CN110110210A - The method and apparatus that push shows information - Google Patents

Abstract

The present invention provides a kind of method and apparatus that push shows information, belong to field of computer technology.The described method includes: when receiving the acquisition request of displaying information of terminal transmission, obtain the account information of contextual information and the current logon account of the terminal, wherein, the account information include the essential information of account, before current point in time in preset duration the counting of each Starting mode and browsing displaying information, page iden-tity is carried in the acquisition request, the contextual information is other contents in the corresponding page of the page iden-tity in addition to showing information；According to the page iden-tity, the contextual information and the account information, in the displaying information aggregate of storage, displaying information to be presented is determined；The displaying information to be presented is sent to the terminal.Using the present invention, the validity for showing information can be improved.

Description

Method and device for pushing and displaying information

technical field

The invention relates to the field of computer technology, in particular to a method and device for pushing and displaying information.

Background technique

With the development of computer technology and network technology, network news business has been widely promoted. Users can install news applications in the terminal, and browse news information of various channels through the application, such as sports news information, entertainment news, etc. information etc. Usually, when a news application program in a terminal displays news information, it also displays some display information, such as advertisement information, for users to browse.

In related technologies, after the general terminal detects that a news application is started, the terminal will send a display information acquisition request to the display information server, and the server can randomly select some display information to be displayed from the display information collection and push them to the The terminal, the terminal can display the display information pushed by the server in the application.

In the process of realizing the present invention, the inventors have found that the related technologies have at least the following problems:

Since the server randomly pushes the displayed information to the terminal, the displayed information may not meet the actual needs of the user, and the user will not click on it, resulting in a relatively low click-through rate of the displayed information.

Contents of the invention

The embodiment of the present invention provides a method and device for pushing and displaying information. In order to solve the problems of related technologies, the technical solution of the present invention is as follows:

On the one hand, a method for pushing and displaying information is provided, and the method includes:

When receiving the display information acquisition request sent by the terminal, the context information and the account information of the account currently logged in by the terminal are acquired, wherein the account information includes the basic information of the account, and each activation within the preset time period before the current time point counting of ways and display information browsed, the acquisition request carries a page identifier, and the context information is content other than the display information in the page corresponding to the page identifier;

Determine the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information;

Sending the display information to be displayed to the terminal.

In one aspect, a method of determining a click-through rate is provided, the method comprising:

Obtain a set of page identification, context information, account information, and display information. The account information includes the basic information of the account, the count of each startup mode and the display information browsed within the preset time before the current time point. The context information is all other content in the page corresponding to the above page identifier except display information;

Inputting the account information, the context information and the display information set into a preset click-through rate estimation model to obtain the click-through rate of each display information in the display information set.

In one aspect, an apparatus for pushing and displaying information is provided, the apparatus comprising:

An acquisition module, configured to acquire context information and account information of an account currently logged in by the terminal when receiving a display information acquisition request sent by the terminal, wherein the account information includes basic information of the account, preset account information before the current time point Set the count of each startup mode and the display information browsed within the duration, the acquisition request carries a page identifier, and the context information is content other than the display information in the page corresponding to the page identifier;

A determining module, configured to determine the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information;

A sending module, configured to send the display information to be displayed to the terminal.

In one aspect, a device for determining a click-through rate is provided, the device comprising:

The obtaining module is used to obtain page identification, context information, account information and a collection of display information. The account information includes the basic information of the account, the count of each startup mode and the display information browsed within the preset time period before the current time point. The context information is other content in the page corresponding to the page identifier except display information;

A determining module, configured to input the account information, the context information and the display information set into a preset click-through rate estimation model to obtain the click-through rate of each display information in the display information set.

In one aspect, a server is provided, the server includes a processor and a memory, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to realize the method of pushing and displaying information and determining the click-through rate Methods.

In one aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the instruction is loaded and executed by a processor to implement a method for pushing display information and a method for determining a click-through rate.

In the embodiment of the present invention, when determining the display information to be displayed, it is determined in combination with the account information and context information of the account, which relatively meets the needs of the user and increases the probability of the user clicking on the display information, thereby making the display information displayed The effectiveness is relatively high.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1a is a schematic diagram of an application scenario for pushing and displaying information provided by an embodiment of the present invention;

Fig. 1b is a schematic flowchart of a method for pushing and displaying information provided by an embodiment of the present invention;

FIG. 2 is a schematic display of display information provided by an embodiment of the present invention;

FIG. 3 is a schematic diagram of a training click rate prediction model provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of a data processing method provided by an embodiment of the present invention;

Fig. 5 is a schematic segmentation diagram of a minimum description principle provided by an embodiment of the present invention;

FIG. 6 is a schematic flowchart of a click-through rate estimation provided by an embodiment of the present invention;

Fig. 7 is a schematic structural diagram of a device for pushing and displaying information provided by an embodiment of the present invention;

Fig. 8 is a schematic structural diagram of a device for pushing and displaying information provided by an embodiment of the present invention;

FIG. 9 is a schematic structural diagram of a device for determining a click-through rate provided by an embodiment of the present invention;

Fig. 10 is a schematic structural diagram of a server provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the implementation manner of the present invention will be further described in detail below in conjunction with the accompanying drawings.

In related technologies, after a user opens a news application, the terminal will send a news information acquisition request to the news information server, and at the same time send a display information acquisition request to the information display server. After receiving the news information acquisition request, the news information server may acquire the news information, and then send the news information to the terminal. After the information display server receives the display information acquisition request sent by the terminal, the information display server can obtain the pre-stored display information set, randomly determine the display information to be displayed from the display information set, and then send the waiting information to the terminal. Display information for the display. After receiving the display information to be displayed, the terminal may display the display information to be displayed. In this way, the display information to be displayed is randomly determined by the server displaying the information, which may not meet the user's needs, and the user is unlikely to click on it, thus resulting in relatively low effectiveness of display information display.

In the solution provided by the embodiment of the present invention, after receiving the display information acquisition request sent by the terminal, the display information server can obtain context information, account information of the terminal's current login account, and a preset display information set. The account information includes The basic information of the account, the count of each startup method within the preset time period, and the displayed information browsed, and then according to the page identification, context information, account information, display information set and the preset click-through rate estimation model carried in the acquisition request, in In the display information collection, the display information to be displayed is determined, and the display information to be displayed is sent to the terminal. In this way, the display information to be displayed is determined by combining the account information and context information of the account, relatively meeting the user's needs, increasing the probability of the user clicking on the display information, and thus making the display information display more effective.

Before implementation, first introduce the scene in the embodiment of the present invention, as shown in Figure 1a, the terminal can send a request for obtaining display information to a server for displaying information, and send a request for obtaining news information to a server for news information, news The information server can determine the news information to be displayed and send the news information to be displayed to the terminal. The server displaying the information can determine the displayed information to be displayed and send the displayed information to the terminal. The displayed display information is displayed.

An embodiment of the present invention provides a method for pushing and displaying information, and the executing subject of the method is a server. The server may be a server for displaying information corresponding to a news application program, and a processor, a transceiver, and a memory may be arranged in the server. For receiving and sending messages, the memory can be used to store the data required and generated in the process of pushing and displaying information.

As shown in Figure 1b, the processing flow of this method can be as follows:

Step 101, when receiving a display information acquisition request sent by a terminal, acquire context information and account information of an account currently logged in by the terminal.

In the implementation, when the user wants to watch news on the terminal, he can install a news application on the terminal, and then open the application, the terminal will display the login interface of the application, and the user can use the account he applied for After logging in successfully, the terminal can determine the page to be displayed, then determine the page ID of the page to be displayed, and then send a request for news information to the server for news information, and at the same time send a display to the server for displaying information For information acquisition requests, both of the two acquisition requests carry page identifiers, which are used to indicate the pages to be displayed for displaying information.

The news information server may send the news information to be displayed to the terminal based on the page identifier. When the information display server receives the display information acquisition request sent by the terminal, it may acquire current context information and account information of the terminal's current login account. Among them, the context information refers to the content included in the page that the terminal requests to display, the channel to which the page belongs, etc. The channels include military news channels, social news channels, etc. In addition, for the convenience of understanding, the context information can also be understood as the display information to be displayed Content other than display information on the page. The information display server may determine the content included in the page requested by the terminal to be displayed based on the page identifier carried in the display information acquisition request.

The account information includes the basic information of the account, the count of each start-up method and the display information browsed within the preset time period before the current time point. The basic information of the account includes the age, gender, interests, etc., geographical Location information, terminal model, connected network, etc. The preset duration can be preset by technicians and stored in the server, such as 1 month, 1 week, etc.

There are two types of startup methods: active startup method and passive startup method. The active startup method refers to the method of opening the application program by triggering the icon of the news application program, and the passive startup method refers to the method of opening the application program by triggering a link. Through different types of links, the passive startup methods can be divided into the following categories: the way to open the application by triggering the push message of the news application, the way to open the application by triggering the link of the news information shared by other friends, The way to open the application through the link of display information shared by other friends, etc. The count of a certain startup method refers to the number of times the application is started using the startup method.

Step 102, according to the page identifier, context information and account information, determine the display information to be displayed in the display information set.

In the implementation, after obtaining the account information and context information of the currently logged-in account, the display information server can obtain the set of stored display information, and then obtain the preset click-through rate estimation model, and combine the context information, account information, display information The set is input to the click-through rate estimation model, and the click-through rate corresponding to each display information in the display information set is obtained, and the page ID carried in it is parsed from the acquisition request, and then the parsed page ID and the click corresponding to each display information are used rate, and determine the display information to be displayed from the display information set. Wherein, the click-through rate refers to the ratio of the number of times the display information is clicked to the number of times it is displayed, which can be used to reflect the attention degree of the display information.

It should be noted that the above-mentioned stored display information collection includes the display information stored by the display information operator in the display information server. The display information can be advertisement information, etc., and the data of display information generally includes pictures, links and texts of the display information. Wait.

In addition, since the display collection stored in the server displaying information will be updated along with the release of information, the stored display information collection can also change periodically (period can be one minute, two minutes, etc.), then the display information The display information set acquired by the server is the display information set stored when the display information acquisition request is received.

Optionally, the display information server may determine the classification information corresponding to the browsed display information according to the obtained browsed display information, for example, the browsed display information is display information of electronic products, and the classified information is electronic products. The display information server can input the context information, the basic information in the account information, the count of various startup methods, the classification information of the browsed display information, and the display information set into the click-through rate estimation model, and obtain the The click-through rate corresponding to the information.

Optionally, the information display server may determine the number of display information to be displayed according to the page identifier carried in the acquisition request, and then determine the display information to be displayed. The corresponding step 102 may be processed as follows:

According to the context information, the account information, the display information set and the preset click rate estimation model, in the display information set, select the first number of display information, and determine the first number of display information as the display information to be displayed.

In implementation, the information display server can determine the information contained in the received acquisition request based on the correspondence between the page identifier and the number of display information to be displayed (this correspondence can be pre-stored in the information display server by the technician). The page identifier corresponds to the first number of display information to be displayed. Then input the context information, account information, and display information set into the click-through rate estimation model to obtain the click-through rate corresponding to each display information in the display information set, and then use the click-through rate corresponding to each display information to obtain the click-through rate from the display information set , select the first number of display information, and determine the first number of display information as the display information to be displayed.

Optionally, the server displaying information may determine the display information to be displayed by using the click unit price corresponding to the display information, and the corresponding step 102 may be processed as follows:

According to the context information and account information, determine the click-through rate corresponding to each display information in the stored display information set; determine the display to be displayed according to the click-through rate corresponding to each display information, the click unit price and page ID corresponding to each display information information.

In implementation, after the display information server obtains the account information of the currently logged in account and the stored display information set, it can obtain the preset click-through rate estimation model, and then input the context information, account information, and display information set into the click-through rate The estimation model is used to obtain the click-through rate corresponding to each display information in the display information set.

When the information display operator stores the display information in the display information server, it calibrates the corresponding click unit price for each display information. The click unit price of the display information indicates the amount paid by the provider of the display information for the operator when the display information is clicked once.

The display information server can obtain the click unit price of each display information, and determine the display information to be displayed according to the corresponding click rate of each display information, the click unit price corresponding to each display information, and the page identifier.

Optionally, the product of the click-through rate corresponding to each display information and the click unit price can be used to determine the display information to be displayed, and the corresponding processing can be as follows:

For each display information, determine the first product of the click-through rate corresponding to the display information and the click unit price corresponding to the display information; determine the display information to be displayed according to the first product corresponding to the page identifier and each display information.

In an implementation, for each display information, the display information server may multiply the determined click-through rate by the click unit price to obtain a first product, so that for each display information, there is a corresponding first product. The display information server may determine the second number corresponding to the page identifier carried in the received acquisition request according to the correspondence between the page identifier and the number of display information to be displayed (this correspondence may be pre-stored by the technician in the display information server), and then sort the first products corresponding to each display information in the display information set from large to small, select the second number of first products starting from the largest product, and divide the second number of first products The display information corresponding to the product is determined as the display information to be displayed. In this way, the display information to be displayed can be determined in combination with the click unit price, so the determined display information to be displayed is relatively more accurate.

Optionally, in order to make the determined display information to be displayed more accurate, the click-through rate corresponding to the display information may also be corrected first, and the corresponding processing may be as follows:

According to the preset correction model, the click-through rate corresponding to each display information is corrected separately, and the corrected click-through rate of each display information is obtained; for each display information, the corrected click-through rate and display information of the display information are determined The second product of the corresponding unit price; the display information to be displayed is determined according to the page identifier and the second product corresponding to each display information.

In implementation, after the display information server obtains the account information of the currently logged in account and the stored display information set, it can obtain the preset click-through rate estimation model, and then input the context information, account information, and display information set into the click-through rate The estimation model is used to obtain the click-through rate corresponding to each display information in the display information set.

The information display server can obtain a preset correction model, use the click-through rate corresponding to each display information as an input of the correction model, and obtain the corrected click-through rate of each display information. The information display server can obtain the click unit price of each display information. For each display information, the information display server can multiply the determined corrected click rate and the click unit price to obtain the second product. In this way, for each Display information corresponds to a second product. The display information server may determine the second number corresponding to the page identifier carried in the received acquisition request according to the correspondence between the page identifier and the number of display information to be displayed (this correspondence may be pre-stored by the technician in the display information server), and then sort the second products corresponding to each display information in the display information set from large to small, select the second number of second products starting from the largest product, and divide the second number of second products The display information corresponding to the product is determined as the display information to be displayed. In this way, since the click rate has been corrected, the determined click rate is more accurate, and thus the determined display information is more accurate.

Optionally, in order to make the determined displayed information more accurate, the preset quality coefficient of the displayed information may also be considered, and the corresponding processing may be as follows:

For each display information, determine the second product of the display information and the third product of the preset quality coefficient of the display information; determine the display information to be displayed according to the page identifier and the third product corresponding to each display information.

In the implementation, when the display information is stored in the display information server, the operator of the display information calibrates the corresponding preset quality coefficient for each display information. The preset quality coefficient represents the quality of the display information. Generally, Belonging to (0, 1), the larger the preset quality coefficient of the displayed information, the higher the quality; on the contrary, the smaller the preset quality coefficient of the displayed information, the lower the quality. Small, healthy ads have a higher default quality factor.

For each display information, the display information server may multiply the preset quality coefficient of the display information by the second product to obtain a third product, so that each display information corresponds to a third product. The display information server may sort the third products corresponding to each display information in the display information set from large to small, select the second number of third products from the largest product, and divide the second number of third products The corresponding display information is determined as the display information to be displayed. In this way, since the quality factor is considered, the determined display information is more accurate.

Optionally, the correction method using the correction model can be as follows:

According to the corresponding relationship between the click-through rate range and the correction value in the preset correction model, the correction value corresponding to the click-through rate range corresponding to the click-through rate of each display information is determined, and according to the correction value corresponding to each display information, each The click-through rate corresponding to the displayed information is corrected respectively, and the corrected click-through rate of each displayed information is obtained.

In implementation, the preset correction model is set with a corresponding relationship between the click rate range and the corrected value, and the process of determining the corresponding relationship will be described in detail later.

After the display information server determines the click-through rate corresponding to each display information, it can determine the click-through rate range to which the click-through rate of each display information belongs, and then use the corresponding relationship between the click-through rate range and the correction value to determine the corresponding click-through rate of each display information. Correction values corresponding to the click-through rate ranges to which the click-through rate belongs. The correction value can be in the form of a numerical value (deviation value) or a coefficient. For any display information, if the correction value is a deviation value, the display information server can add the click-through rate corresponding to the display information to the correction value. The corrected click rate corresponding to the displayed information is obtained. If the correction value is a correction coefficient, for any display information, the display information server can multiply the click-through rate corresponding to the display information by the correction value to obtain the corrected click-through rate corresponding to the display information. In this way, each The corrected click-through rate corresponding to the displayed information.

Step 103, sending the display information to be displayed to the terminal.

In implementation, as shown in FIG. 2 , after the display information server determines the display information to be displayed, it may send the display information to the terminal. For each display information to be displayed, the publisher sets a corresponding display position. After receiving the display information to be displayed, the terminal can determine the display position of the display information to be displayed, and display it at the display position of the currently displayed interface. Display information to be displayed.

Optionally, as shown in FIG. 3, in the embodiment of the present invention, a method for determining the click-through rate prediction model is also provided, and the corresponding processing process may be as follows:

Step 301, acquire a sample set.

In the implementation, the information display server can obtain the recorded click logs and exposure logs. Every time the terminal detects the click display information, it will send the click log to the information display server. The information display server will record the click logs, and then can record every The account information of each displayed information that was clicked. If a certain displayed information is not clicked, the account information corresponding to the displayed information is 0. In addition, an exposure log is also recorded on the server displaying the information, and the exposure log is used to record the number of times the display information is displayed, the display time, and the like.

The server displaying information can obtain a sample collection through click logs and exposure logs. The sample collection includes the content of multiple display information within a period of time, the account information corresponding to each display information, and the corresponding account information of each display information. contextual information. The account information includes age, gender, interests, display information of historical browsing, and time of browsing display information (that is, time of clicking display information), counting of various activation methods, location information, etc. Context information refers to other content on the page where the information is displayed and the channel to which the page belongs. Each of the above-mentioned information (age, gender, etc.) can be called a feature of each dimension corresponding to the information displayed. In addition, for each display information, the actual click rate is calculated correspondingly, that is, the number of times clicked is divided by the number of times displayed. In this way, for each display information, the content of the display information, the account information of the user who clicks the display information, context information, and actual click rate (the actual click rate is a label of the display information, not a feature) are marked.

In addition, before performing the subsequent step 302, the obtained data can also be filtered. If the same display information is clicked twice by the same user for less than a preset duration (such as 2 seconds, etc.), it is considered to be a malicious click. Delete the two click records.

The characteristics and labels (click-through rate) of each displayed information are represented in the form of a table, as shown in Table 1:

Table I

Step 302, according to the sample set, perform iterative training to obtain a click-through rate prediction model.

In the implementation, after the server displaying the information obtains the sample set, since it is determined that the click-through rate belongs to a two-category problem (that is, there are only two types of output, representing two categories respectively, click (category 1) and no-click (category 0)), You can use the logistic regression algorithm for training, specifically through the Sigmoid function to convert the click-through rate into 0-1. The formula can be expressed as follows:

Assuming that the features corresponding to each display information in the sample set satisfy the linear boundary, for the linear boundary, the boundary form is as follows:

In formula (2), x _i represents each dimension feature, θ _i represents the parameter to be trained corresponding to each dimension feature, T represents transposition, the above θ and x are vectors, θ=[θ ₀ ,θ ₁ ,θ ₂ ,...,θ _n ] ^T .

Construct the prediction function as:

For the input sample set, the probabilities of the classification results being category 1 and category 0 (that is, the probability of the display information being clicked and the probability of not being clicked) are respectively:

P(y=1|x,θ)=h _θ (x), P(y=0|x,θ)=1-h _θ (x) (4)

Combine the probabilities based on formula (4) to get:

P(y|x,θ)＝(h _θ (x)) ^y (1-h _θ (x)) ^1-y (5)

The likelihood function for formula (5) is:

In formula (6), m represents the number of displayed information in the sample set, and each displayed information has n features.

L(θ) in formula (6) is obtained by taking the logarithm:

Then perform maximum likelihood estimation on formula (7), so that l(θ) takes the maximum value, and θ is obtained, and θ is the optimal value.

Wherein, when performing maximum likelihood estimation, the embodiment of the present invention also provides an iterative update process of θ:

Assuming that the loss function is J(θ), the loss function J(θ) is taken as:

Because a negative coefficient (-1/m) is multiplied, when the minimum value of J(θ) is taken, it is the maximum value of l(θ), and the optimal value of θ is obtained.

Then use the gradient descent method to find the minimum value of J(θ).

The update process of θ is given below:

Among them, in formula (9), j represents the number of iterations.

In this way, combined with formula (10), formula (9) can be expressed as:

In this way, based on the above multiple iterations, the minimum value of J(θ) is obtained, and the optimal value of θ is obtained, and the optimal value of θ is substituted into the prediction function A click-through rate prediction model is also obtained, which can be used to predict the click-through rate of displayed information later.

Optionally, since in the embodiment of the present invention, the format of some feature data in the sample set may not be the format of the training data, it is also possible to first process the data in various formats in the sample set to obtain the model training Available formats, corresponding processing can be as follows:

Process the data in the sample set to obtain data that can be used for training; according to the data that can be used for training, iteratively train to obtain a click-through rate prediction model.

In implementation, the server for displaying information may process multiple data for displaying information, process data for account information corresponding to each displaying information, and process data for context information corresponding to each displaying information , to get the processed data.

Then use the processed data to perform iterative training to obtain a click-through rate prediction model. How to train and obtain the click-through rate prediction model has been described in detail above, and will not be repeated here.

Optionally, the processing method for the data of various characteristics is as follows:

The data of each feature is processed according to the processing method corresponding to each feature.

In implementation, as shown in Figure 4, the data of general characteristics can be divided into disordered type, continuous type and text type, etc., for disordered type, there are the following several processing methods: feature domain (limited data range), Outlier processing (marking outliers as null values or undetected features), vectorization (classification processing, clustering processing, etc.), hash strategy (hash strategy can have hash space and hash The choice of function, there are many choices of hash function, such as MurMuHash32, MD5SUM, etc., the hash space has global hash and local hash), feature effective frequency (preliminary selection by controlling the frequency of feature appearance, that is to say, the occurrence of Only a certain number of times will be counted).

As shown in Figure 4, for the continuous type, there are the following processing methods: outlier processing (marking outliers as null values or as undetected features), eigenvalue upper and lower bounds (determining the upper bound of eigenvalues and Lower bound), feature discretization (equal frequency discretization, equidistant discretization, entropy-based partitioning, etc.), feature normalization (extreme value offset and relative width).

Multiple data displaying information are generally pictures and texts, etc., so data belonging to the disordered type and text type can be processed according to any of several processing methods of the disordered type. The contextual information generally includes the content displayed on the page, which is generally pictures and texts, and also belongs to unordered data and text data. For the data of the account information corresponding to each display information, since the account information includes age, gender, interest, geographic location information, terminal model, count of various startup methods, time when the display information is clicked, display information of historical browsing, etc. . Therefore, there are both disordered data and continuous data in the account information data, so they are all processed by their own corresponding processing. The following describes the counting processing methods of each startup mode:

According to the minimum description length principle, the counts of each startup mode are processed.

Generally, the counts of each startup mode are one-hot encoded, and the one-hot encoded is a one-bit effective code. For example, there are three colors, yellow, green and red, and yellow is 0, green is 1, and red is 2. After one-hot After encoding, yellow can be expressed as (1, 0, 0), green can be expressed as (0, 1, 0), and red can be expressed as (0, 0, 1). Since the corresponding vector increases by one dimension each time it is started, the vector will be lengthy, so it is necessary to process the eigenvalues of such features. Here, the principle of minimum description length is used for processing and segmentation processing. As shown in Figure 5, add the feature value to be split to the queue to be split, first check whether the queue to be split is empty, if it is empty, end, if not empty, obtain the feature value from it, and perform segmentation. Carry out the first split, calculate the entropy of the feature value before split, E=SUM(p _i *log p _i ), calculate the entropy after split, SplitE=(|S1|*E ₁ + |S2|*E ₂ )/|S|, S1 represents the length of the first segment after segmentation, S2 represents the length of the second segment after segmentation, S represents the total length before segmentation, and E ₁ represents the segmentation After the entropy of the first segment, E ₂ represents the entropy of the second segment after segmentation. Then calculate the information gain E-SplitE, if the information gain is greater than the preset value (the preset value can be preset by the technician and stored in the server displaying the information), put the split feature value into the queue to be split , continue to split until the information gain after the Nth split is less than the preset value, determine the N-1th split as the best split, and then re-judge whether the queue to be split is empty, if the information If the gain is less than the preset value, it will re-judge whether the queue to be split is empty.

For example, if the count of one active start is 1, and the sum of the active start modes is 100, then the vector that is actively started once after one-hot encoding is a 100-dimensional vector. After segmentation, the segmentation result of the active start mode is 0, 2, 6, 12, 34, 100, that is, 100 is divided into 5 segments, namely (0, 2], (2, 6], (6, 12], (12, 34], (34, 100 ], in this way, the vector of active start once is (1, 0, 0, 0, 0), and the dimension of active start 5 times is (0, 1, 0, 0, 0), so the dimension of the vector will be reduced.

In addition, in the embodiment of the present invention, a complete training and use flowchart is also provided, as shown in Figure 6, which can be divided into a data processing platform, a machine learning platform, and a display information push platform, and the data processing platform is used to collect sample sets , the machine learning platform is used to train data to obtain a click-through rate estimation model, and the display information push platform is used to push display information based on the click-through rate estimation model. specific:

The display information server obtains click logs, exposure logs, multiple display information and account information corresponding to each display information through the data processing platform, and then marks the obtained data, that is, marks each of the multiple display information. Display the account information, click rate and context information corresponding to the information. The server displaying information transmits the above-mentioned sample set to the machine learning platform through the data processing platform, and the machine learning platform can process the data in the sample set obtained to obtain data that can be used for training, that is, to obtain standardized training log, and then train to obtain a click-through rate prediction model. The click-through rate estimation model is sent to the display information push platform, and the display information can be pushed through the click-through rate estimate model in the display information push platform.

In addition, a correction model is also set in the machine learning platform, and the method of determining the correction model can be as follows:

The machine learning platform predicts the click-through rate of the display information in the sample set through the click-through rate estimation model, obtains the estimated click-through rate, compares it with the actual click-through rate, and then divides the actual click-through rate according to the clustering method into For multiple CTR ranges, the difference between the actual CTR contained in a certain CTR range and the corresponding estimated CTR is made, and the multiple difference values obtained are averaged, and the obtained average value is the click The corrected value corresponding to the click-through rate range, and the corrected value corresponding to each click-through rate range can be obtained based on the same method. In this way, the corresponding relationship between the click-through rate range and the correction value in the correction model is obtained.

In addition, after the correction model is determined, the correction model can be sent to the display information push platform through the machine learning platform.

It should be noted that, in the embodiment of the present invention, the data obtained in the sample set are recorded using distributed methods.

It should also be noted that, in the embodiment of the present invention, the server for the news information mentioned and the server for displaying the information may also be the same server.

In addition, since a start-up method is added when training the click-through rate prediction model, the AUC (Area under curve, area under the curve) corresponding to the trained click-through rate prediction model can be increased. Generally, the larger the value of AUC, the better the classifier, so combined with the startup method, the click rate prediction model can be more accurate.

In the embodiment of the present invention, when determining the display information to be displayed, it is determined in combination with the account information and context information of the account, which relatively meets the needs of the user and increases the probability of the user clicking on the display information, thereby making the display information displayed The effectiveness is relatively high.

In addition, in the embodiment of the present invention, a method for determining the click-through rate based on the click-through rate estimation model is also provided. The specific method can be as follows:

Obtain a collection of page identification, context information, account information, and display information. Account information includes basic information of the account, the count of each startup method within the preset time period before the current time point, and display information browsed. The context information is the page corresponding to the page identification. In addition to the display information in other content; input the account information, context information and display information set into the preset click-through rate estimation model, and obtain the click-through rate of each display information in the display information set.

For the specific processing process, reference may be made to the process of determining the click-through rate described in the foregoing embodiments, which will not be repeated here.

In the embodiment of the present invention, since the click-through rate is calculated in combination with the activation method, the calculated click-through rate can be made more accurate.

Based on the same technical concept, the embodiment of the present invention also provides a device for pushing and displaying information, as shown in Figure 7, the device includes:

The obtaining module 710 is configured to obtain context information and account information of an account currently logged in by the terminal when receiving a display information acquisition request sent by the terminal, wherein the account information includes basic information of the account, and the account information before the current time point The count of each startup mode and the displayed information browsed within the preset time period, the acquisition request carries a page identifier, and the context information is other content in the page corresponding to the page identifier except the displayed information;

A determining module 720, configured to determine the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information;

A sending module 730, configured to send the display information to be displayed to the terminal.

Optionally, the determining module 720 is configured to:

According to the context information and the account information, determine the click rate corresponding to each display information in the stored display information set;

The display information to be displayed is determined according to the click rate corresponding to each display information, the click unit price corresponding to each display information, and the page identifier.

Optionally, the determining module 720 is configured to:

For each display information, determine the first product of the click rate corresponding to the display information and the click unit price corresponding to the display information;

The display information to be displayed is determined according to the first product corresponding to the page identifier and each display information.

Optionally, the determining module 720 is configured to:

Correcting the click-through rate corresponding to each display information according to a preset correction model to obtain the corrected click-through rate of each display information;

For each display information, determine the second product of the corrected click-through rate of the display information and the unit price corresponding to the display information;

The display information to be displayed is determined according to the second product corresponding to the page identifier and each display information.

Optionally, the determining module 720 is configured to:

For each display information, determining a second product of the display information and a third product of a preset quality factor of the display information;

The display information to be displayed is determined according to the third product corresponding to the page identifier and each display information.

Optionally, the determining module 720 is configured to:

According to the correspondence between the click-through rate range and the correction value in the preset correction model, determine the correction value corresponding to the click-through rate range to which the click-through rate corresponding to each display information belongs;

According to the correction value corresponding to each piece of display information, the click-through rate corresponding to each piece of display information is respectively corrected to obtain the corrected click-through rate of each piece of display information.

Optionally, the determining module 720 is configured to:

determining a first number of display information to be displayed according to the page identifier;

According to the context information and the account information, in the display information set, select the first number of display information, and determine the first number of display information as the display information to be displayed.

Optionally, the determining module 720 is configured to:

Inputting the page identifier, the context information, the account information and the stored display information set into a preset click-through rate estimation model to obtain the display information to be displayed.

The obtaining module 710 is also used for:

Acquiring a sample set, wherein the sample set includes a plurality of display information, and account information, context information, and display times corresponding to each display information in the plurality of display information;

As shown in Figure 8, the device also includes:

The training module 740 is configured to perform iterative training to obtain the click-through rate prediction model according to the sample set.

Optionally, the training module 740 is configured to:

Processing the data in the sample set to obtain data that can be used for training;

According to the data available for training, the click-through rate prediction model is obtained through iterative training.

In the embodiment of the present invention, when determining the display information to be displayed, it is determined in combination with the account information and context information of the account, which relatively meets the needs of the user and increases the probability of the user clicking on the display information, thereby making the display information displayed The effectiveness is relatively high.

It should be noted that: when the device for pushing display information provided by the above embodiment pushes display information, it only uses the division of the above-mentioned functional modules for illustration. In practical applications, the above-mentioned functions can be assigned to different functional modules according to needs. To complete means to divide the internal structure of the device into different functional modules to complete all or part of the functions described above. In addition, the device for pushing display information provided in the above embodiment and the embodiment of the method for pushing display information belong to the same idea, and its specific implementation process is detailed in the method embodiment, and will not be repeated here.

Based on the same technical concept, the embodiment of the present invention also provides a device for determining the click-through rate, as shown in Figure 9, the device includes:

The obtaining module 910 is used to obtain page identification, context information, account information and a collection of display information. The account information includes the basic information of the account, the count of each startup mode and the display information browsed within the preset time period before the current time point, The context information is content other than display information in the page corresponding to the page identifier;

The determining module 920 is configured to input the account information, the context information and the display information set into a preset click-through rate estimation model to obtain the click-through rate of each display information in the display information set.

In the embodiment of the present invention, since the click-through rate is calculated in combination with the activation method, the calculated click-through rate can be made more accurate.

It should be noted that: the device for determining the click-through rate provided by the above-mentioned embodiments only uses the division of the above-mentioned functional modules as an example to illustrate when determining the click-through rate. In practical applications, the above-mentioned functions can be assigned to different functional modules according to needs To complete means to divide the internal structure of the device into different functional modules to complete all or part of the functions described above. In addition, the device for determining the click-through rate provided by the above embodiment and the embodiment of the method for determining the click-through rate belong to the same idea, and its specific implementation process is detailed in the method embodiment, and will not be repeated here.

Please refer to FIG. 10 , which shows a schematic structural diagram of a server involved in an embodiment of the present invention, and the server may be used to implement the method for pushing display information provided in the above embodiments. Specifically:

The server 1000 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (central processing units, CPU) 1022 (for example, one or more processors) and memory 1032, one or more The above storage medium 1030 (for example, one or more mass storage devices) for storing application programs 1042 or data 1044 . Wherein, the memory 1032 and the storage medium 1030 may be temporary storage or persistent storage. The program stored in the storage medium 1030 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1022 may be configured to communicate with the storage medium 1030 , and execute a series of instruction operations in the storage medium 1030 on the server 1000 .

The server 1000 can also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input and output interfaces 1058, one or more keyboards 1056, and/or, one or more operating systems 1041 , such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM and so on.

The server 1000 may include a memory, and one or more programs, wherein one or more programs are stored in the memory, and configured to be executed by one or more processors, including for performing the above-mentioned Push display information and methods to determine click-through rate.

An embodiment of the present invention also provides a computer-readable storage medium, at least one instruction is stored in the computer-readable storage medium, and the instruction is loaded and executed by a processor to implement the above-mentioned method for pushing display information and determining the click-through rate Methods.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (15)

1. A method for pushing display information, characterized in that the method comprises: When receiving the display information acquisition request sent by the terminal, the context information and the account information of the account currently logged in by the terminal are acquired, wherein the account information includes the basic information of the account, and each activation within the preset time period before the current time point counting of ways and display information browsed, the acquisition request carries a page identifier, and the context information is content other than the display information in the page corresponding to the page identifier; Determine the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information; Sending the display information to be displayed to the terminal. 2. The method according to claim 1, wherein the determining the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information includes: According to the context information and the account information, determine the click rate corresponding to each display information in the stored display information set; The display information to be displayed is determined according to the click rate corresponding to each display information, the click unit price corresponding to each display information, and the page identifier. 3. The method according to claim 2, wherein the display to be displayed is determined according to the click-through rate corresponding to each display information, the click unit price corresponding to each display information, and the page identifier information, including: For each display information, determine the first product of the click rate corresponding to the display information and the click unit price corresponding to the display information; The display information to be displayed is determined according to the first product corresponding to the page identifier and each display information. 4. The method according to claim 2, wherein the display to be displayed is determined according to the click-through rate corresponding to each display information, the click unit price corresponding to each display information, and the page identifier information, including: Correcting the click-through rate corresponding to each display information according to a preset correction model to obtain the corrected click-through rate of each display information; For each display information, determine the second product of the corrected click-through rate of the display information and the unit price corresponding to the display information; The display information to be displayed is determined according to the second product corresponding to the page identifier and each display information. 5. The method according to claim 4, wherein the determining the display information to be displayed according to the second product corresponding to the page identifier and each display information includes: For each display information, determining a second product of the display information and a third product of a preset quality factor of the display information; The display information to be displayed is determined according to the third product corresponding to the page identifier and each display information. 6. The method according to claim 3, characterized in that, according to the preset correction model, the click-through rate corresponding to each display information is respectively corrected to obtain the corrected click-through rate of each display information. CTR, including: According to the correspondence between the click-through rate range and the correction value in the preset correction model, determine the correction value corresponding to the click-through rate range to which the click-through rate corresponding to each display information belongs; According to the correction value corresponding to each piece of display information, the click-through rate corresponding to each piece of display information is respectively corrected to obtain the corrected click-through rate of each piece of display information. 7. The method according to claim 1, wherein the determining the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information includes: determining a first number of display information to be displayed according to the page identifier; According to the context information and the account information, in the display information set, select the first number of display information, and determine the first number of display information as the display information to be displayed. 8. The method according to claim 1, wherein the determining the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information includes: Inputting the page identifier, the context information, the account information and the stored display information set into a preset click-through rate estimation model to obtain the display information to be displayed. 9. The method of claim 8, further comprising: Acquiring a sample set, wherein the sample set includes a plurality of display information, and account information, context information, and display times corresponding to each display information in the plurality of display information; According to the sample set, iteratively train to obtain the click rate prediction model. 10. The method according to claim 9, wherein the iterative training to obtain the click-through rate prediction model according to the sample set comprises: Processing the data in the sample set to obtain data that can be used for training; According to the data available for training, the click-through rate prediction model is obtained through iterative training. 11. A method for determining click-through rate, characterized in that the method comprises: Obtain a set of page identification, context information, account information, and display information. The account information includes the basic information of the account, the count of each startup mode and the display information browsed within the preset time before the current time point. The context information is all other content in the page corresponding to the above page identifier except display information; Inputting the account information, the context information and the display information set into a preset click-through rate estimation model to obtain the click-through rate of each display information in the display information set. 12. A device for pushing and displaying information, characterized in that the device comprises: An acquisition module, configured to acquire context information and account information of an account currently logged in by the terminal when receiving a display information acquisition request sent by the terminal, wherein the account information includes basic information of the account, preset account information before the current time point Set the count of each startup mode and the display information browsed within the duration, the acquisition request carries a page identifier, and the context information is content other than the display information in the page corresponding to the page identifier; A determining module, configured to determine the display information to be displayed in the stored display information set according to the page identifier, the context information and the account information; A sending module, configured to send the display information to be displayed to the terminal. 13. A device for determining a click-through rate, wherein the method comprises: The obtaining module is used to obtain page identification, context information, account information and a collection of display information. The account information includes the basic information of the account, the count of each startup mode and the display information browsed within the preset time period before the current time point. The context information is other content in the page corresponding to the page identifier except display information; A determining module, configured to input the account information, the context information and the display information set into a preset click-through rate estimation model to obtain the click-through rate of each display information in the display information set. 14. A server, characterized in that the server comprises a processor and a memory, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement any one of claims 1 to 10. The method for pushing and displaying information and the method for determining the click-through rate as claimed in claim 11. 15. A computer-readable storage medium, characterized in that at least one instruction is stored in the storage medium, and the instruction is loaded and executed by a processor to realize pushing display information according to any one of claims 1 to 10 and the method for determining the click-through rate as claimed in claim 11.