CN112612826A

CN112612826A - Data processing method and device

Info

Publication number: CN112612826A
Application number: CN202011522170.8A
Authority: CN
Inventors: 李健伟; 高梓尧; 刘子岳; 黄岑
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2020-12-21
Filing date: 2020-12-21
Publication date: 2021-04-06
Anticipated expiration: 2040-12-21
Also published as: CN112612826B

Abstract

The application discloses a data processing method and device. The method comprises the following steps: acquiring a first historical behavior characteristic of a user account before the current moment; inputting a first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of a user account output by the survival probability model in each prediction time period in a prediction period, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; acquiring the active time of a user account for accessing a service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in a prediction period; and determining the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the influence degree and the influence direction of the first historical behavior characteristics and/or the service characteristics of the service on the activity gain. Through the application, the effect of effectively improving the activity of the user is achieved.

Description

Data processing method and device

Technical Field

The present application relates to the field of data processing, and in particular, to a data processing method and apparatus.

Background

In recent years, the field of internet videos is rapidly developed, wherein various social software, short video APP (Application), and the like become the first choice for hundreds of millions of users to entertain and entertain every day, and many internet enterprises are rapidly developing related services such as short videos and content push, wherein how to improve the user activity and increase the user viscosity become the core concern of each service.

At present, a main means for improving the user liveness is through a personalized recommendation strategy, a common method for recommending high-quality push content comprises consumption-based sequencing, conversion rate-based sequencing and the like, the methods are popular and easy to understand, but a great recommendation algorithm short board exists, so that video selection depends on short-term feedback (click rate, duration, praise and the like), the video selection is not matched with a target for improving the user liveness, and meanwhile, videos selected by the method are all head videos, so that the Martian effect of platform flow distribution is very serious.

The inventors have found that in order to solve the drawbacks of ordering recommended service content based on consumption and conversion rate, the solutions of the related art are mainly to describe video quality from the viewpoint of the inherent features of the video, such as:

1. title information of the video;

2. content attribute information, content feedback information and content source information;

3. the image quality information of the video includes a frame rate and a code rate.

However, these methods describe more characteristics such as the definition of the video itself, and no effective solution has been proposed for what contents can enhance the user liveness.

Disclosure of Invention

The present application mainly aims to provide a data processing method and apparatus, so as to solve the problem that when content is pushed to a user, it is not possible to know which content can improve the user activity.

In order to achieve the above object, according to an aspect of the present application, there is provided a data processing method including: acquiring a first historical behavior characteristic of a user account before the current moment; inputting the first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of the user account in each prediction time period in a prediction period, which is output by the survival probability model, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of a sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before a first historical moment, the historical activity information comprises a time interval between the second historical moment and the first historical moment, and the second historical moment is a historical moment when the sample user account is firstly active within a preset time period after the first historical moment; acquiring the active time of the user account for accessing the service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in the prediction period; and determining the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain.

Optionally, the determining the degree and the direction of the influence of the first historical behavior feature and/or the traffic feature of the traffic on the activity gain includes: training a tree model by taking the first historical behavior characteristic and the service characteristic of the service as input and the liveness gain as output; and explaining the trained tree model, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business on the activity gain.

Optionally, taking the first historical behavior feature and the service feature of the service as inputs, and taking the activity gain as an output, the training tree model includes: the step of training the tree model by taking the dimension characteristic of the first historical behavior and the service dimension characteristic of the service as input and the activity gain as output, explaining the trained tree model, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain comprises the following steps: and explaining the trained tree model, and determining the influence degree and the influence direction of the dimensional characteristics of the first historical behaviors and/or the service dimensional characteristics of the service on the activity gain.

Optionally, interpreting the trained tree model, and determining the degree and direction of influence of the first historical behavior feature and/or the business feature of the business on the activity gain, includes: inputting the trained tree model, the first historical behavior feature and the business feature of the business into a model interpreter, and obtaining the influence degree and the influence direction of the first historical behavior feature and/or the business feature of the business output by the model interpreter on the liveness gain.

Optionally, the modifying the first survival curve by using the active time to obtain a second survival curve of the user account in the prediction period includes: acquiring a first survival prediction probability of the user account in each prediction time period in a prediction period; acquiring a second survival prediction probability of the user account in the prediction time period of the active moment; correcting each of the first survival prediction probabilities using the second survival prediction probability; and obtaining the second survival curve based on the corrected first survival prediction probability.

Optionally, after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and the direction of the influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, the method further includes: under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a positive direction, adding pushing of the service comprising the service characteristic to a user account corresponding to the first historical behavior characteristic; and under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a negative direction, reducing the pushing of the service comprising the service characteristic to a user account corresponding to the first historical behavior characteristic.

Optionally, after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and the direction of the influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, the method further includes: if the influence direction of the service characteristics of the service on the liveness gain is a positive direction, pushing the service comprising the service characteristics; if the influence direction of the first historical behavior characteristic on the liveness gain is a positive direction, pushing a service to a client logged in the user account; and if the influence direction of the first historical behavior characteristic and the service characteristic of the service on the liveness gain is a positive direction, pushing the service comprising the service characteristic to a client logged in the user account.

Optionally, after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and the direction of the influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, the method further includes: if the influence degree of the service characteristics of the service on the liveness gain is greater than a first preset degree and the influence direction is a positive direction, pushing the service comprising the service characteristics; if the influence degree of the first historical behavior characteristic on the liveness gain is greater than a second preset degree and the influence direction is a positive direction, pushing a service to a client logging in the user account; and if the influence direction of the first historical behavior characteristic and the service characteristic of the service on the liveness gain is a positive direction, the influence degree of the service characteristic of the service on the liveness gain is greater than a first preset degree, and the influence degree of the first historical behavior characteristic on the liveness gain is greater than a second preset degree, pushing the service including the service characteristic to a client logging in the user account.

Optionally, after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and the direction of the influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, the method further includes: if the influence direction of the service characteristics of the service on the liveness gain is a negative direction, reducing the pushing of the service comprising the service characteristics; if the influence direction of the first historical behavior characteristic on the liveness gain is a negative direction, reducing the pushing of the business to the client logging in the user account; and if the influence direction of the first historical behavior characteristic and the service characteristic of the service on the liveness gain is a negative direction, reducing the service comprising the service characteristic to be pushed to the client logged in the user account.

Optionally, after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and the direction of the influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, the method further includes: if the influence degree of the service characteristics of the service on the liveness gain is greater than a third preset degree and the influence direction is a negative direction, reducing the pushing of the service comprising the service characteristics; if the influence degree of the first historical behavior characteristic on the liveness gain is greater than a fourth preset degree and the influence direction is a negative direction, reducing the business pushed to the client logging in the user account; if the first historical behavior characteristic and the business characteristic of the business have negative influence on the liveness gain, the influence degree of the business characteristic of the business on the liveness gain is greater than a third preset degree, and the influence degree of the first historical behavior characteristic on the liveness gain is greater than a fourth preset degree, the business including the business characteristic is pushed to the client logged in the user account.

In order to achieve the above object, according to another aspect of the present application, there is also provided a data processing apparatus including: the first acquisition unit is configured to acquire a first historical behavior characteristic of a user account before the current moment; an input unit configured to perform input of the first historical behavior feature to a pre-trained survival probability model, acquire a first survival prediction probability of the user account output by the survival probability model for each prediction time period in a prediction period, and acquire a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of a sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before a first historical moment, the historical activity information comprises a time interval between the second historical moment and the first historical moment, and the second historical moment is a historical moment when the sample user account is firstly active within a preset time period after the first historical moment; a second obtaining unit, configured to perform obtaining of an active time at which the user account accesses a service for the first time after the current time, and modify the first survival curve by using the active time to obtain a second survival curve of the user account within the prediction period; the determining unit is configured to determine the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determine the degree and the direction of the influence of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain.

Optionally, the determining unit includes: a training module configured to perform training of a tree model with the first historical behavior feature, the business feature of the business as inputs, and the liveness gain as an output; and the determining module is configured to execute the interpretation trained tree model and determine the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business on the activity gain.

Optionally, the training module is configured to train the tree model with the dimensional feature of the first historical behavior and the business dimensional feature of the business as inputs and the liveness gain as an output, and the determining module is configured to perform interpretation of the trained tree model and determine the degree and direction of influence of the dimensional feature of the first historical behavior and/or the business dimensional feature of the business on the liveness gain.

Optionally, the determining module is configured to perform: inputting the trained tree model, the first historical behavior feature and the business feature of the business into a model interpreter, and obtaining the influence degree and the influence direction of the first historical behavior feature and/or the business feature of the business output by the model interpreter on the liveness gain.

Optionally, the second obtaining unit includes: a first obtaining module configured to perform obtaining a first prediction probability of survival for the user account for each prediction time period within a prediction period; a second obtaining module configured to perform obtaining a second survival prediction probability of the user account in the prediction time period of the active time; a revising module configured to perform revising each of the first survival prediction probabilities using the second survival prediction probability; a third obtaining module configured to obtain the second survival curve based on the corrected first survival prediction probability.

Optionally, the apparatus further comprises: a first pushing unit, configured to perform pushing of adding a service including the service feature to a user account corresponding to the first historical behavior feature when the influence direction of the first historical behavior feature and/or the service feature of the service on the liveness gain is a positive direction after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve and determining the influence degree and the influence direction of the first historical behavior feature and/or the service feature of the service on the liveness gain; the first pushing unit is further configured to perform: and under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a negative direction, reducing the pushing of the service comprising the service characteristic to a user account corresponding to the first historical behavior characteristic.

Optionally, the apparatus further comprises: a second pushing unit configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and a second survival curve, and determining a degree and a direction of influence of the first historical behavior feature and/or a service feature of the service on the activity gain, if the direction of influence of the service feature of the service on the activity gain is a positive direction, pushing the service including the service feature; the second pushing unit is further configured to perform: when the influence direction of the first historical behavior characteristic on the liveness gain is a positive direction, pushing a service to a client logged in the user account; the second pushing unit is further configured to perform: and when the first historical behavior characteristic and the business characteristic of the business influence the liveness gain in a positive direction, pushing the business including the business characteristic to a client logged in the user account.

Optionally, the apparatus further comprises: a third pushing unit configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and the second survival curve and determining a degree of influence and a direction of influence of the first historical behavior feature and/or the service feature of the service on the activity gain, pushing the service including the service feature when the degree of influence of the service feature of the service on the activity gain is greater than a first preset degree and the direction of influence is a positive direction; the third pushing unit is further configured to execute pushing of a service to a client logged in the user account when the influence degree of the first historical behavior characteristic on the liveness gain is greater than a second preset degree and the influence direction is a positive direction; the third pushing unit is further configured to execute pushing the service including the service characteristic to the client logged in the user account when the first historical behavior characteristic and the service characteristic of the service have a positive influence on the liveness gain, the influence degree of the service characteristic of the service on the liveness gain is greater than a first preset degree, and the influence degree of the first historical behavior characteristic on the liveness gain is greater than a second preset degree.

Optionally, the apparatus further comprises: a fourth pushing unit configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and the second survival curve and determining a degree and a direction of influence of the first historical behavior feature and/or the business feature of the business on the activity gain, reducing to push the business including the business feature when the direction of influence of the business feature of the business on the activity gain is a negative direction; the fourth pushing unit is further configured to reduce pushing of traffic to the client logged in the user account when the influence direction of the first historical behavior characteristic on the liveness gain is a negative direction; the fourth pushing unit is further configured to reduce pushing of the service including the service characteristic to the client logged in the user account when the first historical behavior characteristic and the service characteristic of the service have a negative influence direction on the liveness gain.

Optionally, the apparatus further comprises: a fifth pushing unit configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and the second survival curve and determining a degree and a direction of influence of the first historical behavior feature and/or the service feature of the service on the activity gain, reducing pushing of the service including the service feature when the degree of influence of the service feature of the service on the activity gain is greater than a third preset degree and the direction of influence is a negative direction; the fifth pushing unit is further configured to reduce pushing of a service to a client logged in the user account when the degree of influence of the first historical behavior characteristic on the liveness gain is greater than a fourth preset degree and the direction of influence is a negative direction; the fifth pushing unit is further configured to perform, when the first historical behavior feature and the business feature of the business have a negative influence on the liveness gain, the business feature of the business has a degree of influence on the liveness gain greater than a third preset degree, and the first historical behavior feature has a degree of influence on the liveness gain greater than a fourth preset degree, reducing pushing of the business including the business feature to the client logged in the user account.

In order to achieve the above object, according to another aspect of the present application, there is also provided a computer-readable storage medium including a stored program, wherein the data processing method described in the present application is performed when the program is executed by a processor.

In order to achieve the above object, according to another aspect of the present application, there is also provided an electronic apparatus including: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the data processing method described herein.

In order to achieve the above object, according to another aspect of the present application, there is also provided a computer program product comprising a computer program which, when executed by a processor, performs the data processing method described herein.

The method comprises the steps of obtaining a first historical behavior characteristic of a user account before the current moment; inputting a first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of a user account output by the survival probability model in each prediction time period in a prediction period, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; acquiring the active time of a user account for accessing a service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in a prediction period; based on the difference between the first survival curve and the second survival curve, determining the liveness gain of the user account, determining the influence degree and the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain, calculating the liveness gain through the difference between the predicted survival data and the actual survival data, and further determining the influence weight and the influence direction of the historical behavior characteristic and the service characteristic of the user on the liveness, so that the service content can be pushed according to the preferences of different users, the problem that the liveness of the user can not be improved by what content can not be obtained when the content is pushed to the user in the related technology is solved, and the effect of determining the influence of different service characteristics on the liveness of the user so as to effectively improve the liveness of the user is achieved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application. In the drawings:

FIG. 1 is a flow chart of a data processing method according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a survival probability curve of the present embodiment;

FIG. 3 is a flowchart of a method for identifying influences on user liveness according to the present embodiment;

FIG. 4 is a graph of the survival probability of the present embodiment;

FIG. 5a is a schematic diagram of the importance of each dimension calculated in the present embodiment;

FIG. 5b is another schematic diagram of the calculated directions of influence of the dimensions according to the present embodiment;

FIG. 6 is a schematic diagram of a data processing apparatus according to an embodiment of the present application;

fig. 7 is a schematic diagram of the apparatus of the present embodiment.

Detailed Description

It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.

In order to make the technical solutions better understood by those skilled in the art, 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 only partial 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.

It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the application described herein may be used. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

An embodiment of the present application provides a data processing method, and fig. 1 is a flowchart of the data processing method according to the embodiment of the present application, in a technical solution of the embodiment, an execution main body may be a background server, and in some mobile terminals with relatively strong computing power, the execution main body may also be a mobile terminal. As shown in fig. 1, the method includes the following steps S102-S108.

Step S102: a first historical behavior feature of a user account prior to a current time is obtained.

In one embodiment, the first historical behavior feature may be a behavior feature generated by a user client accessing a service during a historical period prior to the current time, the user client being logged into the user account. In one example, the access data of the user client to access the service includes a user account and a behavior characteristic.

The first historical behavior feature may be data of behaviors such as browsing, clicking, like, or may be a feature extracted based on the data of behaviors such as browsing, clicking, like.

Step S104: the method comprises the steps of inputting first historical behavior characteristics to a pre-trained survival probability model, obtaining first survival prediction probabilities of user accounts in each prediction time period in a prediction period, which are output by the survival probability model, and obtaining a first survival curve of the user accounts in the prediction period based on the first survival prediction probabilities. The survival probability model is obtained by training with second historical behavior characteristics of the sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before the first historical time, the historical activity information comprises a time interval between the second historical time and the first historical time, and the second historical time is the historical time when the sample user account is firstly active within a preset time period after the first historical time.

The survival probability model is obtained by training with actual data as a sample, specifically, training with the second historical behavior characteristic of the sample user account as input and the historical activity information of the sample user account as output, and because the historical activity information to be predicted already has real data, the model parameters can be adjusted according to the real data, so that the difference between the model prediction result and the real result is in a very small range. The second historical behavior feature may be data of behaviors such as browsing, clicking, like, and may further include attribute features of the user such as age, sex, and the like.

The prediction period may be N days after the current time, where N is a positive integer. In one embodiment, the prediction period may be divided into a plurality of consecutive time segments, and the divided time segments may be referred to as prediction time segments. For example, the prediction period is 7 days, data within 7 days after the current time is predicted by the survival probability model, and each prediction time period within the prediction period is … … th day 1, 2 nd day, and 3 rd day 7. In an alternative embodiment, a 7 day prediction period may be set and changed. And then drawing a survival curve for 7 continuous days according to the predicted daily survival probability, wherein the survival curve is the first survival curve.

In one embodiment, the first historical behavior feature is a behavior feature within a preset time period before the current time, the second historical behavior feature is a behavior feature within a preset time period before the first historical time, and the preset time period can be customized according to situations, for example, 2 days, 7 days, 14 days, and the like. The prediction period is N days after the current time, N is a positive integer, and N can be customized according to the situation, such as 3 days, 7 days, 14 days and the like.

For example: assume that the current time is 10 months, 12 days 00 in 2020: 00, the preset time period is 7 days, the prediction period is 7 days, and then the first historical time is the time of 7 days before the current time, namely, 10 in 2020, 10 in month, 5 in day 00: 00, first historical behavior characterized by 00 at 10 months and 12 days of 2020: behavior characteristics 7 days before 00, second historical behavior characteristics 10 months 5 days 00 in 2020: behavioral characteristics 7 days before 00. Assume again that the second historical time at which the sample user account was first active within 7 days after the first historical time is 10 months, 8 days 00 in 2020: 00, then the historical activity information includes 3 days (i.e., the time interval between the second historical time and the first historical time).

And training a survival probability model by using the second historical behavior characteristic within 7 days before the first historical moment as input and using 3 days as output, wherein the survival probability model is trained well when the survival probability model is converged. And then, using the first historical behavior characteristic within 7 days before the current time as an input, predicting the survival probability (namely, the first survival prediction probability) of the user account in each day (prediction time period) within 7 days (prediction period) after the current time by using the trained survival probability model, wherein the predicted survival probability on the first day is P1, the survival probability on the second day is P2, the survival probability on the third day is P3. And drawing a first survival curve predicted within 7 days in the future after the current time by taking the survival probabilities P1, P2, P3, and P7 as vertical coordinates and the prediction time period as horizontal coordinates.

Step S106: and acquiring the active time of the user account for accessing the service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in the prediction period.

In one embodiment, the service may be a video, the time when the user client accesses the service for the first time after the current time is an active time, and the access data of the user client accessing the service includes a user account and a behavior characteristic. In one example, after the user client logged in to the user account accesses the service at the current time or a historical time before the current time, the active time at which the user client logged in to the user account accesses the service for the first time after the current time may be obtained.

For example, the current time is 12 noon 10 months 12 months 2020: 00, assume that the user viewed video 1 in the business at this point by the user client logging into the user account, and at 10/13/11/am in 2020: 00 again accessed this service through the user client logging into the user account, the active time would be 10 months in 2020, 13 am 11: 00.

in one embodiment, the survival prediction probability is modified by using the actual survival data (i.e., the active time) of the user account, and an actual survival curve (i.e., a second survival curve) of the user account is drawn according to the modified survival probability.

Step S108: and determining the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the influence degree and the influence direction of the first historical behavior characteristics and/or the service characteristics of the service on the activity gain.

The first survival curve is corrected through the actual activity data of the user to obtain a second survival curve, the difference of the two survival curves represents the activity gain of the user, and the influence degree and the influence direction of the first historical behavior characteristic and the service characteristic of the service on the activity gain can be reflected. Therefore, the service content which is more likely to increase the user activity can be pushed to the user during the subsequent service pushing so as to meet different preferences of the user.

The embodiment obtains a first historical behavior characteristic of a user account before the current moment; inputting a first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of a user account output by the survival probability model in each prediction time period in a prediction period, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; acquiring the active time of a user account for accessing a service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in a prediction period; determining an activity gain for the user account based on a difference between the first survival curve and the second survival curve, and determining the degree and direction of influence of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain, calculating an activity gain by a difference between the predicted survival data (a first survival curve drawn based on the first survival prediction probability) and the actual survival data (a second survival curve), further determining the influence weight and the influence direction of the historical behavior characteristics and the business characteristics of the user on the activeness, therefore, the service content can be pushed according to the favor of different users, the problem that the user can not know which content can improve the activity of the user when pushing the content to the user in the related technology is solved, and further, the effect of determining the influence of different service characteristics on the user activity is achieved so as to effectively improve the user activity.

In some embodiments, determining the degree and direction of the influence of the first historical behavior characteristic and/or the traffic characteristic of the traffic on the liveness gain includes: training a tree model by taking the first historical behavior characteristic and the service characteristic of the service as input and taking the activity gain as output; and explaining the trained tree model, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business on the activity gain.

When determining the influence of the liveness and the gain of the user account, the influence degree and the influence direction of various factors on the liveness can be calculated through the tree model. In an embodiment, a tree model (e.g., a decision tree model, a regression tree model, Xgboost, Lightgbm, etc.) may be used as an initial model, historical data of an application scenario of the present application is used as a sample to train and obtain the tree model meeting requirements of the application scenario, and by interpreting the trained tree model, a degree and a direction of influence of a first historical behavior feature and/or a business feature of a business on an activity gain may be determined, where the direction of influence is divided into a positive direction and a negative direction, where the positive direction is used for indicating to increase activity, and the negative direction is used for indicating to decrease activity.

In some embodiments, the manner of determining the degree and direction of influence of the first historical behavior feature and/or the traffic feature on the liveness gain is: and explaining the trained tree model, and determining the influence degree and the influence direction of the dimension characteristic of the first historical behavior and/or the service dimension characteristic of the service on the activity gain.

It should be noted that the feature in this embodiment may be embodied in the form of a dimensional feature, the service may be a video, and multiple feature dimensions may be set for the video, for example, 64-dimensional features are set for the video. The historical behavior features and the like are also represented in the form of dimensions, for example, each user is also provided with 64-dimensional features, so that the behavior features of the users are represented in the form of 64-dimensional features. The more dimensionality, the more refined the conclusion is. In this embodiment, the dimensional characteristic of the user behavior refers to user behavior Embedding, and the dimensional characteristic of the video refers to video Embedding.

In this embodiment, the Embedding may be based on a user's behavior of clicking on the video, and a DNN model is used to train the user behavior and the Embedding of the video, and this embodiment may also adopt other ways to extract the user behavior Embedding and the video Embedding, specifically which way, which way is not limited in this application.

In one embodiment, the trained tree model, the first historical behavior feature and the business feature of the business are input into the model interpreter, and the influence degree and the influence direction of the first historical behavior feature and/or the business feature of the business output by the model interpreter on the liveness gain are obtained. The causality inference based on the tree model has higher interpretability, and the model result can be better explained and landed.

The model interpreter in this embodiment may be a SHAP module, and importance (influence degree) and influence direction of each dimension feature on activity gain may be calculated by treeexplanier and shape _ values methods of the SHAP module.

In one embodiment, modifying the first survival curve with the activity time to obtain a second survival curve for the user account over the prediction period comprises: acquiring a first survival prediction probability of each prediction time period of a user account in a prediction period; acquiring the actual activity of the user account in the prediction time period of the active moment; correcting each first survival prediction probability by using the actual liveness; and obtaining a second survival curve based on the corrected first survival prediction probability.

In one example, based on a trained survival probability model, a first daily survival prediction probability for n days in the future of the user account may be predicted, P1, P2, P3, a. In this example, the user account is active on the mth day, the first survival prediction probability that the user account is active on the mth day is predicted to be Pm based on the trained generation probability model, and the ratios of the first survival prediction probabilities to the first survival probabilities (i.e., the second survival prediction probabilities) at the time when the user account is actually active are obtained as follows: and P1/Pm, P2/Pm, P3/Pm, Pm/Pm 1, P, Pn/Pm, and the obtained ratios are used as survival probability values after the first survival prediction probability is corrected. Since the maximum survival probability value does not exceed 1, the modified probability values are subjected to threshold pruning to obtain modified survival probability values of 1, 1. The corrected survival probability value uses the actual activity data of the user, ensures that the difference between the survival probability value and the first survival prediction probability can describe the change of the activity of the user account, simultaneously still keeps the trend of the survival probability of the user account, and reduces the influence of the noise of the activity of the user on the change of the activity.

In some of the above embodiments, after step S108, the method further includes: under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a positive direction, pushing of the service comprising the service characteristic is added to a user account corresponding to the first historical behavior characteristic; and under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain is a negative direction, reducing the pushing of the service comprising the service characteristic to the user account corresponding to the first historical behavior characteristic.

After determining the influence of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain, if the influence direction is a positive direction, pushing of service content with an effect of improving the liveness can be added to improve the liveness of the user; if the direction of influence is negative, pushing of the business content which has a decreasing effect on the liveness is to be avoided or reduced. In one embodiment, the influence directions of different service features on the liveness may be opposite, for example, the first dimension feature has a positive influence on the liveness, the third dimension feature has a negative influence on the liveness, and if the first dimension feature has a high influence on the liveness of the video, the third dimension feature has a high influence on the liveness, that is: for the video, the larger the first dimension characteristic is, the higher the improvement degree of the liveness is; the larger the third dimension characteristic is, the higher the reduction degree of the activity is. In this case, it should be minimized to push the services with high third dimension characteristics, or to increase the services with high first dimension characteristics, so as to improve the user activity to the greatest extent.

The method and the device for processing the video activity can output the combined influence of the video Embedding and the user behavior Embedding on the activity gain. For example, a video group whose one-dimensional characteristic of video Embedding is greater than a value (e.g., 0.2) has the most positive effect on the increase of the activity of the user group whose one-dimensional characteristic of user behavior Embedding is greater than a value (e.g., 0.1), i.e., the greater the video Embedding greater than 0.2, the greater the increase of the activity of the user group whose user behavior Embedding is greater than 0.1. According to the method, the influence of different combinations of the video dimensional characteristics and the user dimensional characteristics on the user liveness can be calculated, the combination of the video dimensional characteristics with the largest forward gain and the user dimensional characteristics is selected, and the user liveness of which users can be improved by which service contents can be obtained, so that the video with the video dimensional characteristics is pushed to the users with the user dimensional characteristics, the effect of pushing personalized high-quality videos to the users is achieved, and the user liveness is further improved to the maximum degree.

The embodiment of the application can know which service contents can improve the user activity, can also know which user activity can be improved, and can also know which service contents can improve the user activity of which user.

Since the user usually logs in the web page or the client through the account, when pushing the service, the service is pushed to the client or the terminal logging in the account of the user, and the pushed service can be displayed on the page, and the specific form of the service is not limited in the application.

In one embodiment, the method and the device can acquire which service contents can improve the user activity, and push the service including the service characteristics if the influence direction of the service characteristics of the service on the activity gain is a positive direction. In this embodiment, if the direction of the influence of the service characteristics of the service on the liveness gain is a positive direction, it indicates that the service has a positive pushing effect on the liveness of the user account, and at this time, the service may be pushed to all the user accounts.

In one embodiment, the method and the device can acquire the user liveness of which users are promoted, and push the service to the client logged in the user account if the influence direction of the first historical behavior characteristics on the liveness gain is a positive direction. In this embodiment, if the direction of the influence of the first historical behavior feature on the liveness gain is a positive direction, it indicates that the liveness of the user account is not influenced by the service, and at this time, the service may be pushed to the client logged in the user account according to a pushing policy of the service.

In another embodiment, the method can acquire which service contents can improve the user activity of which users, and if the first historical behavior characteristics and the service characteristics of the service have positive influence on the activity gain, the service including the service characteristics is pushed to the client logged in the user account. In this embodiment, if the direction of the influence of the first historical behavior feature and the service feature of the service on the liveness gain is a positive direction, it indicates that the service plays a role in promoting the liveness of the user account corresponding to the first historical behavior feature in a positive direction, and at this time, the service can be pushed to the client logged in the user account. In the embodiment of the present application, the user account corresponding to the behavior feature indicates that the behavior feature is the behavior feature of the user. For example, if the behavior feature 1 is the behavior feature of the user account 1, the user account corresponding to the behavior feature 1 is the user account 1.

In some embodiments, the method and the device can acquire which service contents can improve the user activity, and push the service including the service characteristics if the influence degree of the service characteristics on the activity gain is greater than the first preset degree and the influence direction is a positive direction. The first preset degree is set according to actual self-definition, and the embodiment of the application is not particularly limited.

In some embodiments, the method and the device can acquire the user liveness of which users are promoted, and push the service to the client logging in the user account if the influence degree of the first historical behavior characteristic on the liveness gain is greater than the second preset degree and the influence direction is a positive direction. The second preset degree is set according to actual self-definition, and the embodiment of the application is not particularly limited.

In some embodiments, the method can acquire which service contents can promote the user liveness of which users, and push the service including the service characteristics to the client logging in the user account if the first historical behavior characteristics and the service characteristics of the service have positive influence on the liveness gain, the influence degree of the service characteristics of the service on the liveness gain is greater than a first preset degree, and the influence degree of the first historical behavior characteristics on the liveness gain is greater than a second preset degree.

In some other embodiments, if the first historical behavior characteristic and/or the traffic characteristic of the traffic have a slight influence on the liveness gain, the platform push strategy may not be changed. The pushing strategy is executed under the condition that the influence is positive direction and the influence degree of the liveness gain is greater than the preset degree, the first preset degree and the second preset degree can be the same numerical value or different numerical values, and the pushing strategy can be specifically set according to the situation in a self-defined mode.

In other embodiments, the pushing of the service including the service feature is reduced if the direction of the impact of the service feature of the service on the activity gain is a negative direction. The scheme can know which service contents reduce the user activity.

And if the influence direction of the first historical behavior characteristics on the liveness gain is a negative direction, reducing the business push to the client of the login user account. The scheme can know which users have reduced user activity.

And if the influence direction of the first historical behavior characteristic and the service characteristic of the service on the liveness gain is a negative direction, reducing the pushing of the service comprising the service characteristic to the client side logging in the user account. The scheme can know which service contents reduce the user activity of which users.

For example, for a certain APP, if it is determined that the influence direction of the video 2 on the liveness gain of the user account is a negative direction, pushing the service to all user accounts registered in the APP can be reduced.

In one embodiment, if the influence degree of the service characteristics of the service on the activity gain is greater than the third preset degree and the influence direction is a negative direction, the service including the service characteristics is pushed less. The scheme can know which service contents reduce the user activity.

And if the influence degree of the first historical behavior characteristic on the liveness gain is greater than the fourth preset degree and the influence direction is a negative direction, reducing the business push to the client of the login user account. The scheme can know which users have reduced user activity.

If the influence direction of the first historical behavior characteristic and the business characteristic of the business on the liveness gain is a negative direction, the influence degree of the business characteristic of the business on the liveness gain is greater than a third preset degree, and the influence degree of the first historical behavior characteristic on the liveness gain is greater than a fourth preset degree, the business including the business characteristic is pushed to the client side of the login user account. The scheme can know which service contents reduce the user activity of which users. The third preset degree and the fourth preset degree are set according to a user-defined setting, and the embodiment is not limited.

For the case that the influence direction is the negative direction, the purpose of the push strategy is to reduce the service push with the influence direction being the negative direction, so as to avoid influencing the user activity.

Based on the scheme provided by the embodiment, the difference of the videos which are stored by the user and promoted in different dimensions can be selected according to the user accounts with different characteristics, and accordingly, the high-quality videos which meet the personalized preferences of the user are screened. This embodiment is through the historical behavior characteristic of the user of input, watch the video after several days the active condition and watch the video Embedding characteristic, calculate user's liveness change according to the difference in survival curve area, fig. 2 is the schematic diagram of the survival probability curve of this embodiment, curve 2 has adopted in two curves that the user's liveness is higher after the push strategy of this application embodiment, and give video each dimension Embedding to influence weight and influence direction of different characteristic user liveness, help short video APP analyst to look for the business that promotes the user and remain.

Referring to fig. 3, the technical solution of the present embodiment can be used to identify factors affecting the activity of the user account, and the solution is: measuring the activity gain of the user account based on the difference of the areas of the survival probability curves, then taking the activity gain as a Label, a user behavior Embedding and a watching video Embedding as Features input tree models (such as Xgboost and Lightgbm), and combining with a SHAP VALUE algorithm, outputting the video Embedding of each dimension and the influence weight and the influence direction of the Embedding of each user dimension on the activity of the user account.

Fig. 3 is a flowchart of the method for identifying an influence on the activity of a user account according to the present embodiment, and as shown in fig. 3, the method mainly includes the following four processes (r), (c), and (d).

Assume that the current time is 10 months, 12 days 00 in 2020: 00, the preset time period is 7 days, the prediction period is 7 days, and then the first historical time is the time of 7 days before the current time, namely, 10 in 2020, 10 in month, 5 in day 00: 00, first historical behavior characterized by 00 at 10 months and 12 days of 2020: behavior characteristics 7 days before 00, second historical behavior characteristics 10 months 5 days 00 in 2020: behavioral characteristics 7 days before 00. Assume again that the second historical time at which the sample user account was first active within 7 days after the first historical time is 10 months, 8 days 00 in 2020: 00, then the historical activity information includes 3 days (i.e., the time interval between the second historical time and the first historical time).

Firstly, a survival probability model such as Cox is trained based on historical behavior characteristics and historical activity information of a user account (M1), and then a survival probability curve of a target user and an area enclosed by the curve and the horizontal and vertical axes are predicted by using M1 (S1).

The specific process is as follows.

1. Training a model: the data of the user account is sampled randomly, and the data format is selected as table 1.

TABLE 1

The fields are explained as follows.

User _ id is a unique identification of a User, such as a User account.

Event indicates Event occurrence, here indicating whether the user account is active, where 1 indicates Event occurrence and 0 indicates Event non-occurrence. In the example of Table 1, Event is 1, indicating that the user account is again active at the second historical time.

Duration represents the interval Duration of the event, here the interval of the Duration of the user being active again after watching the video, i.e. the time interval between the second historical time and the first historical time. The interval is limited in the example to be only counted up to 7 days at maximum, and can be adjusted according to specific scenes. Here, Event and Duration are historical activity information. When Event is 1, Duration is the time interval between the second historical time and the first historical time, which is 3 days in the example of table 1; when Event is 0, the Duration field is 0.

Feature represents behavior characteristics such as user attributes, user figures, and user behaviors before the first history time.

After the data in table 1 are acquired, for each piece of data, Feature is used as input, Duration is used as output, a Cox model is trained, and when each parameter in the Cox model is converged, Cox is trained well. The trained Cox model may be referred to as the M1 model.

2. Model prediction: using the survival probability model M1 trained in the previous step, and reading in (the data of) the target user account (i.e. the first historical behavior feature), no Duration (model label) data is needed, in the form as shown in table 2.

TABLE 2

The fields in table 2 are explained as follows.

User _ id is a unique identification of a User, such as a User account.

An Event indicates the occurrence of an Event, here indicating whether the user is active.

Feature represents behavior characteristics such as user attributes, user figures, and user behaviors before the current time.

3. And (4) outputting a prediction result: after the playing video is output in the previous step, the survival probability of each user account for each day in the next 7 days can be plotted according to the first survival curve, such as a dotted line (which may be referred to as a prior survival probability curve) in fig. 4, and a survival curve area S1 enclosed by the first survival curve and the horizontal and vertical axes is calculated.

And secondly, correcting the first survival curve based on the actual activity data of the user to obtain a corrected second survival curve (a corrected survival probability curve in the graph 4), and calculating the survival curve area enclosed by the second survival curve and the horizontal and vertical axes (S2).

The specific process is as follows.

1. And calculating the date interval (date _ diff) from the video playing date to the first time when the video is played by the user client who logs in the user account. Assuming that a user client (e.g. video APP) logging in a user account logs in the user client again on the 3 rd day after playing a video, the day interval in which the user account is active for the first time is considered to be 3 days.

2. The first survival curve (prior survival probability curve) predicted by the model is corrected, and a survival curve area S2 enclosed by the second survival curve and the abscissa and ordinate axes is calculated. The correction process is specifically as follows.

Based on the trained survival probability model M1, the first daily survival prediction probabilities of the user account in the 7 future days are predicted to be P1, P2, P3, the. The user account is active on day 3, the first survival curve is modified using the user's first probability of survival prediction on day 3P 3 (i.e., the second probability of survival prediction). The ratio of each first survival prediction probability to the second survival prediction probability is obtained as follows: and P1/P3, P2/P3, P3/P3 are 1, P7/P3, and the obtained ratios are used as survival probability values after the first survival prediction probability is corrected. Since the maximum survival probability value does not exceed 1, the probability values after the correction are subjected to threshold pruning, and the probability values after the correction are respectively 1, 1.

Thirdly, measuring the activity gain of the user after watching the video (namely the user client who logs in the user account plays the video) by using the area difference between S2 and S1, and calculating the influence of Embedding of each dimension of the video on the gain by using a tree model (Xgboost, Lightgbm).

The specific process is as follows.

1. For each user account, an area difference is calculated between the survival curve area of the second survival curve S2 and the curve area of the first survival curve S1 for the user account as an activity gain for the user account.

2. For each user account, inputting the Embedding characteristics of the played video and the Embedding characteristics of the user behaviors as tree models Features, and using the activity gain as the label of the tree models to obtain the trained models (M2). The data format used for training the model is shown in table 3.

TABLE 3

The fields in table 3 are explained as follows.

User _ id is a unique identification of a User, such as a User account.

Video Embedding is a representation that abstracts video into multi-dimensional feature vectors.

User behavior Embedding is a representation that abstracts user behavior features into multidimensional feature vectors.

The activity gain is the difference between S2 and S1.

And fourthly, calculating the influence weight and the influence direction of each Embedding of the video on the user activity of the Embedding of different user behaviors through the SHAP module.

The specific process is as follows.

1. The model M2 and the user behavior Embedding and video Embedding characteristics after label removal are input into the SHAP module.

2. The importance and the influence direction of each Embedding are calculated by TreeExplainer and shape _ values methods of the SHAP module.

The above is the main process of identifying video Embedding and/or user behavior Embedding which affect the user liveness by using the scheme provided by the embodiment of the application. Fig. 5a and 5b are schematic diagrams of importance and influence direction of each calculated dimension according to this embodiment, and videos for increasing the liveness of different user groups can be recommended according to the result. In fig. 5a, the abscissa is the influence degree of each dimension feature on the activity gain, the larger the value is, the more obvious the activity gain influence on the user account is illustrated, the ordinate is the Embedding feature of each dimension of the user behavior and the service, the size of the scatter point represents the size of the corresponding feature value, the larger the point is, the larger the feature value is illustrated, and as can be seen from fig. 5a, the emb27 (i.e., the video Embedding27) is in positive correlation with the activity gain of the user account. In fig. 5b, the abscissa is the influence direction of the video Embedding39 on the liveness gain, and the ordinate is the influence degree of the video Embedding39 on the liveness gain of the user account, such as: the value of the video vector of the emb39 (namely, Embedding39) is obviously and negatively related to the activity gain of the user account, and videos with larger emb39 values can be reduced or filtered in video recommendation and screening.

The embodiment is used for searching what service (such as video) can better improve the liveness of which user accounts, and has the following beneficial effects:

reflecting the influence of the services (such as videos) with different characteristics on users with different characteristics, introducing service (such as videos) and Embedding of user behaviors as model characteristics, determining the service Embedding liked by the user group of each user behavior Embedding, determining the influence of different service characteristics and/or behavior characteristics on the user activity, and achieving the purpose of personalized service screening for the users. The embodiment of the application can also screen high-quality services (such as videos) in batches and reduce resource consumption, for example, abstract the service characteristics and the behavior characteristics into the vector Embedding, can search high-quality contents for users with the same behavior characteristics in batches, reduce the screening cost of the services (such as videos), and simultaneously, do not need to calculate the service characteristics and the user behavior characteristics with higher dimensionality and reduce a large amount of computing resources. The method and the device for improving the user activity can meet the core appeal of the short video industry for improving the user activity, and ensure that the theory of video screening is matched with the requirement of the user on the video from the viewpoint of improving the user activity, so that the method and the device have a larger practical application space; in addition, the activity gain is measured through the survival curve area, discrete data are converted into continuous data, the data discrimination is increased, the sensitivity is higher, and the influence of services (such as videos) on users can be captured more sensitively.

In conclusion, the embodiment can be widely applied to the internet field (for example, the video field), helps each APP to screen the service Embedding for improving the user activity in batch, and has a wide application prospect.

It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowcharts, in some cases, the steps illustrated or described may be performed in an order different than presented herein.

The embodiment of the application provides a data processing device, and the device can be used for executing the data processing method of the embodiment of the application.

Fig. 6 is a schematic diagram of a data processing apparatus according to an embodiment of the present application, as shown in fig. 6, the apparatus including:

a first obtaining unit 10 configured to perform obtaining a first historical behavior characteristic of a user account before a current time;

an input unit 20 configured to perform input of a first historical behavior feature to a pre-trained survival probability model, acquire a first survival prediction probability of a user account output by the survival probability model for each prediction time period in a prediction period, and acquire a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of the sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before the first historical time, the historical activity information comprises a time interval between the second historical time and the first historical time, and the second historical time is the historical time at which the sample user account is firstly active within a preset time period after the first historical time;

a second obtaining unit 30, configured to perform obtaining an active time at which the user account accesses the service for the first time after the current time, and modify the first survival curve by using the active time to obtain a second survival curve of the user account within the prediction period;

the determining unit 40 is configured to determine the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determine the degree and the direction of the influence of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain.

The embodiment adopts a first obtaining unit 10 configured to perform obtaining a first historical behavior characteristic of a user account before a current time; an input unit 20 configured to perform input of a first historical behavior feature to a pre-trained survival probability model, acquire a first survival prediction probability of a user account output by the survival probability model for each prediction time period in a prediction period, and acquire a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of the sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before the first historical time, the historical activity information comprises a time interval between the second historical time and the first historical time, and the second historical time is the historical time at which the sample user account is firstly active within a preset time period after the first historical time; a second obtaining unit 30, configured to perform obtaining an active time at which the user account accesses the service for the first time after the current time, and modify the first survival curve by using the active time to obtain a second survival curve of the user account within the prediction period; the determining unit 40 is configured to determine the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determine the degree and direction of the influence of the first historical behavior characteristics and/or the service characteristics of the service on the liveness gain, so that the problem that in the related art, when the content is pushed to the user, which content can improve the liveness of the user cannot be known is solved, and the effect of determining the influence of different service characteristics on the liveness of the user to effectively improve the liveness of the user is achieved.

Optionally, the determining unit 40 includes: the training module is configured to execute training of a tree model by taking the first historical behavior characteristic and the service characteristic of the service as input and taking the activity gain as output; and the determining module is configured to execute the interpreted and trained tree model and determine the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business on the activity gain.

Optionally, the training module is configured to perform training of the tree model with the dimension characteristic of the first historical behavior and the service dimension characteristic of the service as inputs and with the liveness gain as an output, and the determination module is configured to perform interpretation of the trained tree model and determine the degree and direction of influence of the dimension characteristic of the first historical behavior and/or the service dimension characteristic of the service on the liveness gain.

Optionally, the determining module is configured to perform: inputting the trained tree model, the first historical behavior characteristic and the business characteristic of the business into the model interpreter, and obtaining the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business output by the model interpreter on the activity gain.

Optionally, the second obtaining unit includes: a first obtaining module configured to obtain a first survival prediction probability of each prediction time period in the prediction period of the user account; the second obtaining module is configured to obtain a second survival prediction probability of the prediction time period to which the user account belongs at the active moment; a revising module configured to perform revising each first survival prediction probability using the second survival prediction probability; and the third obtaining module is configured to obtain a second survival curve based on the corrected first survival prediction probability.

Optionally, the apparatus further comprises: the first pushing unit is configured to execute pushing of the service including the service characteristics to the user account corresponding to the first historical behavior characteristics under the condition that the influence direction of the first historical behavior characteristics and/or the service characteristics of the service on the liveness gain is a positive direction after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve and determining the influence degree and the influence direction of the first historical behavior characteristics and/or the service characteristics of the service on the liveness gain; the first pushing unit is further configured to perform: and under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain is a negative direction, reducing the pushing of the service comprising the service characteristic to the user account corresponding to the first historical behavior characteristic.

Optionally, the apparatus further comprises: a second pushing unit configured to perform, after determining the liveness gain of the user account based on a difference between the first survival curve and the second survival curve, and determining a degree and a direction of influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, if the direction of influence of the service feature of the service on the liveness gain is a positive direction, pushing the service including the service feature; the second pushing unit is further configured to perform: when the influence direction of the first historical behavior characteristics on the liveness gain is a positive direction, pushing a service to a client of a login user account; the second pushing unit is further configured to perform: and when the first historical behavior characteristic and the business characteristic of the business influence the liveness gain in a positive direction, pushing the business including the business characteristic to the client logged in the user account.

Optionally, the apparatus further comprises: a third pushing unit configured to perform, after determining the liveness gain of the user account based on a difference between the first survival curve and the second survival curve and determining the degree and the direction of influence of the first historical behavior feature and/or the service feature of the service on the liveness gain, pushing the service including the service feature when the degree of influence of the service feature of the service on the liveness gain is greater than the first preset degree and the direction of influence is a positive direction; the third pushing unit is also configured to execute pushing business to the client end of the login user account when the influence degree of the first historical behavior characteristic on the liveness gain is greater than the second preset degree and the influence direction is a positive direction; the third pushing unit is further configured to execute pushing the service including the service characteristics to the client logged in the user account when the first historical behavior characteristics and the service characteristics of the service have a positive influence on the liveness gain, the influence degree of the service characteristics of the service on the liveness gain is greater than the first preset degree, and the influence degree of the first historical behavior characteristics on the liveness gain is greater than the second preset degree.

Optionally, the apparatus further comprises: a fourth pushing unit configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and the second survival curve, and determining a degree and a direction of influence of the first historical behavior feature and/or the business feature of the business on the activity gain, when the direction of influence of the business feature of the business on the activity gain is a negative direction, reducing pushing of the business including the business feature; the fourth pushing unit is also configured to reduce pushing of the service to the client logging in the user account when the influence direction of the first historical behavior characteristic on the liveness gain is a negative direction; the fourth pushing unit is further configured to reduce pushing of the service including the service characteristics to the client logged in to the user account when the first historical behavior characteristics and the service characteristics of the service have a negative influence on the liveness gain.

Optionally, the apparatus further comprises: a fifth pushing unit, configured to perform, after determining an activity gain of the user account based on a difference between the first survival curve and the second survival curve, and determining an influence degree and an influence direction of the first historical behavior feature and/or the service feature of the service on the activity gain, when the influence degree of the service feature of the service on the activity gain is greater than a third preset degree and the influence direction is a negative direction, reducing to push the service including the service feature; the fifth pushing unit is further configured to reduce pushing of the service to the client logging in the user account when the influence degree of the first historical behavior characteristic on the liveness gain is greater than a fourth preset degree and the influence direction is a negative direction; the fifth pushing unit is further configured to reduce pushing of the service including the service feature to the client logged in the user account when the first historical behavior feature and the service feature of the service have a negative direction of influence on the liveness gain, the degree of influence of the service feature of the service on the liveness gain is greater than a third preset degree, and the degree of influence of the first historical behavior feature on the liveness gain is greater than a fourth preset degree.

The data processing device comprises a processor and a memory, wherein the first acquisition unit, the input unit, the correction unit and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions.

The processor comprises a kernel, and the kernel calls the corresponding program unit from the memory. The kernel can be set to be one or more than one, and the user activity is effectively improved by adjusting the kernel parameters.

The memory may include volatile memory in a computer readable medium, Random Access Memory (RAM) and/or nonvolatile memory such as Read Only Memory (ROM) or flash memory (flash RAM), and the memory includes at least one memory chip.

An embodiment of the present application provides a storage medium on which a program is stored, the program implementing the data processing method when executed by a processor.

The embodiment of the application provides a processor, wherein the processor is used for running a program, and the data processing method is executed when the program runs.

An apparatus is provided in an embodiment of the present application, and fig. 7 is a schematic diagram of the apparatus in this embodiment, as shown in fig. 7, the apparatus includes at least one processor, and at least one memory and a bus connected to the processor; the processor and the memory complete mutual communication through a bus; the processor is used for calling the program instructions in the memory so as to execute the data processing method. The device herein may be a server, a PC, a PAD, a mobile phone, etc.

The present application further provides a computer program product adapted to perform a program for initializing the following method steps when executed on a data processing device: acquiring a first historical behavior characteristic of a user account before the current moment; inputting a first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of a user account output by the survival probability model in each prediction time period in a prediction period, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; acquiring the active time of a user account for accessing a service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in a prediction period; and determining the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the influence degree and the influence direction of the first historical behavior characteristics and/or the service characteristics of the service on the activity gain.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that 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 phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.

The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A data processing method, comprising:

acquiring a first historical behavior characteristic of a user account before the current moment;

inputting the first historical behavior characteristic to a pre-trained survival probability model, acquiring a first survival prediction probability of the user account in each prediction time period in a prediction period, which is output by the survival probability model, and acquiring a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of a sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before a first historical moment, the historical activity information comprises a time interval between the second historical moment and the first historical moment, and the second historical moment is a historical moment when the sample user account is firstly active within a preset time period after the first historical moment;

acquiring the active time of the user account for accessing the service for the first time after the current time, and correcting the first survival curve by using the active time to obtain a second survival curve of the user account in the prediction period;

and determining the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain.

2. The method of claim 1, wherein the determining the degree and direction of the impact of the first historical behavior characteristic and/or the traffic characteristic of the traffic on the liveness gain comprises:

training a tree model by taking the first historical behavior characteristic and the service characteristic of the service as input and the liveness gain as output;

and explaining the trained tree model, and determining the influence degree and the influence direction of the first historical behavior characteristic and/or the business characteristic of the business on the activity gain.

3. The method of claim 2,

taking the first historical behavior characteristic and the service characteristic of the service as input, and taking the activity gain as output, the training tree model comprises: training the tree model by taking the dimension characteristic of the first historical behavior and the service dimension characteristic of the service as input and the activity gain as output,

interpreting the trained tree model, and determining the degree and direction of influence of the first historical behavior feature and/or the business feature of the business on the liveness gain comprises: and explaining the trained tree model, and determining the influence degree and the influence direction of the dimensional characteristics of the first historical behaviors and/or the service dimensional characteristics of the service on the activity gain.

4. The method of claim 2, wherein interpreting the trained tree model to determine the degree and direction of influence of the first historical behavior feature and/or the traffic feature of the traffic on the liveness gain comprises:

inputting the trained tree model, the first historical behavior feature and the business feature of the business into a model interpreter, and obtaining the influence degree and the influence direction of the first historical behavior feature and/or the business feature of the business output by the model interpreter on the liveness gain.

5. The method of claim 1, wherein said modifying said first survival curve with said active time to obtain a second survival curve for said user account over said prediction period comprises:

acquiring a first survival prediction probability of the user account in each prediction time period in a prediction period;

acquiring a second survival prediction probability of the user account in the prediction time period of the active moment;

correcting each of the first survival prediction probabilities using the second survival prediction probability;

and obtaining the second survival curve based on the corrected first survival prediction probability.

6. The method of claim 1, wherein after determining the liveness gain of the user account based on the difference between the first survival curve and the second survival curve, and determining the degree and direction of the impact of the first historical behavior feature and/or the traffic feature of the traffic on the liveness gain, the method further comprises:

under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a positive direction, adding pushing of the service comprising the service characteristic to a user account corresponding to the first historical behavior characteristic;

and under the condition that the influence direction of the first historical behavior characteristic and/or the service characteristic of the service on the liveness gain is a negative direction, reducing the pushing of the service comprising the service characteristic to a user account corresponding to the first historical behavior characteristic.

7. A data processing apparatus, comprising:

the first acquisition unit is configured to acquire a first historical behavior characteristic of a user account before the current moment;

an input unit configured to perform input of the first historical behavior feature to a pre-trained survival probability model, acquire a first survival prediction probability of the user account output by the survival probability model for each prediction time period in a prediction period, and acquire a first survival curve of the user account in the prediction period based on the first survival prediction probability; the survival probability model is obtained by training with second historical behavior characteristics of a sample user account as input and historical activity information of the sample user account as output, the second historical behavior characteristics are historical behavior characteristics of the sample user account before a first historical moment, the historical activity information comprises a time interval between the second historical moment and the first historical moment, and the second historical moment is a historical moment when the sample user account is firstly active within a preset time period after the first historical moment;

a second obtaining unit, configured to perform obtaining of an active time at which the user account accesses a service for the first time after the current time, and modify the first survival curve by using the active time to obtain a second survival curve of the user account within the prediction period;

the determining unit is configured to determine the activity gain of the user account based on the difference between the first survival curve and the second survival curve, and determine the degree and the direction of the influence of the first historical behavior characteristic and/or the service characteristic of the service on the activity gain.

8. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the data processing method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored program, wherein the data processing method of any one of claims 1 to 6 is performed when the program is executed by a processor.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, performs the data processing method of any one of claims 1 to 6.