or equal to a preset frequency threshold, then determining target pushing information for pushing to a user terminal of the first user from a predetermined set of information to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, where the execution frequency of the second operation by the second user is greater than the preset frequency threshold, and then pushing the target pushing information to the user terminal. According to the method, the pushing information used for pushing to the user with sparse behavior can be determined based on the operation information of the user with dense behavior, so that the sparsity of the user behavior is relieved, and the accuracy of pushing the information to the user with sparse behavior is improved.

Drawings

Fig. 1 is a schematic flowchart of an information pushing method according to an embodiment of the present disclosure;

fig. 2 is a schematic flow chart of another information pushing method provided in the embodiment of the present disclosure;

fig. 3A is a schematic flowchart of another information push method according to an embodiment of the present disclosure;

fig. 3B is a schematic structural diagram of a second model involved in an information pushing method according to an embodiment of the present disclosure;

fig. 3C is a schematic training diagram of a first model involved in an information pushing method according to an embodiment of the present disclosure;

fig. 3D is a schematic flowchart of another information pushing method according to an embodiment of the disclosure;

fig. 4 is a schematic structural diagram of an information pushing apparatus according to an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.

Detailed Description

Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that: the relative arrangement of parts and steps, numerical expressions, and values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

It will be understood by those within the art that the terms "first," "second," and the like in the embodiments of the present disclosure are used only for distinguishing between different steps, devices, or modules, and do not denote any particular technical meaning or order therebetween.

It is also understood that in the present embodiment, "a plurality" may mean two or more, and "at least one" may mean one, two or more.

It is also to be understood that any reference to any component, data, or structure in the embodiments of the disclosure, may be generally understood as one or more, unless explicitly defined otherwise or stated otherwise.

In addition, the term "and/or" in the present disclosure is only one kind of association relationship describing an associated object, and means that three kinds of relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in the present disclosure generally indicates that the former and latter associated objects are in an "or" relationship.

It should also be understood that the description of the embodiments in the present disclosure emphasizes the differences between the embodiments, and the same or similar parts may be referred to each other, and are not repeated for brevity.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.

Techniques, methods and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

It should be noted that, in the present disclosure, the embodiments and the features of the embodiments may be combined with each other without conflict. For the purpose of facilitating an understanding of the embodiments of the present disclosure, the present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with the embodiments. It is to be understood that the described embodiments are only a few, and not all, of the disclosed embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.

Further, it should be noted that users (including the first user, the second user, the user performing the operation indicated by the first operation information, the user performing the operation indicated by the second operation information, and the like) described in the present disclosure may be distinguished by the user identification. For example, the user identifier may be a login account, and in this scenario, if different people use the same account to log in, the different people may be considered as the same user; if the same person respectively adopts different accounts to log in, the same person who logs in different accounts can be considered as different users.

Fig. 1 is a schematic flowchart of an information push method according to an embodiment of the present disclosure. As shown in fig. 1, the method specifically includes:

101. the method includes the steps that first operation information of a first operation executed by a first user through a target application is obtained, wherein the execution frequency of the first user aiming at the first operation is smaller than or equal to a preset frequency threshold.

In this embodiment, the first user may be a user whose frequency of the operation (i.e., the execution frequency) executed by the target application (i.e., the first operation) is less than or equal to a preset frequency threshold.

The target application may be any application. By way of example, the target application may be shopping-like software, a shopping-like website, video playback software, an audio playback website, and the like.

The first operation may be various operations performed by the first user through the target application. As an example, the first operation may include: search, purchase, collect, click, browse, view, send barrage, etc.

The first operation information may be various information of the first operation. As an example, the first operation information may include: purchase information, collection information, click information, browsing information, viewing information, transmission barrage information, operation object information, operation time information, and the like.

The preset frequency threshold may be a predetermined frequency. As an example, the preset frequency threshold may be once every three days, 50 times every 90 days, etc.

Here, if the first user used the search function 3 times and watched 20 videos within the past 90 days, the execution frequency of the first operation may be: every 90 days for 23 times.

102. And determining target push information used for pushing to a user terminal of the first user from a predetermined information set to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is greater than the preset frequency threshold.

In this embodiment, the second user may be a user who performs an operation (i.e., the second operation) through the target application, where the frequency (i.e., the execution frequency) of the operation is greater than the preset frequency threshold.

The second operation may be various operations performed by the second user through the target application. As an example, the second operation may include: search, purchase, collect, click, browse, view, send barrage, etc.

The second operation information may be various information of the second operation. As an example, the second operation information may include: purchase information, collection information, click information, browsing information, viewing information, transmission barrage information, operation object information, operation time information, and the like.

Here, the second operation is different from the first operation in that the user is performed differently. The executing user of the second operation is the second user, and the executing user of the first operation is the first user.

The information set to be pushed may include one or more information to be pushed. The information to be pushed may include video, news, images, product information, and the like. The information set to be pushed may be a set of all or part of information to be pushed available for pushing in the target application.

The user terminal of the first user may be a user terminal used by the first user. For example, the user terminal of the first user may be a user terminal logged in with an account of the first user.

The target push information may be one or more pieces of push information in the set of information to be pushed for pushing to the user terminal.

Illustratively, for each piece of information to be pushed in the information set to be pushed, the similarity between the piece of information to be pushed and the second operation information may be calculated, and a preset number of pieces of information to be pushed are selected according to the sequence of similarity from high to low, so as to obtain a subset of the pieces of information to be pushed. Then, for each piece of information to be pushed in the information to be pushed subset, the similarity between the piece of information to be pushed and the first operation information is calculated, and the piece of information to be pushed with the highest similarity to the first operation information is determined as the target pushing information used for pushing to the user terminal of the first user.

103. And pushing the target pushing information to the user terminal.

Optionally, after the user terminal receives the target push information, the user terminal may perform semantic analysis on the target push information to obtain a semantic analysis result, and then perform a corresponding operation based on the semantic analysis result. For example, if the semantic analysis result indicates that "xx variety programs are updated", the user terminal may automatically perform a downloading operation (i.e., the corresponding operation described above), or the user terminal may automatically display prompt information for prompting the user to download the variety programs, and in the case that a confirmation operation of the user for the prompt information is detected, the variety programs are downloaded.

According to the information pushing method provided by the embodiment of the disclosure, first operation information of a first operation executed by a first user through a target application is acquired, wherein an execution frequency of the first operation by the first user is smaller than or equal to a preset frequency threshold, then, target pushing information used for pushing to a user terminal of the first user is determined from a predetermined information set to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is larger than the preset frequency threshold, and then, the target pushing information is pushed to the user terminal. According to the method, the pushing information used for pushing to the user with sparse behavior can be determined based on the operation information of the user with dense behavior, so that the sparsity of the user behavior is relieved, and the accuracy of pushing the information to the user with sparse behavior is improved.

Fig. 2 is a schematic flow chart of another information pushing method according to an embodiment of the present disclosure. As shown in fig. 2, the method specifically includes:

201. the method includes the steps that first operation information of a first operation executed by a first user through a target application is obtained, wherein the execution frequency of the first user aiming at the first operation is smaller than or equal to a preset frequency threshold.

In this embodiment, step 201 is substantially the same as step 101 in the embodiment corresponding to fig. 1, and is not described here again.

202. Generating a first operation characteristic of the first user based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is greater than the preset frequency threshold.

In this embodiment, the first operation characteristic may characterize the first operation performed by the first user. In practice, the first operational characteristic may be characterized using a vector.

For example, the first operation characteristic of the first user may be generated according to a preset rule based on the first operation information and second operation information of a second operation performed by a second user through the target application.

203. And determining target push information for pushing to the user terminal of the first user from a predetermined set of information to be pushed based on the first operation characteristic.

In this embodiment, for each piece of information to be pushed in the set of information to be pushed, the similarity between the piece of information to be pushed and the first operation feature may be calculated, and the piece of information to be pushed with the highest similarity among the calculated similarities may be determined as the target pushing information.

204. And pushing the target pushing information to the user terminal.

In this embodiment, step 204 is substantially the same as step 103 in the embodiment corresponding to fig. 1, and is not described here again.

It should be noted that, in addition to the above-mentioned contents, the present embodiment may further include technical features described in the embodiment corresponding to fig. 1, so as to achieve the technical effect of the information push method shown in fig. 1, and for brevity, please refer to the description related to fig. 1, which is not repeated herein.

According to the information pushing method provided by the embodiment of the disclosure, the first operation characteristic of the first user is generated based on the first operation information and the second operation information, and the target pushing information is determined based on the first operation characteristic, so that the accuracy of information pushing can be further improved.

Fig. 3A is a schematic flow chart of another information pushing method according to an embodiment of the present disclosure. The method can be applied to one or more electronic devices such as smart phones, notebook computers, desktop computers, portable computers and servers. In addition, the execution main body of the method can be hardware or software. When the execution main body is hardware, the execution main body may be one or more of the electronic devices. For example, a single electronic device may perform the method, or multiple electronic devices may cooperate with each other to perform the method. When the execution subject is software, the method may be implemented as a plurality of software or software modules, or may be implemented as a single software or software module. And is not particularly limited herein.

Specifically, as shown in fig. 3A, the method specifically includes:

301. the method includes the steps that first operation information of a first operation executed by a first user through a target application is obtained, wherein the execution frequency of the first user aiming at the first operation is smaller than or equal to a preset frequency threshold.

In this embodiment, step 301 is substantially the same as step 101 in the embodiment corresponding to fig. 1, and is not described herein again.

302. Inputting the first operation information into a pre-trained first model to obtain a first operation characteristic of the first user, wherein the first model represents a corresponding relation between the first operation information and the first operation characteristic, the first model is obtained by training second operation information of a second operation executed by a second user through the target application, and the execution frequency of the second user for the second operation is greater than the preset frequency threshold.

In this embodiment, the first model may include a convolutional neural network or the like.

Illustratively, the first model may be obtained by training as follows:

first, a first initial model of a first model and a first training sample set are obtained. The first initial model may be an initial model of the first model. After training the first initial model and adjusting model parameters (including initial model parameters) of the first initial model, the first model can be obtained. The training samples in the first set of training samples may include input data and target output data. The input data may include sample first operational information and sample second operational information, and the target output data may include sample first operational characteristics.

And then, inputting the input data and the expected output data into the first initial model by adopting a machine learning algorithm so as to train the first initial model and adjust model parameters of the first initial model, thereby obtaining the first model.

303. And determining target push information for pushing to the user terminal of the first user from a predetermined set of information to be pushed based on the first operation characteristic.

In this embodiment, step 303 is substantially the same as step 203 in the corresponding embodiment of fig. 2, and is not described herein again.

304. And pushing the target pushing information to the user terminal.

In this embodiment, step 304 is substantially the same as step 103 in the embodiment corresponding to fig. 1, and is not described here again.

In some optional implementations of this embodiment, the first model is obtained by training in the following manner:

and step one, inputting the second operation information into a pre-trained second model to obtain a second operation characteristic of the second user. Wherein the second model characterizes a correspondence between the second operational information and the second operational characteristic.

And secondly, adjusting model parameters of a first initial model of the first model based on the first operation information and the second operation characteristics so as to train the first initial model.

Here, the model parameters of the first initial model of the first model may be adjusted by calculating a similarity between the first operation information and the second operation characteristic to train the first initial model.

And thirdly, determining the trained first initial model as the first model.

Here, it may be determined that the first initial model training is completed if at least one of the following conditions is satisfied: the similarity between the first operation information and the second operation characteristic is smaller than or equal to a preset similarity threshold, the training time exceeds the preset time, and the training times are larger than or equal to the preset times.

It is to be understood that, in the above alternative implementation, the second operation characteristic of the second user is obtained through the second model, and the first model is obtained based on the first operation information and the second operation characteristic. Therefore, the information pushing accuracy can be further improved by adopting the first model to push the information.

In some application scenarios in the above optional implementation manners, the second model is obtained by training in the following manner:

in a first step, a set of training samples is obtained. And training samples in the training sample set comprise second operation information and corresponding user category information. The user category information represents a user category of an executing user of the operation indicated by the second operation information.

And secondly, training a second initial model of the second model by adopting a machine learning algorithm based on the second operation information and the user category information.

And thirdly, determining the trained second initial model as the second model.

Here, it may be determined that the second initial model training is completed if at least one of the following conditions is satisfied: the similarity between the actual output data and the user category information is smaller than or equal to a preset similarity threshold, the training time exceeds the preset time, and the training times are larger than or equal to the preset times.

The first output data may be obtained as follows: first, second operation information is input into a second initial model, and first output data is obtained. And then, inputting the first output data to a full connection layer to obtain the actual output data of the full connection layer.

It can be understood that, in the application scenario, the second model may be trained and obtained based on the second operation information and the corresponding user category information, and thus, the accuracy of information pushing may be further improved.

In some cases in the application scenarios described above, the following may be employed. To perform the second step:

first, a second initial model of the second model is obtained.

The second initial model may be an initial model of the second model. After training the second initial model and adjusting the model parameters (including the initial model parameters) of the second initial model, the second model can be obtained.

The initial model parameters of the second initial model may be the same or different from the initial model parameters of the first initial model. The model structure of the second initial model may be the same as or different from the model structure of the first initial model.

And then, inputting the second operation information into the second initial model to obtain first output data of the second initial model.

The first output data may be data output by the second initial model after the second operation information is input into the second initial model.

And then, inputting the first output data into a full connection layer to obtain the actual output data of the full connection layer.

The actual output data may be data output by the full link layer after the first output data is input to the full link layer.

And finally, adjusting model parameters of the second initial model by adopting a machine learning algorithm based on the user category information and the actual output data so as to train the second initial model.

Specifically, the function value of the preset loss function may be calculated based on the user category information and the actual output data. And then, using a gradient descent method and a back propagation method to adjust the model parameters of the second initial model by using the function values so as to train the second initial model.

It is to be understood that, in the above case, the second model may be trained based on the second operation information and the user category information, and thus, the accuracy of information push may be further improved.

In some application scenarios in the foregoing optional implementation manners, the following manner may be adopted to adjust model parameters of a first initial model of the first model based on the first operation information and the second operation characteristic, so as to train the first initial model:

first, a first initial model of the first model is obtained.

The first initial model may be an initial model of the first model. After training the first initial model and adjusting model parameters (including initial model parameters) of the first initial model, the first model can be obtained.

And then, inputting the first operation information into the first initial model to obtain second output data of the first initial model.

The second output data may be data output by the first initial model after the first operation information is input to the first initial model.

Then, the following steps (including the first step and the second step) are adopted to adjust model parameters of a first initial model of the first model so as to train the first initial model:

first, when the user performing the operation indicated by the first operation information and the user performing the operation indicated by the second operation information belong to the same user category, a first loss function is used to adjust the model parameters of the first initial model based on the second operation characteristics and the second output data. Wherein a function value of the first loss function is positively correlated with a target similarity, the target similarity being a similarity between the second output data and the second operating characteristic.

And a second step of adjusting the model parameters of the first initial model based on the second operation characteristics and the second output data by using a second loss function when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to different user categories. Wherein the function value of the second loss function is inversely related to a target similarity, the target similarity being a similarity between the second output data and the second operating characteristic.

It can be understood that, in the above application scenario, when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to the same user class, the first loss function having a positive correlation between the function value and the target similarity is used to adjust the model parameters of the first initial model, so that the first model can learn the commonality between the operation features of the users of the same user class; when the executing user of the operation indicated by the first operation information and the executing user of the operation indicated by the second operation information belong to different user classes, the second loss function with the function value being in negative correlation with the target similarity is adopted to adjust the model parameters of the first initial model, so that the first model can learn the difference between the operation characteristics of the users of different user classes, and the accuracy of information push is further improved.

In some cases in the above application scenarios, the first loss function is: l = d (H) _ua ，H _ub ). The second loss function is: l = max (0,m-d (H) _ua ，H _ub ))。

Wherein H _ua Characterizing said second output data, H _ub Characterizing said second operating characteristic, d (H) _ua ，H _ub ) And characterizing the target similarity, wherein m is a preset threshold value.

It can be understood that, in the above case, the accuracy of information push can be further improved by training the first model with the first loss function and the second loss function as above.

In some examples of the above case, the initial model parameters of the second initial model are the same as the initial model parameters of the first initial model, and the model structure of the second initial model is the same as the model structure of the first initial model.

It can be understood that, in the above example, the same initial model may be adopted, and the first model and the second model are trained respectively, so that the first operation characteristic of the first user may be obtained more accurately by using the first model, thereby further improving the accuracy of information pushing.

The following description is made for the purpose of illustrating the embodiments of the present disclosure, but it should be noted that the embodiments of the present disclosure may have the features described below, but the following description should not be construed as limiting the scope of the embodiments of the present disclosure.

In this example, it is considered that the user whose N is less than or equal to 50 times a day is a user whose behavior is sparse (denoted by UB here, i.e., the first user), and the other user (i.e., the user whose N is greater than 50 times a day) is a user whose behavior is dense (denoted by UA here, i.e., the second user). The behavior frequency N represents the number of behaviors (i.e., the first operation or the second operation) for 90 days. The behaviors refer to a set of behaviors of searching, watching and the like of the user. For example, in the past 90 days, the user has used the search function 3 times at the target application, and has seen 20 videos, then this N is 23 for the user.

Here, a frequency threshold is preset, i.e. 50 times every 90 days.

Specifically, referring to fig. 3B, fig. 3B is a schematic structural diagram of a second model involved in an information pushing method provided in the embodiment of the present disclosure. The population (i.e., the user) can be first classified into T categories according to the business requirements, so as to obtain T user categories. Sending the UA behaviors (i.e., the second operation information) of each category into a model (Modela, i.e., a second initial model of the second model) for training, wherein the training target is the crowd category (i.e., the user category information) to which the user belongs, and the training target is a T classification model (i.e., the second model). Assume that the characteristic of each UA user behavior (i.e., the first output data) is F _ua ∈R ^s*d S is the number of user behaviors, d is the feature dimension of each behavior, the feature is sent to DNN (Deep Neural Networks) of a full connection layer to train a T classification model, and y' = Softmax (F = Relu) _ua W1)O)∈R ^T Y' is the prediction result of the model, W1. Epsilon. R ^d*wd Is a trainable linearity parameter, wd is the number of layers of the hidden layer of the linearity layer, and O e R ^wd*T Is an output layer, and the loss function is a cross entropy function

y' is the true class of the predicted sample.

Next, referring to fig. 3C, fig. 3C is a schematic training diagram of a first model involved in an information pushing method according to an embodiment of the present disclosure. Initializing the same model (ModelB, i.e. the first initial model, the first initial model and the second initial model do not share parameters), the input of the model is UB behavior, and the characteristic of UB user behavior (i.e. the first operation characteristic) is F _ub ∈R ^s*d Since s is smaller than N, therefore, for feature vector F _ub Its characteristic element after N is 0. And combining the Model A and the Model B into a double-tower model for joint training, wherein the input of the Model A is a random UA behavior (namely the first operation information), and the input of the Model B is a random UB behavior (namely the second operation information). Then respectively outputting the characteristic H of the DNN hidden layer _ua ＝relu(F _ua W1)∈R ^d*wd And H _ub ＝relu(F _ub W2)∈R ^d*wd ，(W2∈R ^d*wd Is a ModelB trainable linearity parameter), training is performed with a loss function of pairwiseloses: if the selected UA and UB populations are in the same population classification, the loss function (i.e., the first loss function) is: l = d (H) _ua ，H _ub ) (ii) a If the selected UA and UB populations are in different population classifications, the loss function (i.e., the second loss function) is: l = max (0,m-d (H) _ua ，H _ub ))。

Where d is a distance metric function, is the cosine distance, d (H) _ua ，H _ub ) And characterizing the similarity of the targets. Where m is set as a threshold, here taking m =0.6.

Then, information push is performed with reference to fig. 3D, and fig. 3D is a schematic flow chart of another information push method provided in the embodiment of the present disclosure. Specifically, if the user is UA, the operation information (i.e., the second operation information) F is input into the model a model to obtain the crowd characteristic (i.e., the second operation characteristic) H, otherwise, the operation information (i.e., the first operation characteristic) F is input into the model b model to obtain the crowd characteristic (i.e., the first operation characteristic) H of the sparse behavior. The crowd characteristic model in the figure can be a model or a model b model. And then combining the characteristics H and the characteristics of the information to be pushed in the information set to be pushed (namely the characteristics of the query statement in the diagram) and putting the characteristics and the characteristics into a ranking model for prediction to obtain a Score.

Firstly, obtaining the vector characteristic v of the query statement searched by the user currently _q ∈R ^d 。

Secondly, obtaining the vector characteristics u e R of the user ^ud 。

Thirdly, obtaining a characteristic vector Q = [ v ] of the information to be pushed in the information set to be pushed _q1 ,…,v _qi ,…,v _qn ]，1≤i≤n。

Fourthly, multiplying the vector characteristics of the user by a characteristic vector Q of the information to be pushed in the push information set to obtain an attention matrix A = uWQ ^T Wherein W ∈ R ^ud×d ，A∈R ^1×n And T represents transposition.

And fifthly, calculating scores of all the information to be pushed through the attention matrix A.

And sixthly, returning m pieces of information to be pushed with the highest scores as recall results (namely target pushing information).

Therefore, in the above example, by combining the semantics of the information to be pushed, the semantic information of the information to be pushed can be fully mined, the video pushing waiting information which the sparse user potentially wants to watch is found, the recall effect is improved, and the search recommendation effect can be further improved.

It should be noted that, in addition to the above-mentioned contents, the present embodiment may further include technical features described in the embodiment corresponding to fig. 1 and/or fig. 2, so as to achieve the technical effect of the information pushing method shown in fig. 1 and/or fig. 2, please refer to the description related to fig. 1 and/or fig. 2 for brevity, which is not described herein again.

According to the information pushing method provided by the embodiment of the disclosure, the target pushing information is determined through the first model obtained by training based on the second operation information, so that the accuracy of information pushing can be further improved.

Fig. 4 is a schematic structural diagram of an information pushing apparatus according to an embodiment of the present disclosure. The method specifically comprises the following steps:

an obtaining unit 401, configured to obtain first operation information of a first operation performed by a first user through a target application, where an execution frequency of the first operation by the first user is less than or equal to a preset frequency threshold;

a determining unit 402, configured to determine, based on the first operation information and second operation information of a second operation performed by a second user through the target application, target push information to be pushed to a user terminal of the first user from a predetermined set of information to be pushed, where an execution frequency of the second user for the second operation is greater than the preset frequency threshold;

a pushing unit 403, configured to push the target push information to the user terminal.

In one possible embodiment, the first model is trained as follows:

and determining the first initial model after training as the first model.

In one possible embodiment, the second model is trained as follows:

and determining the second initial model after training as the second model.

acquiring a second initial model of the second model;

obtaining a first initial model of the first model;

In one possible embodiment of the method according to the invention,

the first loss function is: l = d (H) _ua ，H _ub )；

The second loss function is: l = max (0,m-d (H) _ua ，H _ub ))；

The information push apparatus provided in this embodiment may be the information push apparatus shown in fig. 4, and may perform all the steps of the information push method shown in fig. 1-3A, so as to achieve the technical effect of the information push method shown in fig. 1-3A, and please refer to the description related to fig. 1-3A for brevity, which is not described herein again.

Fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure, where the electronic device 500 shown in fig. 5 includes: at least one processor 501, memory 502, at least one network interface 504, and other user interfaces 503. The various components in the electronic device 500 are coupled together by a bus system 505. It is understood that the bus system 505 is used to enable connection communications between these components. The bus system 505 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 505 in FIG. 5.

The user interface 503 may include, among other things, a display, a keyboard, or a pointing device (e.g., a mouse, trackball, touch pad, or touch screen, among others.

It will be appreciated that the memory 502 in embodiments of the present disclosure can be either volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory. The non-volatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable PROM (EEPROM), or a flash Memory. Volatile Memory can be Random Access Memory (RAM), which acts as external cache Memory. By way of illustration and not limitation, many forms of RAM are available, such as Static random access memory (Static RAM, SRAM), dynamic Random Access Memory (DRAM), synchronous Dynamic random access memory (Synchronous DRAM, SDRAM), double Data Rate Synchronous Dynamic random access memory (ddr Data Rate SDRAM, ddr SDRAM), enhanced Synchronous SDRAM (ESDRAM), synchlronous SDRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 502 described herein is intended to comprise, without being limited to, these and any other suitable types of memory.

In some embodiments, memory 502 stores elements, executable units or data structures, or a subset thereof, or an expanded set thereof as follows: an operating system 5021 and application programs 5022.

The operating system 5021 includes various system programs, such as a framework layer, a core library layer, a driver layer, and the like, and is used for implementing various basic services and processing hardware-based tasks. The application 5022 includes various applications, such as a Media Player (Media Player), a Browser (Browser), and the like, for implementing various application services. A program implementing the method of an embodiment of the present disclosure may be included in the application program 5022.

In this embodiment, by calling a program or an instruction stored in the memory 502, specifically, a program or an instruction stored in the application 5022, the processor 501 is configured to execute the method steps provided by the method embodiments, for example, including:

determining target push information used for being pushed to a user terminal of a first user from a predetermined information set to be pushed based on the first operation information and second operation information of a second operation executed by a second user through the target application, wherein the execution frequency of the second user for the second operation is larger than the preset frequency threshold;

and pushing the target pushing information to the user terminal.

The method disclosed by the embodiment of the present disclosure may be applied to the processor 501, or may be implemented by the processor 501. The processor 501 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware or instructions in the form of software in the processor 501. The Processor 501 may be a general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, or discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present disclosure may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present disclosure may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software elements in the decoding processor. The software elements may be located in ram, flash, rom, prom, or eprom, registers, among other storage media that are well known in the art. The storage medium is located in the memory 502, and the processor 501 reads the information in the memory 502 and completes the steps of the method in combination with the hardware.

It is to be understood that the embodiments described herein may be implemented in hardware, software, firmware, middleware, microcode, or any combination thereof. For a hardware implementation, the Processing units may be implemented in one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), digital Signal Processing Devices (DSPDs), programmable Logic Devices (PLDs), field Programmable Gate Arrays (FPGAs), general purpose processors, controllers, micro-controllers, microprocessors, other electronic units configured to perform the above-described functions of the present disclosure, or a combination thereof.

For a software implementation, the techniques described herein may be implemented by means of units performing the functions described herein. The software codes may be stored in a memory and executed by a processor. The memory may be implemented within the processor or external to the processor.

The electronic device provided in this embodiment may be the electronic device shown in fig. 5, and may perform all the steps of the information push method shown in fig. 1-3A, so as to achieve the technical effect of the information push method shown in fig. 1-3A, and for brevity, it is described with reference to fig. 1-3A, which is not described herein again.

The disclosed embodiments also provide a storage medium (computer-readable storage medium). The storage medium herein stores one or more programs. Among others, the storage medium may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, a hard disk, or a solid state disk; the memory may also comprise a combination of the above kinds of memories.

When one or more programs in the storage medium are executable by one or more processors, the information push method executed on the electronic device side is realized.

The processor is configured to execute the information pushing program stored in the memory to implement the following steps of the information pushing method executed on the electronic device side:

and pushing the target pushing information to the user terminal.

Those of skill would further appreciate that the various illustrative components and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, a software module executed by a processor, or a combination of the two. A software module may reside in Random Access Memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

The above-mentioned embodiments, objects, technical solutions and advantages of the present disclosure are described in further detail, it should be understood that the above-mentioned embodiments are merely illustrative of the present disclosure and are not intended to limit the scope of the present disclosure, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.

Claims

1. An information pushing method, characterized in that the method comprises:

and pushing the target pushing information to the user terminal.

2. The method according to claim 1, wherein the determining target push information for pushing to the user terminal of the first user from a predetermined set of information to be pushed based on the first operation information and second operation information of a second operation performed by a second user through the target application comprises:

3. The method of claim 2, wherein generating the first operational characteristic of the first user based on the first operational information and second operational information of a second operation performed by a second user through the target application comprises:

4. The method of claim 3, wherein the first model is trained by:

and determining the first initial model after training as the first model.

5. The method of claim 4, wherein the second model is trained by:

and determining the trained second initial model as the second model.

6. The method of claim 5, wherein the training a second initial model of the second model based on the second operation information and the user category information using a machine learning algorithm comprises:

acquiring a second initial model of the second model;

7. The method of any of claims 4-6, wherein said adjusting model parameters of a first initial model of the first model to train the first initial model based on the first operational information and the second operational characteristics comprises:

acquiring a first initial model of the first model;

8. The method of claim 7,

the first loss function is: l = d (H) _ua ，H _ub )；

The second loss function is: l = max (0,m-d (H) _ua ，H _ub ))；

9. Method according to claim 7 or 8, characterized in that the initial model parameters of the second initial model are the same as the initial model parameters of the first initial model and in that the model structure of the second initial model is the same as the model structure of the first initial model.

10. An information pushing apparatus, characterized in that the apparatus comprises:

the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring first operation information of a first operation executed by a first user through a target application, and the execution frequency of the first operation by the first user is less than or equal to a preset frequency threshold;

11. An electronic device, comprising:

a memory for storing a computer program;

a processor for executing a computer program stored in the memory, and when executed, implementing the method of any of the preceding claims 1-9.

12. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method of any one of the preceding claims 1 to 9.