CN115496539A - User value prediction method, device, equipment and computer readable medium - Google Patents

Abstract

The application relates to a method, a device, equipment and a computer readable medium for predicting user value. The method comprises the following steps: acquiring meta-characteristics and actual behavior characteristics of a target object; modifying the numerical value of the actual behavior characteristic to obtain a first behavior characteristic and a second behavior characteristic, deriving a first virtual object corresponding to the target object based on the meta characteristic and the first behavior characteristic, and deriving a second virtual object corresponding to the target object based on the meta characteristic and the second behavior characteristic; inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model respectively to obtain a first retention day of the first virtual object to a target product in a target period after a target day and a second retention day of the second virtual object to the target product in the target period after the target day; determining a difference between the first number of days to live and the second number of days to live as the user value of the target object. The method and the device solve the technical problem that the advertisement budget utilization rate is low due to inaccurate prediction of the user value.

Description

User value prediction method, device, equipment and computer readable medium

Technical Field

The present application relates to the field of user growth technologies, and in particular, to a method, an apparatus, a device, and a computer-readable medium for predicting a user value.

Background

With the rapid development of internet technology, enterprises relying on the internet rapidly erupt like bamboo shoots in spring after rain, and the survival of the enterprises needs to depend on the payment of high-value users, such as derivative paid products and members for purchasing free products. Enterprises need to accurately find high-value users with higher payment willingness and high sensitivity to advertisement delivery from users of the enterprises, and then preferentially deliver limited advertisement resources to the high-value users, so that the utilization rate of advertisement budget can be improved. In the pulling of new users, the prediction of the user value is also important, and the utilization rate of the advertisement budget can be improved to the maximum extent only by accurately predicting the user value.

At present, in the related art, in the field of user growth, large enterprises and products often adopt promotion methods such as advertisement putting and coupon issuing to attract users to pay, the put objects are often biased to people with high product retention, but some users have high retention without advertisement putting, if the advertisements are put to the non-advertisement-sensitive users, the advertisement budget is wasted, so that the problem of poor advertisement budget utilization rate due to inaccurate user value prediction exists, the more accurate user value prediction can be more accurately positioned to the advertisement-put-sensitive users with high advertisement retention rate and low advertisement-put-free retention rate, and the utilization rate of the advertisements can be improved to the maximum extent by carrying out advertisement putting on the advertisement budget.

Aiming at the problem of low utilization rate of advertisement budget caused by inaccurate prediction of user value, no effective solution is provided at present.

Disclosure of Invention

The application provides a method, a device and equipment for predicting user value and a computer readable medium, which are used for solving the technical problem of low utilization rate of advertisement budget caused by inaccurate prediction of user value.

According to an aspect of an embodiment of the present application, there is provided a method for predicting a user value, including:

acquiring a meta feature and an actual behavior feature of a target object, wherein the meta feature is used for representing a user attribute of the target object, and the actual behavior feature is used for representing an actual use behavior of the target object on a target product;

modifying the numerical value of the actual behavior characteristic to obtain a first behavior characteristic and a second behavior characteristic, deriving a first virtual object corresponding to the target object based on the meta characteristic and the first behavior characteristic, and deriving a second virtual object corresponding to the target object based on the meta characteristic and the second behavior characteristic, wherein the first behavior characteristic is used for representing that the first virtual object is advertised on a target date and uses a target product on the target date, and the second behavior characteristic is used for representing that the second virtual object is not advertised on the target date and does not use the target product on the target date;

respectively inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model to obtain a first retention day of the first virtual object on a target product in a target period after a target day and a second retention day of the second virtual object on the target product in the target period after the target day, wherein the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

determining a difference between the first number of days to live and the second number of days to live as the user value of the target object.

Optionally, before inputting the features of the first virtual object and the second virtual object into a persistence prediction model, respectively, the method further comprises training the persistence prediction model as follows:

selecting a plurality of reference dates in the history dates;

constructing training samples by taking each reference date as a reference to obtain a plurality of groups of training samples;

and combining and inputting the multiple groups of training samples into a target regression model for training to obtain a retention prediction model.

Optionally, a training sample is constructed based on each reference date, and any one of the sets of training samples is constructed as follows:

constructing meta-characteristics of each user and actual behavior characteristics with reference to a reference date based on the user attribute information and historical behavior data;

classifying and splicing the meta-characteristics and the actual behavior characteristics of each user according to the historical activity period of the user before the reference date to construct an inference sample;

and taking the number of remaining days of the user in the target period after the reference date as a label, and splicing the label of each user and the corresponding reasoning sample to obtain a group of training samples taking the reference date as the reference.

Optionally, the target product includes a video streaming platform, and the constructing of the meta-feature of each user and the actual behavior feature based on the reference date based on the user attribute information and the historical behavior data includes:

using at least one of user gender, age, occupation, resident region, device model, consumption level, education level, video member type, video member start-stop time, and video preference as meta-features;

counting at least one of video accumulated playing times, accumulated playing time length, average playing completion rate and video playing number of a user in a plurality of time periods before a reference date to obtain an in-station statistical characteristic as an actual behavior characteristic;

the login days and the silent days of a preset time period before the user reference date are counted to obtain the user login characteristics as actual behavior characteristics;

counting historical click times of the user to the off-site advertisement delivery within a preset time period before a reference date to obtain advertisement click characteristics as actual behavior characteristics;

and counting the number of times that the user is pulled up by the off-site advertisement in a preset period before the reference date to obtain the off-site advertisement pulling-up characteristic as the actual behavior characteristic.

Optionally, classifying and splicing the meta-features and the actual behavior features of each user according to the historical activity period of the user to construct the inference sample includes:

determining a historical activity period for each user;

screening out the statistical characteristics in the target station corresponding to the historical active time period from the statistical characteristics in the station in a plurality of time periods of each user;

and splicing the meta-characteristics, the in-target station statistical characteristics, the user login characteristics, the advertisement click characteristics and the out-station advertisement pull-up characteristics of each user to obtain an inference sample of the corresponding user.

Optionally, the method for obtaining a set of training samples based on the reference date by splicing the label of each user and the corresponding inference sample with the number of remaining days of the user in the target period after the reference date as the label includes:

sampling inference samples of users who have placed advertisements and pulled up on a reference date, users who have placed advertisements and not pulled up on the reference date and users who have not placed advertisements on the reference date according to a preset proportion;

and splicing the sampled inference sample with the label of the corresponding user to obtain a training sample.

Optionally, after obtaining the retention prediction model, the method further comprises verifying the retention prediction model as follows:

constructing a first virtual object and a second virtual object for each user in the verification sample;

inputting the characteristics of the first virtual object and the second virtual object corresponding to each user into a retention prediction model one by one to obtain the third retention days of the first virtual object corresponding to each user on a target product in a target time period after the advertisement is put on the target date and the fourth retention days of the second virtual object corresponding to each user on the target product in a target time period after the advertisement is not put on the target date;

determining the difference value of the third retention days and the fourth retention days of each user as the user value of the corresponding user;

sequencing according to the sequence of the user value from high to low, and determining a target position in the sequencing according to a target ratio so as to determine a user before the target position as a high-value user and determine a user after the target position as a low-value user;

subtracting the average value of the fourth retention days from the average value of the third retention days of all high-value users to obtain a first gain, and subtracting the average value of the fourth retention days from the average value of the third retention days of all low-value users to obtain a second gain;

and determining that the retention prediction model is verified to be effective in the case that the first gain is larger than the second gain.

According to another aspect of the embodiments of the present application, there is provided a device for predicting a user value, including:

the characteristic acquisition module is used for acquiring the meta-characteristics and the actual behavior characteristics of the target object, wherein the meta-characteristics are used for representing the user attributes of the target object, and the actual behavior characteristics are used for representing the actual use behavior of the target object on the target product;

the object derivation module is used for modifying the numerical value of the actual behavior characteristic to obtain a first behavior characteristic and a second behavior characteristic, deriving a first virtual object corresponding to the target object based on the meta characteristic and the first behavior characteristic, and deriving a second virtual object corresponding to the target object based on the meta characteristic and the second behavior characteristic, wherein the first behavior characteristic is used for representing that the first virtual object is advertised on a target date and uses a target product on the target date, and the second behavior characteristic is used for representing that the second virtual object is not advertised on the target date and does not use the target product on the target date;

the system comprises a forecasting module, a storage forecasting module and a processing module, wherein the forecasting module is used for respectively inputting the characteristics of a first virtual object and a second virtual object into a storage forecasting model to obtain a first storage day of the first virtual object on a target product in a target period after a target day and a second storage day of the second virtual object on the target product in the target period after the target day, which are obtained by the prediction of the storage forecasting model, and the storage forecasting model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

and the value determining module is used for determining the difference value of the first retention days and the second retention days as the user value of the target object.

According to another aspect of the embodiments of the present application, there is provided an electronic device, including a memory, a processor, a communication interface, and a communication bus, where the memory stores a computer program executable on the processor, and the memory and the processor communicate with each other through the communication bus and the communication interface, and the processor implements the steps of the method when executing the computer program.

According to another aspect of embodiments of the present application, there is also provided a computer readable medium having non-volatile program code executable by a processor, the program code causing the processor to perform the above-mentioned method.

Compared with the related art, the technical scheme provided by the embodiment of the application has the following advantages:

the method comprises the steps of obtaining element characteristics and actual behavior characteristics of a target object, modifying numerical values of the actual behavior characteristics to obtain first behavior characteristics and second behavior characteristics, deriving a first virtual object corresponding to the target object based on the element characteristics and the first behavior characteristics, and deriving a second virtual object corresponding to the target object based on the element characteristics and the second behavior characteristics; inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model respectively to obtain a first retention day of the first virtual object to a target product in a target period after a target day and a second retention day of the second virtual object to the target product in the target period after the target day, wherein the first retention day and the second retention day are predicted by the retention prediction model; determining a difference between the first number of days to live and the second number of days to live as the user value of the target object. The method and the device estimate the behaviors of the users under two conditions of advertisement putting and advertisement not putting, thereby obtaining the gain which can be brought by the intervention of putting the advertisement on the target object compared with the intervention of not putting the advertisement, taking the gain as the user value, accurately positioning the advertisement putting to the advertisement sensitive users with high retention rate and low retention rate of not putting the advertisement, ensuring that the advertisement budget can reach the maximum utilization rate, and solving the technical problem of low utilization rate of the advertisement budget caused by inaccurate prediction of the user value.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application.

In order to more clearly illustrate the technical solutions in the embodiments or related technologies of the present application, the drawings needed to be used in the description of the embodiments or related technologies will be briefly described below, and it is obvious for those skilled in the art to obtain other drawings without any creative effort.

FIG. 1 is a diagram illustrating an alternative user value prediction method hardware environment according to an embodiment of the present application;

FIG. 2 is a flow chart of an alternative user value prediction method provided in accordance with an embodiment of the present application;

FIG. 3 is a block diagram of an alternative user value prediction device according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of an alternative electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In the following description, suffixes such as "module", "component", or "unit" used to denote elements are used only for the convenience of description of the present application, and have no specific meaning in themselves. Thus, "module" and "component" may be used in a mixture.

In the related technology, in the user growth field, large enterprises and products often adopt promotion methods such as advertisement putting and coupon issuing to attract users to pay, the putting objects are often biased to people with high product retention, but some users have high retention without advertisement putting, if the advertisements are put to the non-advertisement-sensitive users, the advertisement budget is wasted, so that the problem of poor advertisement budget utilization rate due to inaccurate prediction of user value exists, the more accurate prediction of the user value can be accurately positioned to the advertisement-putting-sensitive users with high advertisement retention rate and low advertisement-not-putting-retention rate, and the utilization rate of the advertisement budget can be furthest improved when the users are subjected to advertisement putting.

To solve the problems mentioned in the background, according to an aspect of the embodiments of the present application, an embodiment of a prediction method of a user value is provided.

Alternatively, in the embodiment of the present application, the method for predicting the user value may be applied to a hardware environment formed by the terminal 101 and the server 103 as shown in fig. 1. As shown in fig. 1, a server 103 is connected to a terminal 101 through a network, and may be configured to provide services (such as a data collection service, a model training service, a user value prediction service, etc.) for the terminal or a client installed on the terminal, and a database 105 may be provided on the server or independent from the server, and may be configured to provide a data storage service for the server 103, where the network includes but is not limited to: a wide area network, a metropolitan area network, or a local area network, and the terminal 101 includes, but is not limited to, a PC, a cell phone, a tablet computer, and the like.

In the embodiment of the present application, a method for predicting a user value may be executed by the server 103, or may be executed by both the server 103 and the terminal 101, as shown in fig. 2, where the method may include the following steps:

step S202, acquiring meta-characteristics and actual behavior characteristics of the target object, wherein the meta-characteristics are used for representing user attributes of the target object, and the actual behavior characteristics are used for representing actual use behaviors of the target object on the target product;

step S204, modifying the numerical value of the actual behavior characteristic to obtain a first behavior characteristic and a second behavior characteristic, deriving a first virtual object corresponding to the target object based on the meta characteristic and the first behavior characteristic, and deriving a second virtual object corresponding to the target object based on the meta characteristic and the second behavior characteristic, wherein the first behavior characteristic is used for representing that the first virtual object is advertised on a target date and uses a target product on the target date, and the second behavior characteristic is used for representing that the second virtual object is not advertised on the target date and does not use the target product on the target date;

step S206, inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model respectively to obtain a first retention day of the first virtual object on a target product in a target period after a target date and a second retention day of the second virtual object on the target product in the target period after the target date, wherein the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

step S208, determining a difference between the first remaining day and the second remaining day as the user value of the target object.

Through the steps S202 to S208, the user behaviors under the two conditions of advertisement putting and advertisement not putting are estimated, so that the gain which can be brought by the intervention of putting the advertisement on the target object compared with the intervention of not putting the advertisement is obtained, the gain is used as the user value, the advertisement sensitive users of the type that the retention rate of the put advertisement is high and the retention rate of the put advertisement is low can be accurately positioned, the maximum utilization rate of the advertisement budget can be ensured, and the technical problem that the utilization rate of the advertisement budget is low due to the fact that the user value is not accurately predicted is solved.

In the technical solution provided in step S202, the target object is a target user whose user value is to be predicted, and the meta-feature of the target object refers to information such as personal attribute, social attribute, etc. of the user, such as user gender, age, resident city, mobile phone model, life style, consumption level, education level, occupation, etc. Further, corresponding to the target product, the meta-features of the user can be extended according to different needs of the target product, if the target product is a video streaming media platform, the extended meta-features further include whether the user purchases a video member, the type of the purchased video member, the start-stop time of the purchased video member, the long and short video preferences of the user, the video type preferences, and the like, and if the target product is a network game, the extended meta-features further include whether the user has a consumption behavior, whether the consumption preference is an attribute promotion type prop or a clothing type prop, whether the user purchases a game member, the type of the purchased game member, the start-stop time of the game member, and the like. The actual behavior characteristics of the target object represent the usage behavior of the user on the target product, taking the target product as a video streaming media platform as an example, the actual behavior characteristics of the user include in-station statistical characteristics such as video playing times, video accumulated playing time, average playing rate, and playing video quantity, etc., user login characteristics such as login days, silent days (i.e., days since last login), etc., of the user on the video streaming media platform in a past period of time, advertisement click characteristics including historical click times of the user on advertisements of the target product delivered out of the station, and advertisement pull-up characteristics including the number of times that the user is pulled up (i.e., is pulled to the target product after clicking the advertisements) due to clicking on the advertisements in the past period of time.

In the technical solution provided in step S204, the step of modifying the numerical value of the actual behavior feature to obtain the first behavior feature may specifically be: silent time =0, user type =2,t day pulled by ad =1,t day login =1,t day ad impression =1, i.e. indicating that there is no silent time for the first virtual object that was advertised on the target date and used the target product on the target date (user type =1 indicating that the user was advertised but not pulled, user type =2 indicating that the user was advertised and pulled, user type =3 indicating that the user was not advertised). The second behavior feature obtained by modifying the value of the actual behavior feature may specifically be: user type =3,t day pulled by ad =0,t day used target product =0, t day logged in =0, t day ad served =0, i.e. indicating that the second virtual object was not advertised, pulled by ad, not using target product on target date, such as not watching video, not logged in to account of target product on target date.

In the technical solutions provided in steps S206 and S208, the retention prediction model is obtained by training using meta-features and actual behavior features of a plurality of users as training data in advance, and the features of the first virtual object corresponding to the target object can be used to predict the next behavior of the target object after the advertisement is delivered to the target object, and the features of the second virtual object corresponding to the target object can be used to predict the next behavior of the target object after the advertisement is not delivered to the target object, so that a gain which can be brought by intervention of delivering the advertisement to the target object compared with that without delivering the advertisement is obtained, and the gain is used as a user value.

The high-value users found in the embodiment of the application are users with large gains, namely users who have very high retention scores (first retention days) when advertisements are placed, and users who have very low retention scores (second retention days) when advertisements are not placed, namely users who are very sensitive to advertisement placement. Therefore, the technical scheme of the application can accurately position the advertisement sensitive users, ensure that the advertisement budget can reach the maximum utilization rate, and solve the technical problem of low advertisement budget utilization rate caused by inaccurate user value prediction.

In the embodiment of the application, a plurality of characteristic values are changed simultaneously on the basis of reasoning values when advertisement putting gains are calculated, so that two conditions of a user are simulated, and a single characteristic of whether the advertisements are put or not is set, so that a prediction result is more fit with an actual condition.

Optionally, the present application may also use other indexes to predict the User value, such as DAU (day Active User), that is:

respectively inputting the characteristics of the first virtual object and the second virtual object into a DAU prediction model to obtain a first DAU of the first virtual object on a target product in a target period after a target date and a second DAU of the second virtual object on the target product in the target period after the target date, wherein the DAU prediction model is obtained by using the meta characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

and determining the difference value of the first DAU and the second DAU as the user value of the target object.

The embodiment of the present application further provides a training method of the retention prediction model, which is described in detail below.

Optionally, before inputting the features of the first virtual object and the second virtual object into a persistence prediction model, respectively, the method further comprises training the persistence prediction model as follows:

step 1, selecting a plurality of reference dates from historical dates;

step 2, constructing training samples by taking each reference date as a reference to obtain a plurality of groups of training samples;

and 3, combining and inputting the multiple groups of training samples into a target regression model for training to obtain a retention prediction model.

In the embodiment of the application, a group of training samples is constructed by a reference date, so that the robustness of the user value estimation is ensured by using multiple groups of training data. The target regression model may be an xgboost regression model.

Alternatively, if the reference date is represented by T, the training samples are constructed based on each reference date to obtain a plurality of sets of training samples, and any one set of training samples is constructed as follows:

s1, constructing meta-characteristics of each user and actual behavior characteristics with reference date as reference based on user attribute information and historical behavior data;

s2, classifying and splicing the meta-characteristics and the actual behavior characteristics of each user according to the historical activity period of the user before the reference date to construct an inference sample;

and S3, taking the remaining days of the users in the target time period after the reference date as labels, and splicing the labels of the users and the corresponding reasoning samples to obtain a group of training samples taking the reference date as a reference.

Specifically, taking the target product as a video streaming platform for example, the step S1 of constructing the meta-feature of each user and the actual behavior feature based on the reference date based on the user attribute information and the historical behavior data includes:

using at least one of user gender, age, occupation, resident region, device model, consumption level, education level, video member type, video member start-stop time, and video preference as meta-features;

counting at least one of the video accumulated playing times, the accumulated playing time length, the average playing completion rate and the video playing number of a user in a plurality of time periods before the reference date to obtain an in-station statistical characteristic as an actual behavior characteristic;

the login days and the silent days of a preset time period before the user reference date are counted to obtain the user login characteristics as actual behavior characteristics;

counting historical click times of the user to the off-site advertisement delivery within a preset time period before a reference date to obtain advertisement click characteristics as actual behavior characteristics;

and counting the number of times that the user is pulled up by the off-site advertisement in a preset time period before the reference date to obtain an off-site advertisement pulling-up characteristic as an actual behavior characteristic.

In the embodiment of the present application, taking T as a reference date, the cumulative playing times, the cumulative playing time length, the average playing completion rate, the number of played videos, and the like of the past 7 days, 30 days, and 90 days can be counted as the in-station statistical features. Daily registration of the past 15 days (registration of 1 and non-registration of 0) may be counted on the basis of day T. The historical click times of the user to the off-site advertisement released in the past 15 days can be counted by taking the T day as a reference to serve as the advertisement click feature. The pulling-up condition of the user by the off-site advertisement in the past 15 days can be counted as the off-site advertisement pulling-up characteristic by taking the T day as a reference. Users may also be classified according to whether they are advertised and pulled, i.e., user type =1 indicates that the user is advertised but not pulled, user type =2 indicates that the user is advertised and pulled, and user type =3 indicates that the user is not advertised.

Specifically, the step S2 of classifying and splicing the meta-feature and the actual behavior feature of each user according to the historical activity period of the user to construct the inference sample includes:

step 1, determining the historical activity period of each user;

step 2, screening out the statistical characteristics in the target station corresponding to the historical active time period from the statistical characteristics in the station in a plurality of time periods of each user;

and 3, splicing the meta-characteristics, the in-target-station statistical characteristics, the user login characteristics, the advertisement click characteristics and the out-station advertisement pull-up characteristics of each user to obtain an inference sample of the corresponding user.

In the embodiment of the application, statistical characteristic snapshots of T-7 days, T-30 days and T-90 days are used in sample construction and characteristic splicing, so that the condition that the statistical characteristics of users with long silent time are all null is effectively avoided; (for example, if the statistical features of users who jump beyond 30 days are all 0) in T day, the users who are active in 180 days are used as an inference range for explanation, and the inference sample construction time defines the users according to the time circle, and the features are spliced together to obtain a T day inference sample, that is:

1) The inference sample of active users within 7 days is: t-day meta characteristic + T-day intra-station statistical characteristic + T-day user login characteristic + T-day advertisement click characteristic + T-day extra-station advertisement pull-up characteristic;

2) The reasoning sample of the active users in 8-30 days is as follows: the method comprises the steps of T day intra-station statistical characteristics, T-7 day user login characteristics, T day advertisement click characteristics and T day extra-station advertisement pull-up characteristics;

3) The reasoning samples of active users within 31-90 days are as follows: the method comprises the steps of T day intra-station statistical characteristics, T-30 day user login characteristics, T day advertisement click characteristics and T day extra-station advertisement pull-up characteristics;

4) The reasoning samples of active users in 91-180 days are as follows: the statistical characteristic in the T day station + the user login characteristic for T-90 days + the advertisement click characteristic for T days + the advertisement pull-up characteristic outside the T day station.

Specifically, the step S3 uses the remaining days of the user in the target period after the reference date as the labels, and splices the label of each user and the corresponding inference sample to obtain a group of training samples based on the reference date, including:

step 1, sampling inference samples of users who are released with advertisements and pulled up on a reference date, users who are released with advertisements and not pulled up on the reference date and users who are not released with advertisements on the reference date according to a preset proportion;

and 2, splicing the sampled inference sample with the label of the corresponding user to obtain a training sample.

In the embodiment of the present application, the preset ratio may be selected according to actual needs, for example, a user who puts an advertisement on a reference date and is pulled: users who have placed advertisements on the base date and are not pulled: user not advertised on the reference date = 1. By controlling the sampling proportion of various users, the model can effectively learn the prediction of various users, thereby improving the prediction precision of the model.

Optionally, after obtaining the retention prediction model, the method further comprises verifying the retention prediction model as follows:

step 1, constructing a first virtual object and a second virtual object for each user in a verification sample;

step 2, inputting the characteristics of the first virtual object and the second virtual object corresponding to each user into a retention prediction model one by one to obtain the third retention days of the first virtual object corresponding to each user on a target product in a target time period after the advertisement is put on the target date and the fourth retention days of the second virtual object corresponding to each user on the target product in a target time period after the advertisement is not put on the target date;

step 3, determining the difference value of the third retention days and the fourth retention days of each user as the user value of the corresponding user;

step 4, sequencing according to the sequence of the user value from high to low, and determining a target position in the sequencing according to a target ratio so as to determine a user before the target position as a high-value user and determine a user after the target position as a low-value user;

step 5, subtracting the average value of the fourth retention days from the average value of the third retention days of all high-value users to obtain a first gain, and subtracting the average value of the fourth retention days from the average value of the third retention days of all low-value users to obtain a second gain;

and 6, determining that the verification of the retention prediction model is effective when the first gain is larger than the second gain.

In the embodiment of the present application, the target ratio may be selected according to actual needs, such as a high-value user: low value users =3, then in the ranking the first 30% are high value users and the last 70% are low value users.

In the embodiment of the application, after the retention prediction model is used for determining the high-value user and the low-value user, a release experiment can be performed to further verify the effectiveness and the accuracy of the model. The delivery experiment can deliver 70% of the advertisement budget to high-value users and 30% of the advertisement budget to low-value users, and the verification mode of the experiment effect is the same as the steps 5 and 6 in the model verification.

In the embodiment of the application, as the time is continuously advanced, new training samples are generated every day, so that the latest model can be used for predicting the user value every day, and the prediction precision is higher and higher.

The method and the device estimate the behaviors of the users under two conditions of advertisement putting and advertisement not putting, thereby obtaining the gain which can be brought by the intervention of putting the advertisement on the target object compared with the intervention of not putting the advertisement, taking the gain as the user value, accurately positioning the advertisement putting to the advertisement sensitive users with high retention rate and low retention rate of not putting the advertisement, ensuring that the advertisement budget can reach the maximum utilization rate, and solving the technical problem of low utilization rate of the advertisement budget caused by inaccurate prediction of the user value.

According to another aspect of the embodiments of the present application, as shown in fig. 3, there is provided a user value prediction apparatus including:

the feature obtaining module 301 is configured to obtain a meta feature and an actual behavior feature of the target object, where the meta feature is used to represent a user attribute of the target object, and the actual behavior feature is used to represent an actual usage behavior of the target object on the target product;

the object derivation module 303 is configured to modify a numerical value of the actual behavior feature to obtain a first behavior feature and a second behavior feature, derive a first virtual object corresponding to the target object based on the meta-feature and the first behavior feature, and derive a second virtual object corresponding to the target object based on the meta-feature and the second behavior feature, where the first behavior feature is used to characterize that the first virtual object is advertised on the target date and uses a target product on the target date, and the second behavior feature is used to characterize that the second virtual object is not advertised on the target date and does not use the target product on the target date;

the prediction module 305 is configured to input the characteristics of the first virtual object and the second virtual object into a retention prediction model respectively, and obtain a first retention day of the first virtual object on the target product in a target period after the target day and a second retention day of the second virtual object on the target product in the target period after the target day, which are predicted by the retention prediction model, where the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

a value determining module 307 for determining a difference of the first days to remain and the second days to remain as the user value of the target object.

It should be noted that the feature obtaining module 301 in this embodiment may be configured to execute step S202 in this embodiment, the object deriving module 303 in this embodiment may be configured to execute step S204 in this embodiment, the predicting module 305 in this embodiment may be configured to execute step S206 in this embodiment, and the value determining module 307 in this embodiment may be configured to execute step S208 in this embodiment.

It should be noted here that the modules described above are the same as the examples and application scenarios implemented by the corresponding steps, but are not limited to the disclosure of the above embodiments. It should be noted that the modules described above as a part of the apparatus may operate in a hardware environment as shown in fig. 1, and may be implemented by software or hardware.

Optionally, the apparatus for predicting user value further includes a model training module, configured to:

selecting a plurality of reference dates from the historical dates;

constructing training samples by taking each reference date as a reference to obtain a plurality of groups of training samples;

and combining and inputting the multiple groups of training samples into a target regression model for training to obtain a retention prediction model.

Optionally, the model training module includes a training sample construction unit, configured to construct any one set of training samples according to the following manner:

constructing meta-characteristics of each user and actual behavior characteristics with reference to a reference date based on the user attribute information and historical behavior data;

classifying and splicing the meta-characteristics and the actual behavior characteristics of each user according to the historical activity period of the user before the reference date to construct a reasoning sample;

and taking the number of remaining days of the user in the target period after the reference date as a label, and splicing the label of each user and the corresponding reasoning sample to obtain a group of training samples taking the reference date as the reference.

Optionally, the target product comprises a video streaming platform, and the training sample construction unit is further configured to construct the user characteristics in the following manner:

using at least one of user gender, age, occupation, resident region, device model, consumption level, education level, video member type, video member start-stop time, and video preference as meta-features;

counting at least one of the video accumulated playing times, the accumulated playing time length, the average playing completion rate and the video playing number of a user in a plurality of time periods before the reference date to obtain an in-station statistical characteristic as an actual behavior characteristic;

the login days and the silent days of a preset time period before the user reference date are counted to obtain the user login characteristics as actual behavior characteristics;

counting historical click times of a user on the off-site advertisement released within a preset time period before a reference date to obtain advertisement click characteristics as actual behavior characteristics;

and counting the number of times that the user is pulled up by the off-site advertisement in a preset period before the reference date to obtain the off-site advertisement pulling-up characteristic as the actual behavior characteristic.

Optionally, the training sample construction unit is further configured to construct the inference sample in the following manner:

determining a historical activity period for each user;

screening out the statistical characteristics in the target station corresponding to the historical active time period from the statistical characteristics in the station in a plurality of time periods of each user;

and splicing the meta-characteristics, the in-target station statistical characteristics, the user login characteristics, the advertisement click characteristics and the out-station advertisement pull-up characteristics of each user to obtain an inference sample of the corresponding user.

Optionally, the training sample construction unit is further configured to:

sampling inference samples of users who are advertised and pulled up on a reference date, users who are advertised and not pulled up on the reference date, and users who are not advertised on the reference date according to a preset proportion;

and splicing the sampled inference sample with the label of the corresponding user to obtain a training sample.

Optionally, the model training module is further configured to verify the retention prediction model in the following manner:

constructing a first virtual object and a second virtual object for each user in the verification sample;

inputting the characteristics of the first virtual object and the second virtual object corresponding to each user into a retention prediction model one by one to obtain the third retention days of the first virtual object corresponding to each user on a target product in a target time period after the advertisement is put on the target date and the fourth retention days of the second virtual object corresponding to each user on the target product in a target time period after the advertisement is not put on the target date;

determining the difference value of the third retention days and the fourth retention days of each user as the user value of the corresponding user;

sequencing according to the sequence of the user value from high to low, and determining a target position in the sequencing according to a target ratio so as to determine a user before the target position as a high-value user and determine a user after the target position as a low-value user;

subtracting the average value of the fourth retention days from the average value of the third retention days of all high-value users to obtain a first gain, and subtracting the average value of the fourth retention days from the average value of the third retention days of all low-value users to obtain a second gain;

and determining that the retention prediction model is verified to be effective in the case that the first gain is larger than the second gain.

According to another aspect of the embodiments of the present application, there is provided an electronic device, as shown in fig. 4, including a memory 401, a processor 403, a communication interface 405, and a communication bus 407, where the memory 401 stores a computer program that is executable on the processor 403, the memory 401 and the processor 403 communicate with each other through the communication interface 405 and the communication bus 407, and the processor 403 implements the steps of the method when executing the computer program.

The memory and the processor in the electronic equipment are communicated with the communication interface through the communication bus. The communication bus may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc.

The Memory may include a Random Access Memory (RAM), and may also include a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.

The Processor may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the Integrated Circuit may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component.

There is also provided, in accordance with yet another aspect of an embodiment of the present application, a computer-readable medium having non-volatile program code executable by a processor.

Optionally, in an embodiment of the present application, a computer readable medium is configured to store program code for the processor to perform the following steps:

acquiring a meta feature and an actual behavior feature of a target object, wherein the meta feature is used for representing a user attribute of the target object, and the actual behavior feature is used for representing an actual use behavior of the target object on a target product;

modifying the numerical value of the actual behavior characteristic to obtain a first behavior characteristic and a second behavior characteristic, deriving a first virtual object corresponding to the target object based on the meta characteristic and the first behavior characteristic, and deriving a second virtual object corresponding to the target object based on the meta characteristic and the second behavior characteristic, wherein the first behavior characteristic is used for representing that the first virtual object is advertised on a target date and uses a target product on the target date, and the second behavior characteristic is used for representing that the second virtual object is not advertised on the target date and does not use the target product on the target date;

respectively inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model to obtain a first retention day of the first virtual object on a target product in a target period after a target day and a second retention day of the second virtual object on the target product in the target period after the target day, wherein the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

determining a difference between the first days to survive and the second days to survive as the user value of the target object.

Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment is not described herein again.

When the embodiments of the present application are specifically implemented, reference may be made to the above embodiments, and corresponding technical effects are achieved.

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

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

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is only one logical functional division, and other divisions may be realized in practice, for example, a plurality of modules or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions may be stored in a computer-readable storage medium if they are implemented in the form of software functional units and sold or used as separate products. Based on such understanding, the technical solutions of the embodiments of the present application may be essentially implemented or make a contribution to the prior art, or may be implemented in the form of a software product stored in a storage medium and including several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk or an optical disk, and various media capable of storing program codes. It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.

The previous description is only an example of the present application, and is provided to enable any person skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A method for predicting a user value, comprising:

acquiring meta-characteristics and actual behavior characteristics of a target object, wherein the meta-characteristics are used for representing user attributes of the target object, and the actual behavior characteristics are used for representing actual use behaviors of the target object on a target product;

modifying the value of the actual behavior feature to obtain a first behavior feature and a second behavior feature, deriving a first virtual object corresponding to the target object based on the meta-feature and the first behavior feature, and deriving a second virtual object corresponding to the target object based on the meta-feature and the second behavior feature, wherein the first behavior feature is used for representing that the first virtual object is advertised on a target date and uses the target product on the target date, and the second behavior feature is used for representing that the second virtual object is not advertised on the target date and does not use the target product on the target date;

inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model respectively to obtain a first retention day of the first virtual object on the target product in a target period after the target day and a second retention day of the second virtual object on the target product in the target period after the target day, wherein the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

determining a difference between the first number of days saved and the second number of days saved as a user value of the target object.

2. The method of claim 1, wherein prior to inputting the features of the first virtual object and the second virtual object into a persistence prediction model, respectively, the method further comprises training the persistence prediction model as follows:

selecting a plurality of reference dates from the historical dates;

constructing training samples by taking each reference date as a reference to obtain a plurality of groups of training samples;

and combining and inputting a plurality of groups of training samples into a target regression model for training to obtain the retention prediction model.

3. The method of claim 2, wherein the constructing of the training samples based on each of the reference dates results in a plurality of sets of the training samples, any set of the training samples being constructed as follows:

constructing meta-characteristics of each user and actual behavior characteristics with the reference date as a reference based on the user attribute information and historical behavior data;

classifying and splicing the meta-characteristics and the actual behavior characteristics of each user according to the historical activity period of the user before the reference date to construct an inference sample;

and taking the remaining days of the users in the target period after the reference date as labels, and splicing the labels of each user and the corresponding reasoning samples to obtain a group of training samples taking the reference date as a reference.

4. The method of claim 3, wherein the target product comprises a video streaming platform, and wherein constructing meta-features of each user and actual behavior features based on the reference date based on the user attribute information and historical behavior data comprises:

using at least one of user gender, age, occupation, resident area, device model, consumption level, education level, video member type, video member start-stop time, and video preference as the meta-feature;

counting at least one of the video accumulated playing times, the accumulated playing time length, the average playing completion rate and the video playing number of the user in a plurality of time periods before the reference date to obtain an in-station statistical characteristic as the actual behavior characteristic;

counting the login days and the silent days of the user in a preset time period before the reference date to obtain user login characteristics as the actual behavior characteristics;

counting historical click times of the user to the off-site advertisement delivery within a preset time period before the reference date to obtain advertisement click characteristics as the actual behavior characteristics;

and counting the number of times that the user is pulled up by the off-site advertisement in a preset time period before the reference date to obtain an off-site advertisement pulling-up characteristic as the actual behavior characteristic.

5. The method of claim 4, wherein the classifying and splicing the meta-features and the actual behavior features of each user according to the historical activity periods of the users to construct the inference sample comprises:

determining a historical activity period for each user;

screening out the intra-station statistical characteristics corresponding to the historical activity periods from the intra-station statistical characteristics of a plurality of periods of each user;

and splicing the meta-characteristics, the in-station statistical characteristics, the user login characteristics, the advertisement click characteristics and the out-station advertisement pull-up characteristics of each user to obtain the reasoning sample of the corresponding user.

6. The method according to claim 3, wherein the tagging the number of days remaining in the target period after the reference date of each user, and the splicing each user's tag with the corresponding inference sample to obtain the set of training samples based on the reference date comprises:

sampling the inference samples of users who have placed advertisements and pulled up on the reference date, users who have placed advertisements and not pulled up on the reference date, and users who have not placed advertisements on the reference date according to a preset ratio;

and splicing the sampled inference sample with the label of the corresponding user to obtain the training sample.

7. The method of claim 2, wherein after obtaining the retention prediction model, the method further comprises validating the retention prediction model as follows:

building the first virtual object and the second virtual object for each user in a validation sample;

inputting the characteristics of the first virtual object and the second virtual object corresponding to each user into the retention prediction model one by one to obtain a third retention day of the first virtual object corresponding to each user on a target product in a target time period after the advertisement is delivered on the target date and a fourth retention day of the second virtual object corresponding to each user on the target product in a target time period after the advertisement is not delivered on the target date;

determining a difference between the third days to live and the fourth days to live of each user as a user value for the corresponding user;

sequencing according to the sequence of the user value from high to low, and determining a target position in the sequencing according to a target ratio so as to determine a user before the target position as a high-value user and determine a user after the target position as a low-value user;

subtracting the average of the fourth days to stay from the average of the third days to stay of all the high value users to obtain a first gain, and subtracting the average of the fourth days to stay from the average of the third days to stay of all the low value users to obtain a second gain;

determining that the surviving prediction model is validated if the first gain is greater than the second gain.

8. An apparatus for predicting user value, comprising:

the characteristic acquisition module is used for acquiring meta-characteristics and actual behavior characteristics of a target object, wherein the meta-characteristics are used for representing user attributes of the target object, and the actual behavior characteristics are used for representing actual use behaviors of the target object on a target product;

an object derivation module, configured to modify a numerical value of the actual behavior feature to obtain a first behavior feature and a second behavior feature, derive a first virtual object corresponding to the target object based on the meta-feature and the first behavior feature, and derive a second virtual object corresponding to the target object based on the meta-feature and the second behavior feature, where the first behavior feature is used to characterize that the first virtual object is advertised on a target date and uses the target product on the target date, and the second behavior feature is used to characterize that the second virtual object is not advertised on the target date and does not use the target product on the target date;

the prediction module is used for respectively inputting the characteristics of the first virtual object and the second virtual object into a retention prediction model to obtain a first retention day of the first virtual object on the target product in a target period after the target day and a second retention day of the second virtual object on the target product in the target period after the target day, wherein the retention prediction model is obtained by using the meta-characteristics and the actual behavior characteristics of a plurality of users as training data in advance;

a value determination module to determine a difference between the first number of days remaining and the second number of days remaining as a user value of the target object.

9. An electronic device comprising a memory, a processor, a communication interface and a communication bus, wherein the memory stores a computer program operable on the processor, and the memory and the processor communicate via the communication bus and the communication interface, wherein the processor implements the steps of the method according to any of the claims 1 to 7 when executing the computer program.

10. A computer-readable medium having non-volatile program code executable by a processor, wherein the program code causes the processor to perform the method of any of claims 1 to 7.

Images