KR20190139008A - A system for preventing user leaving based on expectation profit and controlling method thereof - Google Patents

Abstract

According to an embodiment, disclosed are a system for preventing leaving of users based on expected profits and a control method therefor. The system comprises: a prediction module determining a leaving prediction model that predicts some users expected to leave among all users by using a game log and machine learning; a target marketing module dividing the predicted users expected to leave into a control group and an experimental group, and performing target marketing on users belonging to the experimental group by a user-specific differential compensation method; and an expected profit optimization module comparing the leaving rate of users belonging to the control group and the experimental group to determine marketing costs to maximize expected profits, and determining optimization parameters based on the expected profits. Thus, the system can optimize the expected profits by differentially providing compensation according to the expected profits obtainable from the users who are to be prevented from being left.

Description

A system for preventing user leaving based on expectation profit and controlling method

Hereinafter, a departure prevention system and a control method thereof in consideration of expected profits in marketing activities will be described.

Recently, with the development of online network services, services such as games and shopping are increasing online, and competition is intensifying. Therefore, each service principal is trying to manage the relationship with the customer as part of the marketing to attract more customers.

As part of such a scheme, in customer relationship management using online services such as online games and shopping, activities such as target marketing are performed to prevent the user from leaving. For example, a user may be prevented from exiting by paying a higher reward to a user who is more likely to leave the online service. In most cases, the greater the rewards paid, the lower the user's churn rate, while the higher the cost, the lower the profit.

On the other hand, in such a target marketing, cost-benefit is called expected profit. In order to maximize profits of online services, many efforts are being made to optimize the expected profit that can be obtained from users who are subject to departure prevention.

In the present disclosure, a departure prevention system and a control method thereof considering an expected profit in a marketing activity are disclosed. In detail, a system and a control method of optimizing the expected profit by differentially paying a reward according to the expected profit that can be obtained from a user who is an object of departure prevention in the online service may be disclosed.

Of course, the technical problem to be solved is not limited to the technical problems as described above, and various technical problems may be further included within the scope apparent to those skilled in the art.

The departure prevention system in consideration of the expected profit according to the first aspect includes a prediction module that determines a departure prediction model for predicting a departure user among all users using a game log and machine learning; A target marketing module for distinguishing a predicted escape user from the predictive escape model into a control group and an experiment group, and performing target marketing of a user-specific differential compensation method for users belonging to the test group; And an expected profit optimization module for determining a marketing cost for maximizing expected profits by comparing the bounce rates of the users belonging to the control group and the experiment group, and determining an optimization parameter based on the expected profits.

The prediction module may modify the deviation prediction model based on the optimization parameter determined by the expected benefit optimization module.

The departure prediction module may determine, as the departure prediction user, a user whose departure probability is greater than an optimization parameter among all the users according to the departure prediction model.

In addition, the expected profit may be determined in consideration of a conversion rate of a user belonging to the experimental group, a user ratio actually deviating from the expected deviation user, a ratio of a user not actually deviating from the expected deviation user, and a ratio of a user who is not expected to leave. Can be.

In addition, the conversion rate may be a difference between the ratio of the user leaving the experimental group and the control group.

In addition, the conversion rate is determined according to the marketing cost, the optimal conversion rate compared to the marketing cost in consideration of at least one of the expected value for each user, the maximum possible conversion rate of the user and the width of the interval where the genetic conversion rate relative to the marketing cost is effective. This can be determined.

In addition, the marketing cost may be applied differently according to the experimental group.

In addition, the optimization parameter value may be selected as a value within a predetermined range in a multi-dimensional probability density function generated as expected profit value for marketing cost repeatedly performed through machine learning.

In addition, the departure prevention system control method in consideration of the expected benefit according to the second aspect, the operation of determining the departure prediction model for predicting the departure user of the total user using the game log and machine learning; Distinguishing a predicted escape user from the predicted escape model into a control group and an experiment group, and performing target marketing of a differential compensation scheme for each user for the users belonging to the experiment group; And comparing the dropout rates of the users belonging to the control group and the experiment group to determine a marketing cost that maximizes the expected profit, and determining an optimization parameter based on the expected profit.

The third aspect may also provide a computer readable recording medium having recorded thereon a program for executing the method of the second aspect on a computer.

1 is a view illustrating a departure prevention system in consideration of an expected profit according to an exemplary embodiment.
2 is a block diagram illustrating in detail an operation of a departure prevention system in consideration of expected profit according to an exemplary embodiment.
3 is a view for explaining an expected profit optimization prediction model of the departure prevention system according to an embodiment.
4 is a chart illustrating a measurement value of a parameter for an expected profit optimization prediction model, according to an exemplary embodiment.
5 is a diagram illustrating a relationship between a marketing cost and a conversion rate, according to an exemplary embodiment.
6 is a diagram for describing an example of selecting an optimized point in an expected profit optimization prediction model according to an embodiment.
7 is a flowchart illustrating a control method of a departure prevention system in consideration of expected profit according to an exemplary embodiment.

The terminology used in the embodiments is a general term that has been widely used as much as possible in consideration of the functions of the present invention, but may vary according to the intention or precedent of a person skilled in the art, the emergence of new technology, and the like. In addition, in certain cases, there is also a term arbitrarily selected by the applicant, in which case the meaning will be described in detail in the description of the invention. Therefore, the terms used in the present invention should be defined based on the meanings of the terms and the contents throughout the present invention, rather than the names of the simple terms.

When a part of the specification is said to "include" any component, this means that it may further include other components, except to exclude other components unless otherwise stated. In addition, the “…” described in the specification. Wealth ”,“… Module ”means a unit for processing at least one function or operation, which may be implemented in hardware or software, or a combination of hardware and software.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention.

Hereinafter, with reference to the drawings will be described embodiments of the present invention;

1 is a view illustrating a departure prevention system 100 in consideration of an expected profit according to an exemplary embodiment.

Referring to FIG. 1, the departure prevention system 100 considering the expected profit may include at least one server 10, a terminal 20, and a network 30. In addition, the server 10 according to an embodiment may include a prediction module 110, a target marketing module 120, and an expected profit optimization module 130.

For example, the server 10 may provide a service such as a game through an application or software installed in the terminal 10 in cooperation with the terminal 20 through the network 30. In addition, the server 10 may provide target marketing optimized so that users of the terminal 10 do not deviate in consideration of expected benefits in providing a service such as a game.

In detail, the prediction module 110 may determine a departure prediction model that predicts a deviation user among all users by using the game log and machine learning. The prediction module 110 may generate a model that may classify the exit user and the remaining user by using a game log and a machine learning algorithm of the previously exited user and the remaining user, and may predict the user to leave using this model in advance. have. In addition, the prediction module 110 measures a probability that the user will leave, and if the probability is greater than the preset optimization parameter t, the prediction module 110 determines that the user will leave. In this case, the optimization parameter t may have a value between 0 and 1, initially starting from 0.5, and then adjusted through the expected benefit optimization module.

The target marketing module 120 may distinguish the predicted departure user from the predictive departure model into a control group and an experimental group, and perform target marketing by a differential compensation method for each user for the users belonging to the experimental group.

For example, the target marketing module 120 divides the anticipated users into at least two groups (control group, experimental groups 1, 2, ...) to measure the accurate anti-deviation effect. The target marketing module 120 may not perform any marketing on the control group, and may perform differentiated marketing on the experimental group according to the expected value (CLV) of the target user. In this case, the expected value CLV for each user may be estimated in advance in consideration of past payment history and life expectancy.

In addition, the expected profit optimization module 130 may determine the marketing cost to maximize the expected profit by comparing the bounce rate of the user belonging to the control group and the experimental group, and may determine the optimization parameter based on the expected profit.

For example, the expected profit may be determined in consideration of the conversion rate of the users belonging to the experimental group, the ratio of the users who actually dropped out of the expected dropout users, the ratio of the users who did not actually drop out of the expected dropout users, and the ratio of the users who are not the expected dropout users.

In addition, the conversion rate may mean a difference between the ratio of the user leaving the experimental group and the control group. For example, the conversion rate is determined according to the marketing cost, and considering the at least one of the expected value of each user, the maximum possible conversion rate of the user, and the width of the interval in which the genetic conversion rate is more effective than the marketing cost, Can be determined. Meanwhile, the marketing cost may be applied differently according to the experimental group.

In addition, the above-described optimization parameter value may be selected as a value within a predetermined range from the multi-dimensional probability density function generated as expected profit value against marketing cost repeatedly performed through machine learning.

Meanwhile, the terminal 20 according to an embodiment includes a mobile phone, a smart phone, a laptop computer, a tablet PC, a smart TV, a digital broadcasting terminal, a personal digital assistant (PDA), and a portable multimedia player (PMP). ), Navigation, and the like, but are not limited thereto. For example, at least one user may use a game service or the like that is executed in cooperation with the server 20 and the network 30 through an application or software installed in the terminal 10.

The network 30 according to an embodiment may include a wireless network as well as a wired network. For example, the network 30 may be configured through a predetermined mobile communication network including 4G and 5G or short range communication including Wi-Fi and Bluetooth.

Meanwhile, although only components for describing the present embodiment are illustrated in FIG. 1, the components are not necessarily limited to these components, and various components for operating the departure prevention system may be further included.

2 is a block diagram illustrating in detail the operation of the departure prevention system 100 in consideration of the expected profit according to an exemplary embodiment.

Referring to FIG. 2, in step S110, the departure prevention system 100 determines a departure prediction model for predicting an exit user 220 among all the users 210 using game logs and machine learning. , The departure probability determined for each user according to the departure prediction model is classified as an exit prediction user 220 by comparing with the optimization parameter (hereinafter, t). Here, t may have a value between 0 and 1, initially starting from 0.5, and then adjusted through an expected profit optimization model.

Next, the prospective user 220 is separated into the control group 230 and the experimental group 240 again. At this time, the experiment group 240 for measuring the target marketing effect may be composed of a plurality.

In step S120, the departure prevention system 100 performs target marketing only for the users belonging to the experimental group 240. For example, target marketing of a differential compensation scheme for each user may be performed.

In step S130, anticipated anticipation benefit from departure of target marketing of the control group 230 and the experimental group 240 is determined. At this time, an optimization parameter is determined as a value for optimizing the determined expected profit.

 In step S140, the marketing cost is optimized to determine the optimization parameter, and in step S150, the prediction model is optimized using the optimization parameter.

Meanwhile, the values of the expected profit, the optimization parameter, and the marketing cost are not values that can be independently determined, and may be determined through machine learning within a predetermined probability range using a function having a plurality of variables.

Hereinafter, a method of determining an optimization parameter of an expected benefit optimization prediction model will be described in more detail with reference to FIGS. 3 to 6.

3 is a view for explaining an expected profit optimization prediction model of the departure prevention system according to an embodiment.

 Referring to FIG. 3, the total user is divided into a user 310 predicted as an exit and a user 320 predicted as a residual, and the total user is returned to the activity user 330 and the exit user 340 after the actual marketing. Again distinguished.

Therefore, the expected profit optimization model for maximizing the expected profit by using the changes of the users before and after marketing may be determined according to Equation 1 below.

[Equation 1]

Where Profit (t, C) is the expected profit, t is the optimization parameter, C is the marketing cost, CLV is the expected value per user, f (C) is the conversion rate, and g (t) is the probability of churn among all users. Ratio of large users) * (percent of users out of those whose probability of exit is greater than t), and h (t) is the ratio of total users whose probability of departure is greater than t) * K (t) is the ratio of the users whose exit probability is less than or equal to t) * (the percentage of users who remain less than or equal to t) ).

Meanwhile, in order to optimize the expected profit Prot (t, C), it is necessary to determine the optimization parameter t and the marketing cost C. At this time, g (t) and h (t) which are variables of the expected profit optimization module described above. Since k (t) depends on the optimization parameter, we need to determine the point where each variable has the optimal value through machine learning.

 For example, FIG. 4 is a chart illustrating measurement values of parameters for an expected profit optimization prediction model, according to an exemplary embodiment.

Referring to FIG. 4, a horizontal axis threshold refers to an optimization parameter and a vertical axis rate may be a graph indicating a ratio. Here, g (t) represents (the ratio of the users whose exit probability is greater than t) * (the ratio of users who actually exited among the users whose exit probability is larger than t). h (t) represents (ratio of users whose exit probability is greater than t) * (ratio of users whose exit probability is not larger than t). k (t) represents (the ratio of the users whose exit probability is less than or equal to t) of the total users * (the actual remaining user ratio among the users whose exit probability is less than or equal to t).

Referring to the graph shown in FIG. 4, it can be seen that k (t) increases as the optimization parameter increases, and g (t) and h (t) decrease as the optimization parameter increases.

이고, 이탈유저(340)에 대한 기대이익은

로 결정될 수 있다. 또한 이탈로 예측된 유저(310)가 실제 잔존(380)할 경우 활동유저(330)에 대한 기대이익은

으로 결정될 수 있다. 여기서, γ는 유저 전환율을 의미할 수 있다.Meanwhile, referring back to FIG. 3, if the user 310 predicted as an exit is actually exiting 370, the expected profit for the activity user 330 is

The expected profit for the exit user 340 is

Can be determined. In addition, if the user 310 predicted as a departure is actually remaining 380, the expected profit for the activity user 330 is

Can be determined. Here, γ may mean a user conversion rate.

In addition, if the user 320 predicted as remaining is actually remaining 360, the expected profit for the activity user 330 is determined as CLV, and when the user 320 predicted as remaining remains actual departure 350, the expected profit is expected. No profit can be determined. This is because the target marketing cost is not performed for the user 320 predicted as remaining, so that the marketing cost is zero.

Specifically, with reference to FIG. 5, the relationship between the marketing cost and the conversion rate according to an embodiment. Here, the difference between the percentage of users who actually dropped out of the experimental group and the control group is defined as the conversion rate f (C).

First, the conversion rate according to the marketing cost may be determined according to Equation 2 below.

[Equation 2]

는 최대 가능한 유저 전환율,

는 마케팅 비용 대비 유저 전환율의 효과가 높은 구간의 너비,

는 마케팅 비용의 변화 대비 전환율의 상승폭이 최대인 지점을 의미할 수 있다.Where f (C) is the conversion rate, C is the marketing cost, CLV is the expected value per user,

Is the maximum possible user conversion rate,

Is the width of the interval for which user conversion rate is more effective than marketing costs,

May indicate the point at which the rate of increase in the conversion rate is the maximum compared to the change in the marketing cost.

,

,

는 서비스의 특성에 따라 달라질 수 있다. 예를들면,

,

,

의 구체적인 값은 타겟 마케팅 실험을 이용하여 결정된다. 여기서, 적어도 하나이상의 실험군에 대해 마케팅 비용을 다르게 적용하여 수학식 2를 만족하는 미지수

,

,

를 도 5에 도시된 그래프를 이용하여 결정할 수 있다. 여기서, 가로축 campaign cost는 타겟 마케팅 비용, 세로축 conversion rate는 전환율(f(C))을 의미한다.Meanwhile,

,

,

May vary depending on the characteristics of the service. For example,

,

,

The specific value of is determined using the target marketing experiment. Here, an unknown value that satisfies Equation 2 by applying marketing costs differently to at least one experimental group.

,

,

Can be determined using the graph shown in FIG. Here, the horizontal axis campaign cost refers to the target marketing cost, and the vertical axis conversion rate refers to the conversion rate f (C).

6 is a diagram for describing an example of selecting an optimized point in an expected profit optimization prediction model according to an embodiment.

When Equations 1 and 2 described above are repeatedly performed through machine learning, a multi-dimensional probability density function that varies optimization parameters and marketing costs may appear as shown in the graph of FIG. 6.

Therefore, the departure prevention system 100 according to an embodiment of the present invention is a multi-dimensional probability density graph implemented through iterative machine learning, and finally, an optimization parameter having an optimal expected profit from a marketing cost (horizontal axis, cost) ( The vertical axis 610 may be determined. Of course, the optimization parameter 610 is not necessarily determined as a single value and may be determined within a range in which an expected profit is greater than or equal to a predetermined percentage.

7 is a flowchart illustrating a control method of a departure prevention system in consideration of expected profit according to an exemplary embodiment.

Referring to FIG. 7, the control method of the departure prevention system 100 includes steps that are processed in time series in the departure prevention system 100 or the server 10 shown in FIG. 1. Therefore, even if omitted below, it can be seen that the contents described above with respect to the departure prevention system 100 or the server 10 illustrated in FIGS. 1 to 6 also apply to the method illustrated in FIG. 7.

In operation 710, the departure prevention system control method according to an exemplary embodiment may determine a departure prediction model that predicts a departure user among all users using a game log and machine learning.

In operation 720, the method of controlling the departure prevention system according to an embodiment may distinguish a predicted departure user from the prediction model into a control group and an experiment group, and perform target marketing by a differential compensation scheme for each user for the users belonging to the experiment group. can do.

In operation 730, the method of controlling a departure prevention system according to an exemplary embodiment may compare a departure rate of a user belonging to a control group and an experiment group to determine a marketing cost that maximizes an expected profit. Here, the expected profit may be determined in consideration of the conversion rate of the user belonging to the experimental group, the ratio of the user who actually dropped out of the expected dropout user, the ratio of the user who did not actually drop out of the expected dropout user, and the ratio of the user who is not the expected dropout user.

In operation 740, the departure prevention system control method may determine an optimization parameter based on the expected benefit. In addition, the method of controlling the departure prevention system may modify the departure prediction model based on the optimization parameter determined in the expected profit optimization module, and determine a user whose departure probability according to the departure prediction model among all users is larger than the optimization parameter as the departure prediction user. Can be.

On the other hand, the above-described method can be written as a program that can be executed in a computer, it can be implemented in a general-purpose digital computer to operate the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described method can be recorded on the computer-readable recording medium through various means. The computer-readable recording medium may include a storage medium such as a magnetic storage medium (eg, ROM, RAM, USB, floppy disk, hard disk, etc.), an optical reading medium (eg, CD-ROM, DVD, etc.). do.

Those skilled in the art will appreciate that the present invention may be embodied in a modified form without departing from the essential characteristics of the above-described substrate. Therefore, the disclosed methods should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

Claims (10)

A prediction module for determining an escape prediction model predicting an escape user by using the game log and machine learning;
A target marketing module for distinguishing a predicted escape user from the predicted escape model into a control group and an experiment group, and performing target marketing by a differential compensation method for each user for the users belonging to the test group; And
And an expected benefit optimization module that compares a departure rate of the user belonging to the control group and the experimental group to determine a marketing cost for maximizing expected profit and determines an optimization parameter based on the expected profit. The method of claim 1,
The prediction module is to modify the departure prediction model based on the optimization parameters determined in the expected benefit optimization module, departure prevention system. The method of claim 1,
The departure prediction module is configured to determine a user whose departure probability is greater than an optimization parameter among all the users according to the departure prediction model. The method of claim 1, wherein the expected profit is
A departure prevention system that is determined in consideration of the conversion rate of the users belonging to the experimental group, the ratio of users who actually escaped from the expected escape users, the ratio of users who do not actually escape from the expected escape users, and the ratio of users who are not expected to leave. . The method of claim 4, wherein
The conversion rate is a difference between the ratio of the user leaving the experimental group and the control group, escape prevention system. The method of claim 5,
The conversion rate is determined according to the marketing cost, and a point at which the conversion rate is optimal to the marketing cost is determined by considering at least one of the expected value of each user, the maximum possible conversion rate of the user, and the width of the interval in which the genetic conversion rate is more effective than the marketing cost. The departure prevention system. The method of claim 1,
The marketing cost is applied differently according to the experimental group, departure prevention system. The method of claim 1,
Wherein the optimization parameter value is selected as a value within a predetermined range from a multi-dimensional probability density function generated from expected cost-to-market value values repeatedly performed through machine learning. Determining an escape prediction model that predicts an escape user among all users using the game log and machine learning;
Distinguishing a predicted escape user from the predicted escape model into a control group and an experiment group, and performing target marketing of a differential compensation scheme for each user for the users belonging to the experiment group; And
And comparing the departure rates of the users belonging to the control group with the experimental group to determine a marketing cost that maximizes the expected profit, and determining an optimization parameter based on the expected profit. The method of claim 9,
A departure prevention system control program, combined with a computer, which is hardware, and stored in a medium for executing the departure prevention system control method of claim 9.

Images