CN108648000B

CN108648000B - Method and device for evaluating user retention life cycle and electronic equipment

Info

Publication number: CN108648000B
Application number: CN201810371862.3A
Authority: CN
Inventors: 张燕; 谢毅
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2018-04-24
Filing date: 2018-04-24
Publication date: 2022-10-28
Anticipated expiration: 2038-04-24
Also published as: CN108648000A

Abstract

The invention discloses a method and a device for evaluating a user retention life cycle, electronic equipment and a computer readable storage medium, wherein the scheme comprises the following steps: acquiring historical active data of a target user on a specified project; dividing the target user into a class of user groups with the same active duration according to the active duration recorded by the historical active data; clustering historical active data of all users in a user group according to the designated cluster number to obtain a sub-life cycle mode of a target user; and matching the sub-life cycle patterns with the complete life cycle patterns of the appointed cluster number one by one to determine the complete life cycle patterns of the target user. The complete life cycle mode of the target user is obtained through prediction by the scheme of the invention, the behavior habit of the target user is better met, and the actual life cycle mode of the target user is better met, so that the problem that the life cycle mode of the user cannot be accurately predicted in the prior art is solved, and the conversion rate of advertisement putting is effectively improved.

Description

Method and device for evaluating user retention life cycle and electronic equipment

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for evaluating a user retention life cycle, an electronic device, and a computer-readable storage medium.

Background

In recent years, with the rapid development of computer network technology, the internet has been widely used. The users can conveniently and quickly complete various daily requirements such as information acquisition, shopping, payment, ticket reservation and the like through the internet, so that the dependence of the users on the internet is intensified. For the website, how to know the retention life cycle of each user in the first time so as to obtain the attention of the user to the client in the whole life cycle provides accurate and timely support for a website decision maker in the aspects of product sale and publicity strategies.

The retention life cycle refers to the whole development process from the establishment of the relationship with the client to the complete disconnection of the relationship with the client through the Internet. For example, as shown in FIG. 1, a game user's retention life cycle process may be divided into 4 phases, a try-play phase, a build phase, a stabilization phase, and an attrition phase. The stage with high conversion rate can be found out by analyzing the conversion rate (the ratio of the times of completing the conversion behavior to the total click times of the promotion information) of each stage, and the users in the stage are taken as target users to carry out advertisement putting, so that the conversion rate of the advertisement can be greatly improved.

In the prior art, the user is determined in which stage based on the life cycle retention process of the user, and then the advertisement is put on the user in the high conversion rate stage, so that the premise is that the life cycle rules of different users are similar, and the life cycle process of the user can be divided by adopting the same method. However, in reality, different users have different behavior habits, and the investment degree on the client side has different rules along with the change of time. However, in the prior art, the life cycle processes of different users cannot be accurately predicted, and thus, targeted advertisement putting cannot be performed.

Disclosure of Invention

In order to solve the problem that the life cycle process of a user cannot be accurately predicted in the related technology, the invention provides a method for evaluating the user retention life cycle.

In one aspect, the invention provides a method of assessing a user's retention lifecycle, the method comprising:

acquiring historical active data of a target user on a specified project; the historical activity data comprises an activity duration;

dividing the target user into a class of user groups with the same active duration according to the active duration recorded by the historical active data;

clustering historical active data of all users in the user group according to the designated cluster number to obtain a sub life cycle mode of the target user;

and matching the sub-life cycle patterns with the complete life cycle patterns of the appointed cluster number one by one to determine the complete life cycle pattern of the target user.

In another aspect, the present invention also provides an apparatus for evaluating a user retention lifecycle, the apparatus comprising:

the data acquisition module is used for acquiring historical active data of a target user on a specified project; the historical activity data comprises an activity duration;

the user dividing module is used for dividing the target user into a class of user groups with the same active duration according to the active duration recorded by the historical active data;

the pattern clustering module is used for clustering the historical active data of all users in the user group according to the designated cluster number to obtain the sub-life cycle pattern of the target user;

and the pattern matching module is used for matching the sub-life cycle patterns with the complete life cycle patterns of the appointed cluster number one by one to determine the complete life cycle patterns of the target user.

In addition, the present invention also provides an electronic device including:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to perform the above method of evaluating a user retention life cycle.

Furthermore, the present invention also provides a computer-readable storage medium storing a computer program executable by a processor to perform the above method of evaluating a user retention life cycle.

The technical scheme provided by the embodiment of the invention can have the following beneficial effects:

the invention embodies commonalities among users by mining a complete lifecycle pattern for a specified number of clusters. And then matching the sub-life cycle patterns generated according to the historical active data of the target user with the complete life cycle patterns of the appointed cluster number one by one, thereby obtaining the prediction result of the complete life cycle patterns of the target user. The complete life cycle mode of the target user is obtained through prediction in the mode, the behavior habit of the target user is better met, the actual life cycle mode of the target user is better met, the user with high conversion rate can be found out according to the life cycle mode of the user, the accurate advertisement putting is achieved, and the problem that the advertisement cannot be accurately put due to the fact that the life cycle mode of the user cannot be accurately predicted in the prior art is solved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.

Drawings

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

FIG. 1 is a schematic diagram of a life cycle process of a game user provided by the prior art;

FIG. 2 is a schematic illustration of an implementation environment in accordance with the present invention;

FIG. 3 is a block diagram illustrating a server in accordance with an exemplary embodiment;

FIG. 4 is a flow diagram illustrating a method of evaluating a user retention lifecycle according to an example embodiment;

FIG. 5 is a schematic diagram of an application scenario for evaluating a user retention life cycle according to an embodiment of the present invention;

FIG. 6 is a flowchart detailing step 410 of the corresponding embodiment of FIG. 4;

FIG. 7 is a detailed flowchart of step 450 of the corresponding embodiment of FIG. 4;

FIG. 8 is a flowchart of details of step 470 of the corresponding embodiment of FIG. 4;

FIG. 9 is a flowchart of another method for evaluating a user retention life cycle provided by the present invention based on the corresponding embodiment in FIG. 8;

FIG. 10 is a state diagram of the complete life cycle of various game users;

FIG. 11 is a comparison of clustering results for different cluster numbers;

FIG. 12 is a flowchart of another method for evaluating a user retention life cycle provided by the present invention based on the corresponding embodiment of FIG. 9;

FIG. 13 is a comparison graph of complete active data clustering results for different lifecycle lengths;

FIG. 14 is a comparison of 4 complete life cycle patterns of a game user versus the game type

FIG. 15 is a comparison graph of clustering results for multi-game users and single-game users;

FIG. 16 is a comparison graph of ad placement conversion rates;

FIG. 17 is a block diagram illustrating a method apparatus for evaluating a user retention lifecycle, according to an example embodiment;

FIG. 18 is a detailed block diagram of a data acquisition module in the corresponding embodiment of FIG. 17;

FIG. 19 is a block diagram illustrating the details of the pattern clustering module in the corresponding embodiment of FIG. 17;

fig. 20 is a detailed block diagram of a pattern matching module in the corresponding embodiment of fig. 17.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings in which the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.

FIG. 2 is a schematic diagram illustrating an implementation environment to which the present invention relates, according to an exemplary embodiment. The implementation environment to which the present invention relates includes a server 210. The historical active data of the target user and the complete life cycle mode of the designated cluster number are stored in the database of the server 210, so that the server 210 can adopt the method for evaluating the user retention life cycle provided by the invention to obtain the sub life cycle mode of the target user based on the historical active data of the target user, and then match the sub life cycle mode with the complete life cycle mode of the designated cluster number to obtain the complete life cycle mode of the target user.

The enforcement environment may also include sources that provide data, i.e., historical activity data, full lifecycle patterns for a specified number of clusters, etc., as desired. In particular, in this implementation environment, the data source may be the mobile terminal 230. The server 210 may receive historical activity data from the mobile terminal 230 and the full lifecycle pattern specifying the number of clusters may be sent by the mobile terminal 230 directly to the server 210 or may be generated by the server 210 from the full activity data of a plurality of sample users provided by the mobile terminal 230.

It should be noted that the method for evaluating the user retention life cycle provided by the present invention is not limited to deploying corresponding processing logic in the server 210, and may also be processing logic deployed in other machines. For example, processing logic for user lifecycle pattern prediction, etc., is deployed in a computing-capable terminal device.

Referring to fig. 3, fig. 3 is a schematic diagram of a server structure according to an embodiment of the present invention. The server 300 may vary significantly depending on configuration or performance, and may include one or more Central Processing Units (CPUs) 322 (e.g., one or more processors) and memory 332, one or more storage media 330 (e.g., one or more mass storage devices) storing applications 342 or data 344. Memory 332 and storage media 330 may be, among other things, transient storage or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instruction operations for the server 300. Still further, the central processor 322 may be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the server 300. The Server 300 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input-output interfaces 358, and/or one or more operating systems 341, such as a Windows Server ^TM ，Mac OS ^XTM ，Unix ^TM ,Linux ^TM ，FreeBSD ^TM And so on. The steps performed by the server described in the embodiments of fig. 4, 6-9, and 12 below may be based on the server shown in fig. 3And (5) structure.

It will be understood by those skilled in the art that all or part of the steps for implementing the following embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

FIG. 4 is a flow diagram illustrating a method of evaluating a user retention lifecycle, according to an example embodiment. The method for evaluating the user retention lifecycle is applicable and subject to execution, e.g., the method is used in the server 210 of the implementation environment shown in fig. 2. As shown in fig. 4, the prediction method, which may be performed by the server 210, may include the following steps.

In step 410, acquiring historical activity data of a target user on a specified item; the historical activity data includes an activity duration.

The user retention life cycle refers to the participation situation (such as participation probability) of the user every day from the first time of participating in the specified project to the last time of participating in the project. The target user refers to a user who needs to perform retention life cycle pattern evaluation. The designated item may be some type of application for participation by the user, such as a game software app, a social application-type app, a life consumption-type app, a video-audio playing-type app, etc., so as to evaluate the remaining life cycle (including total number of participation days, probability of participation per day, etc.) of the user using the application. The designated project may also be a certain sport the user is engaged in, a certain work the user is engaged in, etc., as desired, to assess the total number of days the user is engaged in a certain sport or work and the probability of engagement per day.

Historical activity data refers to the state of the target user as active on the specified item each day before predicting the target user's lifecycle pattern. For example, if a target user logs in a specific game software app on a certain day, the active data of the day may be considered to be 1, and if no log record exists on the certain day, the active data of the day may be considered to be 0. From the time that a target user registers to use the game software app on the first day, the activity data of the user on the game software app every day is taken as the historical activity data of the target user until the historical activity data is acquired. Similarly, the target user participates in a certain movement on a certain day, and may consider that the active data of the day is 1, and does not participate in the movement on the next day, and may consider that the active data of the day is 0, and so on. From the first day of the target user participating in the sport to the time of acquiring historical activity data, the activity data of each day of the user is taken as the historical activity data of the target user participating in the sport.

The active duration refers to the number of active days of the target user on the specified project recorded in the historical active data, namely the number of days that the target user has participated in the specified project. For example, the target user registers the total number of days to log in a certain game software app to the time historical activity data is currently acquired the first day.

Specifically, the mobile terminal 230 may also collect active data of the target user on the specified project in real time and report the active data to the server 210, and the server 210 may record the reported active data in real time in a local database, and the active conditions reported every day constitute historical active data. When the target user's lifecycle pattern needs to be predicted, server 210 may retrieve historical activity data for the target user for the specified project from a local database.

In step 430, dividing the target user into a class of user groups with the same active duration according to the active duration recorded by the historical active data;

the user group refers to a group of users composed of a plurality of users with the same active time length. It should be noted that the target user with a certain active duration may consider that the user already has a certain life cycle length, and is in the life cycle at this time, but complete life cycle data is not yet formed, so that the complete life cycle mode to which the target user belongs cannot be determined by matching with the complete life cycle mode of the specified cluster number through methods such as similarity calculation or trajectory similarity.

According to the method, the target users are divided into a class of user groups with the same active time length in a clustering judgment mode according to the current active time length of the target users. In other words, the current active duration of all users in the user group is equal. Assuming that the active time length recorded in the historical active data by the target user is 5 weeks, the target user and the users with the current active time length of 5 weeks are divided into a group.

In step 450, clustering historical active data of all users in the user group according to the designated cluster number to obtain a sub-life cycle mode to which the target user belongs;

wherein the number of clustered clusters represents the number of lifecycle patterns. The assigned cluster number can be obtained by clustering the complete active data of a large number of sample users in advance, and if three complete life cycle modes with the cluster number of 3 are generated, the discrimination between clusters is high, and the number of people is reasonably distributed, the assigned cluster number can be 3. Similarly, if the number of clusters is 4, the degree of discrimination between clusters is higher and the population distribution is more reasonable, the designated number of clusters may be 4.

And according to the appointed cluster number, clustering the historical active data of all users in the user group through a clustering algorithm. For example, the users of the user group can be grouped into four classes through a clustering algorithm to obtain sub-life cycle patterns of the four classes of users, and further obtain the sub-life cycle patterns of the target user according to the class to which the target user belongs.

It should be noted that the sub-lifecycle patterns are incomplete, and are the lifecycle patterns of the target user from the beginning of the contact with the specified item to the current moment. The sub-lifecycle mode is relative to the full lifecycle mode. For example, a full life cycle mode for a game user refers to the entire process from the beginning of a user's contact to the end of the game when the user is not playing the game. The sub-lifecycle patterns are only part of this overall process. The invested degree of the user to the game has a certain rule along with the change of time. It can be generally divided into a try-in period, a formation period, a stabilization period and a run-off period. While the sub-lifecycle pattern may only include the previous try-and-form period.

In step 470, the sub-lifecycle patterns are matched with the full lifecycle patterns of the designated cluster number one by one, and the full lifecycle pattern of the target user is determined.

It should be noted that the complete life cycle mode for specifying the number of clusters refers to clustering complete active data of a large number of sample users through a clustering algorithm, and obtaining a clustering result for specifying the number of clusters. Full activity data is full lifecycle data, which refers to the daily participation status of a sample user throughout the process from initial contact with an item to eventual non-participation in the item.

Assuming that the number of designated clusters is 4, that is, by clustering a large amount of sample data, 4 complete life cycle patterns are generated. The sub-lifecycle patterns of the target user are respectively matched with each complete lifecycle pattern, so that the complete lifecycle pattern most matched with the sub-lifecycle patterns can be obtained. The matching full lifecycle pattern is the predicted full lifecycle pattern of the target user from the historical activity data of the target user. Based on the target user's full lifecycle pattern, the target user's engagement probability for a specified project may be predicted each day thereafter. And by dividing the complete life cycle mode, the period of high conversion rate can be found out, and then the advertisement putting is carried out in a targeted manner.

In the prior art, the whole life cycle processes of all users are similar by default, then a high conversion period is found out based on the life cycle process shown in fig. 1, and the users in the period are subjected to advertisement putting, so that the advertisement putting rate is improved to different degrees. However, different users have different behavior habits, the game input degree has different changes in different periods, and if the life cycle rules of all the users are similar by direct default, the accuracy of the life cycle mode of each user is directly influenced, and further, the period with high conversion rate cannot be accurately divided for advertisement delivery.

It should be noted that although each game user has a different personal game lifecycle, there is a commonality between these users, and therefore the present invention embodies the commonality between these users by mining several representative game user lifecycle patterns (i.e., the complete lifecycle pattern for the specified number of clusters described above). And then matching the sub-life cycle mode of the target user with the complete life cycle mode, so as to predict the complete life cycle mode to which the target user belongs. The complete life cycle mode of the target user predicted by the method is more consistent with the behavior habit of the target user.

Furthermore, based on the complete life cycle mode which is more in line with the behavior habits of the users, the period with high conversion rate is divided for advertisement putting, and the advertisement conversion rate can be improved to the maximum extent.

Fig. 5 is a schematic diagram of an application scenario for evaluating a user retention life cycle according to an embodiment of the present invention. As shown in fig. 5, the application scenario includes a data platform 510, a media provider platform 520 (SSP), an advertisement transaction platform 530 (ADX), and an advertisement demander platform 540 (DSP). The advertisement demander platform 540 serves advertisers who can set target audiences of advertisements, delivery areas, advertising bids, etc. through the advertisement demander platform 540. The media provider platform 520 serves ad slots owners where media with rich media resources and user traffic can place their own ad slots, control the presentation of advertisements, set subsidies, and so on. The ad trafficking platform 530 interfaces the ad demander platform 540 with the media provider platform 520.

The Data Management Platform) 510 can provide user characteristic Data for the advertisement demander Platform 540, so that the advertisement demander Platform 540 can select a target audience of an advertisement in a targeted manner. Further, the data platform 510 may analyze the life cycle data of a large number of sample users by using the method for evaluating the user retention life cycle provided by the present invention, and mine the complete life cycle pattern of the specified cluster number and the corresponding transformation rule, so as to predict the complete life cycle pattern of the target user according to the historical active data of the target user, thereby effectively distinguishing users with different transformation capabilities.

The advertisement demander platform 540 may select users with high conversion capability as advertisement delivery users according to the conversion capability of different users. The advertisement trading platform 530 may select an appropriate advertisement according to the bid of the advertisement demander platform 540, and the media supplier platform 520 controls the advertisement to be displayed to the advertisement delivery user. Thereby. The method and the system realize the advertisement delivery to the users with high conversion capacity, greatly improve the conversion rate of the advertisement delivery, help the advertisers to maximize the ROI (return on investment), and form the virtuous circle of the advertisement delivery.

FIG. 6 is a flow chart illustrating a method of evaluating a user retention lifecycle, according to another exemplary embodiment. As shown in fig. 6, step 410 in the embodiment corresponding to fig. 4 specifically includes:

in step 411, acquiring a participation status record of the target user on the specified project;

wherein, the participation status record refers to the recorded status of the target user participating in the specified project every day. The participation status is divided into participation and non-participation. Assuming that the target user takes part in the specified project for the first time and obtains the participation state record at the current time, the total time is 25 days, and the state of whether the target user participates in the specified project every day within 25 days can be obtained.

When the user participates in the designated project, the user can be reported to the server 210 in real time by the mobile terminal 230 to which the user belongs, and the participation status record can be updated in real time by the server 210. When the life cycle pattern prediction needs to be performed on the target user, the server 210 has a database of itself to obtain the stored participation state record.

In step 412, historical activity data of the target user for the specified item is generated by sequentially encoding the engagement status records.

The step of sequentially coding the participation state records refers to the step of generating corresponding characters according to the participation states of the target user to the designated items every day. For example, a number 1 is generated if participating, and a number 0 is generated if not participating. By sequentially encoding the participation status records, a string such as 11101010110 may be generated, each numeral representing the present day participation status, which may serve as historical activity data of the target user for the specified project.

FIG. 7 is a flow chart illustrating a method of evaluating a user retention lifecycle, according to another exemplary embodiment. As shown in fig. 7, step 450 in the embodiment corresponding to fig. 4 specifically includes:

in step 451, clustering historical active data of all users in the user group through a spectral clustering algorithm to generate a sub-life cycle pattern with a specified cluster number;

the spectral clustering algorithm is established on the basis of spectrogram theory, and compared with the traditional clustering algorithm, the spectral clustering algorithm has the advantages that clustering can be performed on sample spaces with any shapes, and the global optimal solution is converged. The algorithm first defines an affinity matrix describing the similarity of pairs of data points from a given data set and computes the eigenvalues and eigenvectors of the matrix. Then select the appropriate feature vector to cluster the different data points. Assuming that the optimal cluster number is 4 when the complete active data of a large number of sample users are clustered by the spectral clustering algorithm, the historical active data of all users in a user group can be clustered by the spectral clustering algorithm to obtain 4 types of clustering results, i.e. a sub-life cycle pattern with the specified cluster number is generated.

In step 452, the sub-lifecycle pattern to which the target user belongs is obtained from the sub-lifecycle patterns of the specified cluster number.

Assuming that 4 types of sub-life cycle patterns are generated, the sub-life cycle patterns of the target user can be obtained according to the category to which the target user belongs during clustering.

It should be noted that, for some new users, for example, two weeks after the users just participated in a certain game, the data volume recorded in the participation state is small, the life cycle characteristic curve is not obvious, and if the users are forced to plan the life cycle pattern, an erroneous prediction may be generated. According to the conversion rate of different periods of the existing life cycle mode, when the user is in the game starting trial playing period, the conversion rate of advertisement putting is very high, so that the users with small data volume of the participation state records can be divided into one type, the advertisement putting is carried out on the users, and the conversion rate of the advertisement can be improved.

FIG. 8 is a flow chart illustrating a method of evaluating a user retention lifecycle, according to another exemplary embodiment. As shown in fig. 8, step 470 in the embodiment corresponding to fig. 4 specifically includes:

in step 471, calculating a similarity between the sub-lifecycle patterns of the target user and each full lifecycle pattern;

it should be noted that each full lifecycle mode can include the probability of participation of the user each day from the first participation to the last participation in a project. And the sub-lifecycle patterns may include the probability of the user participating in the project each day from the first participation to the current time.

Assuming that 4 complete life cycle patterns can be obtained by clustering the complete active data of all sample users through a spectral clustering algorithm, the feature vector a of each complete life cycle pattern can be generated according to the daily participation probability of each complete life cycle pattern. And generating a characteristic vector b according to the daily participation probability in the sub-life cycle mode of the target user, and calculating the Euclidean distance or cosine similarity of the characteristic vectors a and b to obtain the similarity between the sub-life cycle mode and each complete life cycle mode.

In step 472, the complete life cycle pattern with the highest similarity to the sub life cycle pattern is screened out from the complete life cycle patterns with the designated cluster number, so as to obtain the complete life cycle pattern of the target user.

Assuming that the number of the designated clusters is 4, according to the similarity between the sub-lifecycle patterns and each complete lifecycle pattern, the complete lifecycle pattern with the highest similarity to the sub-lifecycle pattern of the target user can be screened out from the 4 complete lifecycle patterns. And taking the screening result as a prediction result of the complete life cycle mode of the target user.

FIG. 9 is a flowchart illustrating a method for evaluating a user retention lifecycle according to yet another exemplary embodiment. As shown in fig. 9, on the basis of the embodiment shown in fig. 4, the following steps may be further included before step 470:

in step 901, acquiring complete active data of a plurality of sample users;

wherein the plurality of sample users may include a plurality of users participating in different projects. By way of example, the plurality of sample users may include users participating in sports-type games, users participating in racing-type games, users participating in card-type games, and users participating in role-playing-type games. Full activity data refers to the actual participation status of a sample user each day from the first time the user participates in a project to the last time the user participates in the project. If participating in the day may be marked as 1 and if not, as 0, whereby the complete activity data may be a string of characters in the form 110100 \ 8230, 1010001, each number representing in sequence the first day participation status, the second day participation status, the third day participation status, the nth day participation status.

In step 902, screening out candidate users having equal life cycles and only being active in a single item at the same time according to the complete active data of the plurality of sample users;

it should be noted that in order to effectively analyze the entire life cycle state of the user, a sample user with complete active data is then selected. According to the existing user life cycle analysis, the length of the blank window period is three weeks, that is, if a user does not participate in a specified project for three consecutive weeks, the life cycle of the user is considered to be finished. Thereby, active data of the whole process from the beginning of participation to the last participation of the user can be obtained.

Further, in order to eliminate the data source heterogeneity caused by the life cycles with different lengths, users with equal-length life cycles, that is, users with the same number of active days in the complete active data, may be selected. In addition, since the impact of the number of items in which the user is simultaneously participating on the analysis is unknown, it is possible to select only users that are active on a single item at the same time. Further, users meeting the two conditions can be selected from the sample users as candidate users. The specific data format is as follows:

user 1: (use State on day 1, use State on day 2, \ 8230;, use State on day n)

And (4) a user 2: (use State on day 1, use State on day 2, \ 8230;, use State on day n)

…

And a user m: (day 1 use state, day 2 use state, \ 8230;, day n use state);

in step 903, the complete active data of all candidate users are clustered, and a complete life cycle pattern of the specified cluster number is generated.

Specifically, all the candidate users may be divided into a plurality of game users according to the category of the game to be participated in, and the daily participation probability of the user of each type of game in the life cycle may be calculated according to the complete activity data of each candidate user. FIG. 10 is a state diagram of the overall life cycle of various game users. It should be noted that, because of the variety of games, in order to clearly reflect the different life cycle modes of the different types of games, fig. 10 only schematically lists the user's overall life cycle states of 5 different types of games, which respectively include games of the categories of role playing 1001, shooting-flying shooting 1002, playing card 1003, tactical-defense 1004, racing 1005, and the like. The abscissa represents the number of active days and the ordinate represents the probability of daily participation. It can be seen from the figure that the participation probability of the first day and the last day is 100%, the participation probability of each category of games is different in the middle period, and the curves shown in fig. 10 are clustered through the spectral clustering algorithm, so that curves of various complete life cycle modes with appropriate cluster numbers can be generated.

Optionally, the complete active data of all candidate users are clustered by a spectral clustering algorithm, so that three or four complete life cycle patterns can be generated.

It is to be explained that how many clusters represent the kind of the generated full lifecycle pattern. As shown in fig. 11, when the candidate user complete activity data with the activity day number of 60 days is clustered, the second cluster mining is not sufficient when the cluster number of clusters is 2, and there is a large refinement space because the cluster number of clusters is 3, and the cluster can be further subdivided into two types. When the number of clusters is 5 or more, most clusters are too thin and the occupancy ratio is extremely small, and thus the coverage rate is low in practical application, which is not suitable. When the number of clusters is 3 or 4, the discrimination between clusters is high, and the number distribution is reasonable. Thus, if a coarser analysis is performed to select a cluster number equal to 3, a general analysis may select a cluster number of 4.

Further, on the basis of the corresponding embodiment in fig. 9, as shown in fig. 12, after step 903 and before step 470, the method for evaluating the user retention life cycle provided by the present invention may further include the following steps:

in step 1201, acquiring a verified user with different life cycle lengths, active in different project categories, active in multiple projects at the same time or active in only a single project at the same time;

it should be noted that, in order to verify the correctness of the complete life cycle pattern of the generated designated cluster number, the analysis is mainly performed according to factors such as different life cycle lengths, activity in different project categories, activity in multiple projects at the same time, and activity in a single project at the same time. Thus, the authenticated users may include three types of users whose lifecycle lengths (i.e., full active days) are 60 days, 90 days, and 120 days. Authenticating users may include several classes of users active in different categories of items, such as different categories of games (sports, card, role-playing, and educational, etc.). Authenticating a user may include simultaneously activating multiple items, such as a user participating in multiple games simultaneously. The authenticated users may be simultaneously active on a single item, such as a single game user. By controlling the variable method, the verified user with one condition is obtained each time.

In step 1202, clustering is performed on the complete active data of the verification user according to the number of the designated clusters, so as to obtain a clustering result of the number of the designated clusters;

referring to the above embodiment, the complete active data of the verified user is clustered according to the designated cluster number by using a spectral clustering algorithm. For example, the number of designated clusters may be 4, and 4-cluster results are obtained.

In step 1203, the clustering result of the specified cluster number is compared with the complete life cycle pattern of the specified cluster number, and the correctness of the complete life cycle pattern is verified.

Specifically, the 4 clustering results are compared with the generated 4 complete life cycle patterns one by one, and the correctness of each complete life cycle pattern is verified. For example, the similarity between the clustering result and the complete life cycle pattern may be calculated, and when the similarity is greater than a threshold, the complete life cycle pattern is considered to be correct after verification is passed.

FIG. 13 is a comparison graph of the clustering results of complete active data for different life cycle lengths. As can be seen from fig. 13, the results for the 4-class clustering are completely the same, although the full life cycle lengths are different at 60 days, 90 days and 120 days, thereby indicating that the different life cycle lengths have less effect on the full life cycle pattern.

FIG. 14 is a comparison of 4 complete life cycle patterns of a game user versus the game type. Since the actual game is of a large variety, as shown in fig. 14, a schematic illustration of only 6 categories (strategy-defense, role play-MMO, chess and card, shooting-flight shooting, leisure-elimination, intellectual development) of games is made by clustering the complete active data of users who participated in the games and have a complete life cycle length of 60. As shown in fig. 14, 4 types (label 0, label1, label2, and label 3) were copolymerized, and the population ratio analysis was performed for each type of the cluster result, the population ratio of each of the 6 game types in the label0 cluster result was close, the population ratio of each of the 6 game types in the label1 cluster result was also close, the population ratio of each of the 6 game types in the label2 cluster result was still relatively close, and the population ratio of each of the 6 game types in the label3 cluster result was also close. It can be found that the number distribution of people of different game categories on the same cluster result is very similar, that is, the different game categories are distinguished in the cluster by small degrees. The impact of the participating item categories on the full lifecycle pattern can therefore be considered small.

FIG. 15 is a comparison of the clustering results of multi-game users and single-game users, and as shown in FIG. 15, by clustering the lifecycle data of single-game players and multi-game players, it can be found that the clustering results are similar, and also shows that this factor has little effect on the complete lifecycle pattern.

In summary, it can be known from the analysis of the experimental results of the life cycle length, the game category, and the single game player and the multi-game player that the complete life cycle pattern of the designated cluster number generated by the method provided by the present invention is not affected by these factors, and has a strong generalization ability. Therefore, the complete life cycle mode of the target user can be accurately predicted by matching the sub life cycle mode and the complete life cycle mode of the target user no matter what the life cycle length of the target user is, what the participating game item is, and whether the target user is a multi-game player.

The advertisement conversion rate of the users in each life cycle mode can be analyzed according to the feedback data of the users by putting the advertisements to the users in each life cycle mode according to needs. Taking the example of delivering a game advertisement, the definition of the conversion rate may be the number of game activation users/the number of exposure users.

Fig. 16 is a comparison graph of ad placement conversion rate. As can be seen from fig. 16, the conversion rate of the users in the trial play period (trial play period group in the figure) is relatively high, in accordance with the existing conclusion. By adopting the method provided by the invention, 4 complete life cycle modes are excavated, and the users of each life cycle mode are divided into one group and sequentially represented as cable-0, cable-1, cable-2 and cable-3. As shown in FIG. 16, the conversion rate of the user group of label-1 is higher than that of the ad group randomly delivered (i.e. the ad group randomly delivered in the figure), and the conversion rate of the other three groups of users is lower than that of the ad group randomly delivered. This shows that by predicting the complete life cycle mode to which the target user belongs, users with high conversion rate and users with low conversion rate can be effectively distinguished, and then the target user with high conversion rate is subjected to game advertisement delivery, so that the advertiser is helped to maximize the ROI, and the virtuous circle of advertisement delivery is formed.

Furthermore, if the feedback data of the advertisement delivery is more, each complete life cycle mode can be further refined, and the process of each complete life cycle mode is divided. For example, the complete lifecycle pattern of the cable-0 group is divided into two phases, a growth phase and a stabilization phase. The complete lifecycle pattern of the label-1 group is divided into two phases, a familiarity phase and an active phase, and so on. By means of the division, the users in the high conversion period can be found more finely, and the advertisement is put to the users in the high conversion period, so that the advertisement conversion rate is improved.

The following is an embodiment of an apparatus of the present invention, which can be used to execute an embodiment of the method for evaluating the user retention life cycle executed by the server 210 of the present invention. For details not disclosed in the embodiments of the apparatus of the present invention, please refer to the embodiments of the method for evaluating the user retention life cycle of the present invention.

Fig. 17 is a block diagram illustrating an apparatus for evaluating a user retention life cycle, which may be used in the server 210 of the implementation environment shown in fig. 2, according to an example embodiment, and which performs all or part of the steps of any one of the methods for evaluating a user retention life cycle shown in fig. 4, 6-9, and 12. As shown in fig. 17, the apparatus includes, but is not limited to: a data acquisition module 1710, a user partitioning module 1730, a pattern clustering module 1750, and a pattern matching module 1770.

A data obtaining module 1710, configured to obtain historical activity data of a target user on a specified project; the historical activity data comprises an activity duration;

the user dividing module 1730 is configured to divide the target user into a class of user groups with the same active duration according to the active duration recorded by the historical active data;

the mode clustering module 1750 is used for clustering the historical active data of all users in the user group according to the designated cluster number to obtain the sub-life cycle mode of the target user;

a pattern matching module 1770, configured to match the sub-lifecycle patterns with the complete lifecycle patterns of the designated cluster number one by one, to determine the complete lifecycle pattern of the target user.

The implementation processes of the functions and actions of each module in the device are specifically described in the implementation processes of the corresponding steps in the method for evaluating the user retention life cycle, and are not described herein again.

The data obtaining module 1710 can be, for example, one of the physical structure input/output interfaces 358 in fig. 3.

The user partitioning module 1730, the pattern clustering module 1750, and the pattern matching module 1770 may also be functional modules for performing corresponding steps in the above method for evaluating a user retention life cycle. It is understood that these modules may be implemented in hardware, software, or a combination of both. When implemented in hardware, these modules may be implemented as one or more hardware modules, such as one or more application specific integrated circuits. When implemented in software, the modules may be implemented as one or more computer programs executing on one or more processors, such as programs stored in memory 332 for execution by central processor 322 of FIG. 3.

In an exemplary embodiment, as shown in fig. 18, the data obtaining module 1710 includes:

a state obtaining unit 1711, configured to obtain a participation state record of the target user for the specified item;

a state encoding unit 1712, configured to generate historical activity data of the target user for the specified item by sequentially encoding the participation state records.

In an exemplary embodiment, as shown in fig. 19, the pattern clustering module 1750 includes:

the sub-pattern clustering unit 1751 is used for clustering the historical active data of all users in the user group through a spectral clustering algorithm to generate a sub-life cycle pattern with a specified cluster number;

a sub-pattern obtaining unit 1752, configured to obtain the sub-lifecycle pattern to which the target user belongs from the sub-lifecycle patterns of the specified cluster number.

In an exemplary embodiment, as shown in fig. 20, the pattern matching module 1770 includes:

a similarity calculation unit 1771, configured to calculate a similarity between the sub-lifecycle mode and each full lifecycle mode of the target user;

and a matching and screening unit 1772, configured to screen out a complete life cycle pattern with the highest similarity to the sub life cycle patterns from the complete life cycle patterns with the specified number of clusters, to obtain the complete life cycle pattern of the target user.

Further, the above apparatus for evaluating a user retention life cycle may further include:

the system comprises a sample acquisition module, a data processing module and a data processing module, wherein the sample acquisition module is used for acquiring complete active data of a plurality of sample users;

the user screening module is used for screening out candidate users which have equal life cycles and are only active in a single project at the same time according to the complete active data of the plurality of sample users;

and the user clustering module is used for clustering the complete active data of all the candidate users to generate the complete life cycle mode of the appointed cluster number.

Wherein the user clustering module may include:

and the spectral clustering unit is used for clustering the complete active data of all the candidate users through a spectral clustering algorithm to generate three or four complete life cycle modes.

the verification user acquisition module is used for acquiring verification users with different life cycle lengths, which are active in different project categories, which are active in a plurality of projects at the same time or which are active in only a single project at the same time;

a clustering result obtaining module, configured to cluster the complete active data of the verification user according to the specified cluster number, and obtain a clustering result of the specified cluster number;

and the clustering result comparison module is used for comparing the clustering result of the appointed cluster number with the complete life cycle mode of the appointed cluster number and verifying the correctness of the complete life cycle mode.

Optionally, the present invention further provides an electronic device, which can be used in the server 210 in the implementation environment shown in fig. 1 to execute all or part of the steps of the method for evaluating the user retention life cycle shown in any one of fig. 4, 6 to 9, and 12. The electronic device includes:

a processor;

a memory for storing processor-executable instructions;

The specific manner in which the processor of the electronic device performs operations in this embodiment has been described in detail in relation to the embodiment of the method for evaluating a user retention life cycle, and will not be described in detail here.

In an exemplary embodiment, a storage medium is also provided that is a computer-readable storage medium, such as may be transitory and non-transitory computer-readable storage media, including instructions. The storage medium stores a computer program executable by the central processor 322 of the server 300 to perform the above-described method of evaluating a user's retention life cycle.

It will be understood that the invention is not limited to the precise arrangements that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims

1. A method of evaluating a user's retention lifecycle, the method comprising:

clustering historical active data of all users in the user group according to the designated cluster number to obtain a sub-life cycle mode of the target user;

2. The method of claim 1, wherein the obtaining historical activity data of the target user for the specified item comprises:

acquiring the participation state record of the target user to the specified project;

and generating historical activity data of the target user on the specified item by sequentially coding the participation status records.

3. The method of claim 1, wherein the clustering historical activity data of all users in the user group according to a specified cluster number to obtain the sub-lifecycle pattern to which the target user belongs comprises:

clustering historical active data of all users in the user group through a spectral clustering algorithm to generate a sub-life cycle mode with a specified cluster number;

and obtaining the sub-life cycle mode of the target user from the sub-life cycle modes of the appointed cluster number.

4. The method of claim 1, wherein the matching the sub-lifecycle patterns to full-lifecycle patterns of a specified number of clusters one-to-one, determining the lifecycle pattern of the target user comprises:

calculating the similarity between the sub-life cycle mode and each complete life cycle mode of the target user;

and screening out the complete life cycle mode with the highest similarity to the sub life cycle modes from the complete life cycle modes with the appointed cluster number to obtain the complete life cycle mode of the target user.

5. The method of claim 1, wherein prior to matching the sub-lifecycle patterns to full-lifecycle patterns of a specified number of clusters one-to-one, and determining the full-lifecycle pattern of the target user, the method further comprises:

acquiring complete active data of a plurality of sample users;

screening out candidate users which have equal life cycles and are only active in a single project at the same time according to the complete active data of the plurality of sample users;

and clustering the complete active data of all candidate users to generate the complete life cycle mode of the appointed cluster number.

6. The method of claim 5, wherein the clustering the full activity data of all candidate users to generate the full lifecycle pattern for the specified number of clusters comprises:

and clustering the complete active data of all candidate users through a spectral clustering algorithm to generate three or four complete life cycle modes.

7. The method of claim 5, wherein after clustering the full activity data of all candidate users to generate the full lifecycle pattern for the specified number of clusters, the method further comprises:

acquiring verification users with different life cycle lengths, active in different project categories, active in multiple projects simultaneously or active in only a single project simultaneously;

clustering the complete active data of the verification user according to the specified cluster number to obtain a clustering result of the specified cluster number;

and comparing the clustering result of the appointed cluster number with the complete life cycle mode of the appointed cluster number, and verifying the correctness of the complete life cycle mode.

8. An apparatus to evaluate a user retention lifecycle, the apparatus comprising:

and the pattern matching module is used for matching the sub-life cycle patterns with the complete life cycle patterns of the appointed cluster number one by one and determining the complete life cycle pattern of the target user.

9. The apparatus of claim 8, wherein the data acquisition module comprises:

the state acquisition unit is used for acquiring the participation state record of the target user on the specified project;

and the state encoding unit is used for sequentially encoding the participation state records to generate historical active data of the target user on the specified project.

10. The apparatus of claim 8, wherein the pattern clustering module comprises:

the sub-mode clustering unit is used for clustering historical active data of all users in the user group through a spectral clustering algorithm to generate a sub-life cycle mode with a specified cluster number;

and the sub-pattern obtaining unit is used for obtaining the sub-life cycle pattern to which the target user belongs from the sub-life cycle patterns of the appointed cluster number.

11. The apparatus of claim 8, wherein the pattern matching module comprises:

the similarity calculation unit is used for calculating the similarity between the sub life cycle mode and each complete life cycle mode of the target user;

and the matching screening unit is used for screening out the complete life cycle pattern with the highest similarity to the sub life cycle pattern from the complete life cycle patterns with the appointed cluster number to obtain the complete life cycle pattern of the target user.

12. The apparatus of claim 8, further comprising:

the system comprises a sample acquisition module, a data acquisition module and a data processing module, wherein the sample acquisition module is used for acquiring complete active data of a plurality of sample users;

13. The apparatus of claim 12, wherein the user clustering module comprises:

14. An electronic device, characterized in that the electronic device comprises:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to perform the method of assessing a user retention life cycle of any of claims 1-7.

15. A computer-readable storage medium, characterized in that it stores a computer program executable by a processor to perform the method of assessing a user's retention life cycle according to any one of claims 1 to 7.