WO2019076173A1

WO2019076173A1 - Content pushing method and apparatus, and computer device

Info

Publication number: WO2019076173A1
Application number: PCT/CN2018/105810
Authority: WO
Inventors: 潘岸腾
Original assignee: 广州优视网络科技有限公司
Priority date: 2017-10-20
Filing date: 2018-09-14
Publication date: 2019-04-25
Also published as: CN107613022B; CN107613022A

Abstract

The present invention provides a content pushing method and apparatus, and a computer device. The content pushing method comprises: obtaining characteristics of a user exposing an application and a click behavior of the application; generating a sample according to the characteristics of the user and the click behavior of the application; inputting the sample into a grouping model to determine a group to which the sample belongs in the grouping model; inputting the sample into a click rate prediction model corresponding to the group to obtain the predicted click rate of the application by a user; and pushing content to the user according to the predicted click rate of the application. According to the content pushing method, after a group to which a user belongs is determined, a predicted click rate is obtained according to the click rate prediction model of the group, the accuracy of the predicted click rate is improved, and content is accurately pushed to the user.

Description

Content pushing method, device and computer device

Technical field

The present invention relates to the field of Internet technologies, and in particular, to a content pushing method, apparatus, and computer device.

Background technique

LR (Logistic Regression) model is one of the discrete selection method models, and the logistic regression model is the earliest discrete selection model, and it is also the most widely used model. The logistic regression model is also a classification model in machine learning. Due to its simplicity and efficiency, it is widely used in practice.

At present, in the application store, the most frequently used model for app (APP) click rate estimation is the LR model. The idea is to cross the user feature and the application feature as the input feature of the final model, and train the LR model. A big drawback of this approach is that it is a model for all users. In many cases, the optimal LR model of the A user group is X. The optimal LR model of the B user group is Y. X and Y are likely to be conflicts. of. That is, the X model is a positive feature on a feature and the Y model is a negative feature. However, if there is only one LR model in order to simultaneously satisfy the A user group and the B user group to generate a compromise model Z, the model Z is not the optimal model. Therefore, the traditional LR model is not accurate for the user's click rate estimation of the application, which makes it impossible to accurately push the relevant content to the user according to the traditional LR model.

Summary of the invention

The object of the present invention is to provide a content pushing method, device and computer device, which can determine an estimated click rate according to a click rate prediction model of the group after determining the group to which the user belongs, and improve the accuracy of the estimated click rate, and accurately Users push relevant content.

In order to achieve the above object, the present invention provides the following technical solutions:

A content pushing method includes the steps of: acquiring a feature of a user of an exposure application and a click behavior on the application; generating a sample according to the feature of the user and a click behavior of the application; inputting the sample into the group In the model, determining a group to which the sample belongs in the group model; inputting the sample into a click rate estimation model corresponding to the group, and obtaining an estimated click rate of the user to the application; Push content to the user based on the estimated clickthrough rate of the app.

In one embodiment, the pushing content to the user based on an estimated click rate of the application comprises: pushing the application to the user based on an estimated click rate of the application.

In one embodiment, the pushing the application to the user according to an estimated click rate of the application comprises: confirming that the estimated click rate of the application is greater than a threshold, and pushing the application to the user .

In one embodiment, the pushing the application to the user according to an estimated click rate of the application, including calculating the estimated click rate of the user to all applications in the application mall, according to the estimated click The rate is sorted from high to low, and the user is pushed to the top of the preset number of applications.

In one embodiment, before acquiring the feature of the user of the exposure application and the click behavior on the application, the method further includes: acquiring a feature of the historical user of the exposure application and a click behavior on the application; according to the history The user's characteristics and the click behavior of the application generate samples; the sample is used to train a cluster model containing several groups and a population-specific click rate prediction model.

In one of the embodiments, the training the clustering model comprising a plurality of populations using the sample comprises: training the clustering model using the sample according to a decision tree algorithm.

In one embodiment, the training the clustering model comprising a plurality of groups and the population corresponding to the click rate prediction model using the sample comprises: training the clustering model comprising several groups using the sample, and grouping the samples into groups To classify, use the sample of the group corresponding classification to train the group's click rate prediction model.

In one embodiment, the using the population corresponding to the classified sample to train the population's click rate prediction model comprises: using the sample of the group corresponding classification, using the logistic algorithm to train the click rate prediction model.

In one embodiment, the click rate prediction model is a click rate estimation model obtained according to a formula of logistic regression; the formula of the logistic regression is:

Where X represents the input variable and β represents the weight vector of the different features.

In one embodiment, the using the population corresponding to the classified samples, using the logistic algorithm, training the click rate prediction model, including: training the group by using a logistic algorithm according to a gradient descent method, training The click rate prediction model in the corresponding group.

In one of the embodiments, the characteristics of the user include a natural attribute, a social attribute, or a preference attribute of the user.

In one embodiment, the generating a sample according to the feature of the user and the click behavior of the application includes: generating a sample by using the feature of the user as an input variable and using the click behavior as a target variable.

A content pushing device, comprising: a first obtaining module, configured to acquire a feature of a user exposing an application and a click behavior on the application; and a generating module, configured to: according to the feature of the user and the click behavior of the application, Generating a sample; a determining module, configured to input the sample into a grouping model, determining a group to which the sample belongs in the grouping model; and a second obtaining module, configured to input the sample to the group corresponding to the group In the click rate estimation model, the estimated click rate of the user to the application is obtained; and the pushing module is configured to push content to the user according to the estimated click rate of the application.

A computer device comprising: one or more processors; a memory; one or more applications, wherein the one or more applications are stored in the memory and configured to be by the one or more Executing by the processor, the one or more applications are configured to perform the content push method described in any of the above embodiments.

According to an embodiment of the present invention, a content pushing method is provided, which generates a sample according to a feature of a user and a click behavior of the application, and inputs the sample into the grouping model to confirm that the sample belongs to a group in the grouping model. Then, according to the click rate estimation model in the group, the estimated click rate of the user is obtained, and then the corresponding content is pushed to the user according to the estimated click rate of the application. After the content pushing method determines the group to which the user belongs, the estimated click rate is obtained according to the click rate estimation model of the group, the accuracy of the estimated click rate is improved, and the relevant content is accurately pushed to the user. For example, relevant content may include applications, similar applications, application introductions, and advertisements and information related to the applications. In addition, the content push method performs group model training according to different user groups, and trains a click rate prediction model for different groups in the group model, so that different user groups correspond to different click rate estimation models, according to corresponding clicks. The rate prediction model obtains a more accurate estimated click rate for the user to accurately push relevant content to the user.

DRAWINGS

The above and/or additional aspects and advantages of the present invention will become apparent and readily understood from

1 is a flowchart of a method for pushing a content according to an embodiment of the present invention;

2 is a flowchart of a method for pushing a content according to another embodiment of the present invention;

3 is a flow chart of a method for establishing a clustered logistic regression click rate prediction model method according to an embodiment of the present invention;

4 is a flowchart of a usage scenario of a model of a content pushing method according to an embodiment of the present invention;

FIG. 5 is a flowchart of a method for a content push method according to still another embodiment of the present invention; FIG.

6 is a flowchart of a method for a content push method according to still another embodiment of the present invention;

FIG. 7 is a schematic structural diagram of a content pushing apparatus according to an embodiment of the present invention; FIG.

FIG. 8 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed ways

The embodiments of the present invention are described in detail below, and the examples of the embodiments are illustrated in the drawings, wherein the same or similar reference numerals are used to refer to the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the drawings are intended to be illustrative of the invention and are not to be construed as limiting.

Those skilled in the art should understand that the content in the content pushing method of the present invention includes: an application, a similar application, an application introduction, an advertisement related to the application, information, and the like.

FIG. 1 is a flowchart of a method for a content push method according to an embodiment of the present invention. As shown in FIG. 1, a content pushing method of the present invention includes the steps of: S10-S50.

S10. Obtain the characteristics of the user of the exposure application and the click behavior of the application.

When the server detects the user exposing the application, it acquires the characteristics of the user who exposed the application and obtains the user's click behavior on the application. Here, the user's click behavior on the application includes the user clicking on the application and the user not clicking on the application. The application can be various applications in the app store, including game apps, social apps, reading apps, system apps, and shopping apps.

The characteristics of the user include the user's natural attributes, the user's social attributes, or the user's preferred attributes. The user's natural attributes such as the age of the user, the gender of the user, and so on. The social attributes of the user, such as the user's cultural level, the user's occupation, and the user's geographic area. User preferences such as "military fans", "technology enthusiasts", "football fans" and "game fans".

In one embodiment, the server obtains features of the user exposing the application and builds a mathematical model that characterizes the user based on the characteristics of the user. The portrayal of the user is in fact to find out the characteristics of the user. The methods for obtaining user characteristics include the following methods:

Method 1: Characterize the user's mathematical model through the user's natural attributes. For example, the user's mathematical model is characterized according to the age of the user and the gender of the user. Among them, discretization processing is performed for continuous quantities. For example, the age is discretized into “children”, “youth”, “youth”, “middle age” and “old age”.

Method 2: Characterize the user's mathematical model through the user's social attributes. For example, the user's mathematical model is characterized according to the user's cultural level, the user's occupation, and the user's geographic location.

Method 3: Characterize the user's mathematical model through the user's preference attributes. For example, the user's mathematical model is characterized according to the user's preference for the article, such as the user's favorite military article, the user is portrayed as "military fans." Users like technology articles and portray users as "tech enthusiasts." Users like to watch football news and portray users as "football fans."

In a specific embodiment, a user portrait is formed by summarizing the users depicted by different methods and different dimensions. The user image is used as a mathematical model for depicting the user. For example, the characteristics of user a are “youth”, “college students”, “residents living in Guangzhou” and “technical enthusiasts”. These four characteristics are the mathematical characterization model of user a.

S20. Generate a sample according to the characteristics of the user and the click behavior of the application.

After the server obtains the characteristics of the user and the user's click behavior on the application, the sample is generated according to the characteristics of the user and the click behavior of the application. Specifically, the user's feature is taken as an input variable, and the user's click behavior of the application is used as a target variable to generate a sample.

For a user, expose an application as a sample. Taking the user's characteristics as the input variable and the user's click behavior as the target variable, the user clicks the target variable to be 1, and the user does not click the target variable to be 0. For example, a user has characteristics (labels) as “youth” and “military fans”, and the application exposed to the user is “WeChat”. If the user does not click on the "WeChat" application, the record sample is in the form: x = (1, 1, ...), y = 0. x represents one of the samples. The two fronts of x indicate that the user has the label "Youth" and "Military Fan". Labels that are not available in x are indicated by 0. y=0 indicates a negative sample.

S30. Input the sample into a grouping model, and determine a group to which the sample belongs in the group model.

After the server generates a sample based on the characteristics of the user and the user's click behavior on the application, the server inputs the sample into the cluster model. Also determine the population to which the sample belongs in the clustering model. Specifically, there are multiple groups in the clustering model, and the sample values are obtained in the plurality of groups of the group model, and the result values of the samples in the plurality of groups are obtained, and the sample belongs to the group belonging to the group model according to the result value.

In an embodiment, the clustering model can generate samples by collecting features of several users and corresponding user click behaviors to the application. According to the sample, the clustering model is trained to obtain an accurate clustering model for multiple user groups, and then an application click rate prediction model is established for different user groups. The application click rate estimation model established by different user groups can be the same click rate estimation model or different click rate estimation models.

In other embodiments, the clustering model can also be a predictive clustering model. That is, a clustering model that has been formed for different user groups.

S40. Enter the sample into a click rate estimation model corresponding to the group, and obtain an estimated click rate of the user to the application.

After the server obtains the sample belonging to the group to which the cluster model belongs, the sample is input into the click rate estimation model corresponding to the group, thereby obtaining the estimated click rate of the user to the application. In the clustering model, the click rate prediction model corresponding to each group is the click rate estimation model obtained by the standard formula of logistic regression. In other embodiments, the click rate prediction model corresponding to each group may also be a click rate estimation model obtained by other formulas. In addition, the click rate prediction model corresponding to each group in the cluster model may not be identical.

S50: Push content to the user according to an estimated click rate of the application.

In this embodiment, according to step S10 to step S40, the server may obtain the estimated click rate of the user for each application in the application mall. According to the estimated click rate of the user for each application, the corresponding content can be pushed to the user.

Among them, the content here can include the application itself and the content related to the application. For example, similar applications and application introductions, etc., may also include corresponding advertisements and information content obtained according to the application.

In an embodiment, according to step S10 to step S40, the server may obtain the estimated click rate of the user for each application in the application mall. The application can be pushed to the user based on the estimated click rate of the user for each application.

The above content pushing method generates a sample according to the characteristics of the user and the click behavior of the application, and inputs the sample into the group model to confirm that the sample belongs to the group in the group model. The estimated click rate of the user to the application is obtained according to the click rate estimation model in the group. Finally, the content is pushed to the user based on the estimated click rate of the application to the user. After determining the group to which the user belongs, the content pushing method obtains the estimated click rate according to the click rate estimation model of the group, improves the accuracy of the estimated click rate, and can accurately push the corresponding user according to the estimated click rate of the application. Content. For example, the application can be pushed to the user accurately based on the estimated click rate of the application.

FIG. 2 is a flowchart of a method for a content push method according to another embodiment of the present invention. As shown in FIG. 2, before step S10, the method further includes: steps S101-S105.

S101. Acquire a feature of a historical user of the exposure application and a click behavior to the application.

The server obtains the characteristics of all users exposing the history of the application, as well as the click behavior of each user to the application. Here, the user's click behavior on the application includes the user clicking on the application and the user not clicking on the application. The application can be various applications in the app store, including game apps, social apps, reading apps, system apps, and shopping apps.

In a specific embodiment, users drawn according to different methods and different dimensions are aggregated to form a user portrait. The user image is used as a mathematical model for depicting the user. For example, the characteristics of user a are “youth”, “college students”, “residents living in Guangzhou” and “technical enthusiasts”. These four characteristics are the mathematical characterization model of user a.

S103. Generate a sample according to the characteristics of the historical user and the click behavior of the application.

After the server obtains the characteristics of all the users of the history and the click behavior of each user to the application, the sample is generated according to the characteristics of each user and the click behavior of the application. Specifically, the user's feature is taken as an input variable, and the user's click behavior of the application is used as a target variable to generate a sample.

S105. Train the cluster model including several groups and the click rate prediction model corresponding to the group using the sample.

In the present embodiment, the sample corresponds to a sample generated based on the characteristics of all users of the history and the click behavior of the application. Based on this sample, a clustering model containing several populations can be trained. At the same time, according to the sample, the corresponding click rate estimation model of each group in the cluster model can also be trained. Specifically, the sample training model including several groups is used to classify the samples by group, and the group corresponding to the group is used to train the group's click rate estimation model.

The sample rate prediction model of the group is trained by using the sample corresponding to the group, including: using the sample corresponding to the group, and using the logistic method to train the click rate estimation model according to the gradient descent method. The click rate prediction model is a click rate estimation model obtained from the standard formula of logistic regression. The standard formula of logistic regression is:

The above-mentioned statistic algorithm is used to obtain the click rate prediction model of the group, which is only one of the methods for obtaining the click rate estimation model. In addition to the above-mentioned logic algorithm, other algorithms can be used to obtain the click rate of the group. Estimate the model.

In the present embodiment, the sample training clustering model is used according to a decision tree algorithm. In other embodiments, the sample training clustering model may also be used according to the Naive Bayes algorithm, the SVM (Support Vector Machines) algorithm, or the neural network algorithm. That is to say, the algorithm for using the sample training cluster model is not limited, and the present scheme merely provides several algorithms as explanations.

In summary, in the present embodiment, samples are generated by collecting the characteristics of the historical user and the corresponding user's click behavior on the application. According to the sample, the grouping model including several groups is trained, and the corresponding click rate estimation model of each group in the group model is trained to generate a click rate estimation model for different user groups, and then clicks according to different user groups. The rate estimation model obtains the estimated click rate of the corresponding user to the application, and according to the estimated click rate, the corresponding content can be accurately pushed to the user. For example, based on the estimated click rate, the corresponding application can be accurately pushed to the user.

The following provides a specific implementation of establishing a clustering model:

The input variables of the training samples are generated for all exposure events of the most recent day as described above. For each sample, each sample is assigned to a group during each iteration. For example: group number 3, sample x is a negative sample, the value calculated by the LR model of "group 1" is 0.3 in a certain iteration, and the value calculated by the LR model of "group 2" is 0.2, using "group The value calculated by the LR model of 3" is 0.1, then the error of the LR model of "group 3" is the smallest, and the category of the sample x is "group 3". A clustering model is established using a decision tree algorithm.

C(X) = Card(X).

Card(X) is a decision tree algorithm. The training of this method uses the industry-wide model Card classification algorithm, which will not be described here.

In an embodiment, as shown in FIG. 3, the method for establishing a clustered logistic regression click rate prediction model includes: steps S301-S311.

S301, initialization: randomly divide the samples into n groups, group LR (Logistic Regression, logistic regression), the number of iterations i=0, the maximum number of iterations is m, and the iteration error threshold is c.

In the present embodiment, samples of all users in history are randomly divided into n groups. The number of iterations of the group LR is i. At initialization, the number of iterations i=0. Here, assume that the maximum number of iterations is m and the iteration error threshold is c.

S303. Train an LR model for each group of samples (the positive sample is 1 and the negative sample is 0).

In the present embodiment, a corresponding LR model is trained based on samples of each group. Among them, the positive sample is 1 and the negative sample is 0.

S305, each sample is substituted into each LR model, and the click probability is predicted, and the sample is divided into the group closest to the target value of the sample (the negative sample is closest to 0, and the positive sample is closest to 1).

According to the random grouping of the samples, and training the corresponding LR model of the group according to the samples of each group, each sample in the sample is substituted into the LR model of each group, and each sample is obtained corresponding to each group according to the group. The predicted click rate obtained by the LR model. Each sample is compared according to the predicted click rate obtained by each group, and the sample is divided into the group closest to the target value of the sample (where the negative sample is closest to 0 and the positive sample is closest to 1).

S307, calculating the error of all sample predictions. The error is Msm(i). Msm(i)=sum(|The predicted value of the sample in the LR model of the group - the true value of the sample|).

In the present embodiment, the error value of the predicted click rate of all samples in each group and the true click rate of the sample is calculated. Assuming that the error value is Msm(i), then Msm(i)=sum(|the predicted value of the sample in the LR model of the group - the true value of the sample|).

S309, a judgment condition is established, that is, |Msm(i) - Msm(i-1)|<c Or i>=m.

In the present embodiment, a judgment condition is established for the iterative process. Wherein, the judgment condition is |Msm(i) - Msm(i-1)|<c or i>=m. When the condition does not satisfy the judgment condition, the process returns to step S303. When the condition satisfies the judgment condition, the following step S311 is performed.

S311, training a clustering model for predicting which group of unknown samples belongs to: treating each group as a class, there are a total of n categories, and in the above steps, which group each sample belongs to, as the target variable, each The sample features are used as input variables of the classification, substituted into the classification algorithm, and trained the cluster model.

In the present embodiment, when the condition satisfies the judgment condition in step 309, that is, all the samples are classified into the belonging group. Further, a clustering model is trained to predict the population to which the unknown sample belongs. Specifically, each group is regarded as a class, assuming that there are a total of n categories, in which the group has obtained which group each sample belongs to, as the target variable, each sample feature is used as a classification input variable and substituted into the classification. In the algorithm, the clustering model is trained. In the present embodiment, the classification algorithm selects a decision tree algorithm. In other embodiments, a naive Bayesian algorithm, an SVM algorithm, or a neural network algorithm may also be selected.

FIG. 4 provides a usage scenario of a model of a content push method. The content here is a sample. Sample x can be an application. As shown in FIG. 4, the following steps are included: S401-407.

S401, the input feature data of the unknown sample x is sorted.

S403, substituting the feature data feature of the unknown sample x into the grouping model, and obtaining the unknown sample x belonging to the Kth group.

S405. Substituting the feature data of the unknown sample x into the LR model of the Kth group to obtain an estimated click rate of the unknown sample x.

That is, the input feature data of the unknown sample x is acquired first. The feature data of the unknown sample x is substituted into the cluster model, thereby obtaining the unknown sample x belonging to the group in the cluster model (K-group). The feature data of the unknown sample x is then substituted into the click rate prediction model (LR model) of the group to obtain the estimated click rate of the unknown sample x.

S407, pushing the sample x to the user according to the estimated click rate of the sample x.

According to step S401 to step S405, the estimated click rate of the user on the sample x can be obtained. The sample x is pushed to the user based on the estimated click rate of the user on the sample x.

In an embodiment, step S50 includes step A: pushing the application to the user according to an estimated click rate of the application. Specifically, as shown in FIG. 5, step A includes: S501, confirming that the estimated click rate of the application is greater than a threshold, and pushing the application to the user.

The server may obtain the estimated click rate of the user for each application in the application store according to step S10 to step S40. When the user's estimated click rate for each application in the application store is greater than a threshold (set in advance according to actual demand), the server pushes the application to the user. Therefore, the user can be pushed to the application of interest to the user in the application mall.

In an embodiment, as shown in FIG. 6, step A includes the steps of:

S503: Calculate the estimated click rate of the user for all applications in the application store, sort the application from high to low according to the estimated click rate, and push the preset number of applications to the user.

The server respectively obtains the estimated click rate of the user to the plurality of applications in the application store according to step S10 to step S40. Therefore, the server can calculate the estimated click rate of the user for all applications in the app store. And, the application is sorted from high to low according to the estimated click rate of each application, thereby obtaining the top-preset number of applications, and pushing the preset number of applications to the user. For example, in an app store, you need to push 8 apps to your users. The server obtains an estimated click rate for all applications of the application store in accordance with the scheme of the present invention. Sort all applications from high to low based on the estimated clickthrough rate of each app. Further, get the top 8 apps and push the 8 apps to the user in the app store.

The application pushing method provided by the embodiment may filter the application of the mobile application store according to the estimation of the click rate of the application, so as to push the application that the user is interested in to the user.

The invention also provides a content pushing device, as shown in FIG. The content pushing device includes a first obtaining module 701, a generating module 703, a determining module 705, a second obtaining module 707, and a pushing module 709.

The first obtaining module 701 is configured to acquire a feature of a user of the exposure application and a click behavior to the application. When the server detects the user exposing the application, it acquires the characteristics of the user who exposed the application and obtains the user's click behavior on the application. Here, the user's click behavior on the application includes the user clicking on the application and the user not clicking on the application. The application can be various applications in the app store, including game apps, social apps, reading apps, system apps, and shopping apps.

The generating module 703 is configured to generate a sample according to the characteristics of the user and the click behavior of the application. After the server obtains the characteristics of the user and the user's click behavior on the application, the sample is generated according to the characteristics of the user and the click behavior of the application. Specifically, the user's feature is taken as an input variable, and the user's click behavior of the application is used as a target variable to generate a sample.

A determination module 705 is configured to input the sample into a cluster model to determine a population to which the sample belongs in the cluster model. After the server generates a sample based on the characteristics of the user and the user's click behavior on the application, the server inputs the sample into the cluster model. Also determine the population to which the sample belongs in the clustering model. Specifically, there are multiple groups in the clustering model, and the sample values are obtained in the plurality of groups of the group model, and the result values of the samples in the plurality of groups are obtained, and the sample belongs to the group belonging to the group model according to the result value.

The second obtaining module 707 is configured to input the sample into a click rate estimation model corresponding to the group, and obtain an estimated click rate of the user for the application. After the server obtains the sample belonging to the group to which the cluster model belongs, the sample is input into the click rate estimation model corresponding to the group, thereby obtaining the estimated click rate of the user to the application. In the clustering model, the click rate prediction model corresponding to each group is the click rate estimation model obtained by the standard formula of logistic regression. In other embodiments, the click rate prediction model corresponding to each group may also be a click rate estimation model obtained by other formulas. In addition, the click rate prediction model corresponding to each group in the cluster model may not be identical.

The push module 709 is configured to push content to the user based on the estimated click rate of the application. In this embodiment, according to step S10 to step S40, the server may obtain the estimated click rate of the user for each application in the application mall. According to the estimated click rate of the user for each application, the corresponding content can be pushed to the user.

The invention also provides a computer device. The computer device includes one or more processors, memory, and one or more applications. Wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications configured to perform any of the embodiments described above Content push method.

FIG. 8 is a schematic structural diagram of a computer device according to an embodiment of the present invention. For example, servers, personal computers, and network devices. As shown in FIG. 8, the device includes a processor 803, a memory 805, an input unit 807, and a display unit 809. It will be understood by those skilled in the art that the device structure device illustrated in FIG. 8 does not constitute a limitation on all devices, and may include more or less components than those illustrated, or may combine certain components. The memory 805 can be used to store the application 801 and the various functional modules, and the processor 803 runs the application 801 stored in the memory 805 to perform various functional applications and data processing of the device. The memory can be internal or external, or both internal and external. The internal memory may include a read only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, or a random access memory. The external storage may include a hard disk, a floppy disk, a ZIP disk, a USB disk, a magnetic tape, and the like. The memories disclosed herein include, but are not limited to, these types of memories. The memory disclosed herein is by way of example only, and not limitation.

The input unit 807 is for receiving an input of a signal and receiving a keyword input by the user. The input unit 807 can include a touch panel as well as other input devices. The touch panel can collect touch operations on or near the user (such as the user using any suitable object or accessory such as a finger or a stylus on the touch panel or near the touch panel), and according to a preset The program drives the corresponding connection device; other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as play control buttons, switch buttons, etc.), trackballs, mice, joysticks, and the like. The display unit 809 can be used to display information input by the user or information provided to the user as well as various menus of the computer device. The display unit 809 can take the form of a liquid crystal display, an organic light emitting diode, or the like. The processor 803 is a control center of the computer device that connects various parts of the entire computer using various interfaces and lines, executes or executes software programs and/or modules stored in the memory 803, and calls data stored in the memory to execute Various functions and processing data.

In an embodiment, the computer device includes one or more processors 803, and one or more memories 805, one or more applications 801. Wherein the one or more applications 801 are stored in a memory 805 and configured to be executed by the one or more processors 803 configured to perform the above embodiments Content push method.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. The integrated modules, if implemented in the form of software functional modules and sold or used as stand-alone products, may also be stored in a computer readable storage medium.

A person skilled in the art can understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be executed by a program to execute related hardware. The program may be stored in a computer readable storage medium, and the storage medium may include Memory, disk or disc, etc.

The above is only a part of the embodiments of the present invention, and it should be noted that those skilled in the art can also make several improvements and retouchings without departing from the principles of the present invention. It should be considered as the scope of protection of the present invention.

Claims

A content pushing method, comprising the steps of:

Get the characteristics of the user who exposed the application and the click behavior on the application;

Generating a sample according to the characteristics of the user and the click behavior of the application;

Entering the sample into a grouping model to determine a population to which the sample belongs in the grouping model;

And inputting the sample into a click rate estimation model corresponding to the group, and obtaining an estimated click rate of the user to the application;

Push content to the user based on the estimated clickthrough rate of the app.
The content push method according to claim 1, wherein the pushing the content to the user according to the estimated click rate of the application comprises:

The application is pushed to the user based on the estimated click rate of the application.
The content push method according to claim 1 or 2, wherein the pushing the application to the user according to an estimated click rate of the application comprises:

Confirm that the estimated click rate of the application is greater than the threshold and push the application to the user.
The content pushing method according to any one of claims 1 to 3, wherein the pushing the application to the user according to an estimated click rate of the application comprises:

Calculating the estimated click rate of the user to all applications in the application mall, sorting the applications from high to low according to the estimated click rate, and pushing the preset number of applications to the user.
The content pushing method according to any one of claims 1 to 4, wherein before the acquiring the feature of the user of the exposure application and the click behavior on the application, the method further comprises:

Get the characteristics of the historical user of the exposure application and the click behavior on the application;

Generating a sample according to the characteristics of the historical user and the click behavior of the application;

The sample is used to train a clustering model comprising a plurality of populations and a population corresponding to a click rate prediction model.
The content pushing method according to any one of claims 1 to 5, wherein the training the clustering model comprising a plurality of groups using the sample comprises:

The sample training clustering model is used according to a decision tree algorithm.
The content pushing method according to any one of claims 1 to 6, wherein the training the clustering model including a plurality of groups and the click rate prediction model corresponding to the group using the sample comprises:

The sample is used to train a clustering model comprising several populations, the samples are grouped by group, and the population corresponding category is trained to train the population's click rate prediction model.
The content pushing method according to any one of claims 1 to 7, wherein the using the sample corresponding to the group to train the click rate prediction model of the group comprises:

The click rate prediction model is trained using a logistic algorithm using a sample corresponding to the population.
The content pushing method according to any one of claims 1-8, wherein the click rate prediction model is a click rate estimation model obtained according to a formula of logistic regression; The formula is:

Where X represents the input variable and β represents the weight vector of the different features.
The content pushing method according to any one of claims 1-9, wherein the using the group corresponding to the classification, using the logical algorithm to train the click rate prediction model, comprises:

The group is trained according to the gradient descent method using the logistic algorithm, and the click rate prediction model in the corresponding group is trained.
The content pushing method according to any one of claims 1 to 10, characterized in that the characteristics of the user include a natural attribute, a social attribute or a preference attribute of the user.
The content push method according to any one of claims 1 to 11, wherein the generating a sample according to the characteristics of the user and the click behavior of the application comprises:

A sample is generated by taking the characteristics of the user as an input variable and using the click behavior as a target variable.
A content pushing device, comprising:

a first obtaining module, configured to acquire a feature of a user who exposes the application and a click behavior to the application;

a generating module, configured to generate a sample according to the characteristics of the user and the click behavior of the application;

a determining module, configured to input the sample into a grouping model, and determine a group to which the sample belongs in the grouping model;

a second obtaining module, configured to input the sample into a click rate estimation model corresponding to the group, and obtain an estimated click rate of the user to the application;

A push module for pushing content to the user based on an estimated click rate of the application.
A computer device, comprising:

One or more processors;

Memory

One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications configured to execute A content pushing method according to any one of claims 1 to 12.