User dynamic layering method based on multiple data dimensions and computer program product
By employing a multi-data-dimensional dynamic user segmentation method, combined with sliding window decay, interval segmentation scoring, and quantile benchmarks, the accuracy problem of single-dimensional segmentation is solved, enabling refined evaluation and accurate segmentation of user value.
Patent Information
- Application Number
- CN202511517915.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-23
- Publication Date
- 2025-11-21
AI Technical Summary
Existing user segmentation methods are based on only a single data dimension, resulting in inaccurate segmentation results and difficulty in distinguishing between historically active users and recently active users, making them susceptible to the influence of extreme data.
A user dynamic stratification method based on multiple data dimensions is adopted. The effective activity score is calculated by dynamically decaying through a sliding window, the login score is determined by interval segmentation scoring, and the task score is adjusted by using a preset quantile benchmark. Finally, the user level is determined by weighted calculation.
It achieves more accurate user segmentation, distinguishing between historically active and recently active users, reducing the impact of extreme data, and improving the reliability of segmentation results.
Smart Images
Figure CN120996855A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of user behavior data analysis, and particularly relates to a user dynamic layering method based on multiple data dimensions and a computer program product. BACKGROUND
[0002] With the wide application of big data analysis technology, many enterprises analyze the behavior data of users on the websites, shopping malls or APPs operated by the enterprises, layer the users, mine high-value users, and accordingly formulate targeted operation strategies / user incentive methods to promote user retention or activity. The existing user layering method usually sets different level thresholds according to a single data dimension to layer the users, for example, sets multiple level thresholds according to user cumulative points or cumulative login times, and then divides the level of the user according to whether the historical total or the statistical period total of the user in this data dimension reaches the corresponding level threshold. This layering method is based on a single data dimension for layering, that is, only from a single dimension to evaluate the value of the user, which is too one-sided, and the layering result cannot distinguish between historical active users and recent active users, is easily affected by extreme data, and leads to inaccurate layering results. SUMMARY
[0003] The purpose of the present application is to provide a user dynamic layering method based on multiple data dimensions and a computer program product for implementing the method when executed, and the layering result obtained by using the method is more accurate.
[0004] The present application provides a user dynamic layering method based on multiple data dimensions, comprising the following steps: X, obtaining the active data, login times and task completion number of the user, and then executing the following steps S1-S3 without any order: S1, calculating the effective active score of the user according to the active data by using a sliding window dynamic attenuation method, specifically, the more distant the time point of the active data, the smaller the weight coefficient when calculating the effective active score; S2, determining the login score of the user according to the login times according to the login times interval segmentation scoring rule, specifically, the higher the login times interval, the slower the login score growth rate; S3, determining the preset quantile reference according to the task completion number of all users in the same period, and then: if the task completion number of the user exceeds the preset quantile reference, setting the task score of the user to the highest value, if the task completion number of the user does not exceed the preset quantile reference, deriving the task score according to the ratio of the task completion number of the user to the preset quantile reference, specifically, the larger the task completion number of the user, the higher the task score; Y, determining the level to which the user belongs according to the effective active score, the login score and the task score of the user, specifically, the level is positively correlated with the effective active score, the login score and the task score.
[0005] Further, in step S1, the effective active score C is calculated as follows:
[0006] Wherein: J i Active data of the ith month; n is the cumulative active month number of the user; λ is the decay coefficient, and the value is 0.2.
[0007] Further, in step S2, the login frequency interval segmentation scoring rule is as follows: Login frequency <5: login score D=login frequency*5; 5≤login frequency≤15: login score D=25+(login frequency-5)*4; 16≤login frequency<30: login score D=65+(login frequency-15)*2; Login frequency≥30: login score D=100.
[0008] Further, in step S3, the preset quantile reference is 90% quantile, if the task completion number of the user exceeds the 90% quantile of the task completion number of all users in the same period, the task score R of the user is set to 100, otherwise the task score R of the user is calculated according to the following formula:
[0009] Wherein, L is the task completion number of the user, Q 90 The 90% quantile of the task completion number of all users in the same period.
[0010] Further: In step S1, the active ranking score M of the user is also determined according to the effective active score C of the user. In step Y, the level to which the user belongs is specifically determined according to the active ranking score M, the login score D and the task score R of the user.
[0011] Further, the active ranking score M of the user is determined as follows:
[0012] Wherein, P is the ranking of the effective active score of the user among all users, and N is the number of all users.
[0013] Further, in step Y, specifically: The three of the active ranking score M of the user, the login score D and the task score R are weighted to obtain the comprehensive score S of the user; The comprehensive score interval corresponding to each level is obtained, and the user is divided into the level corresponding to the comprehensive score interval in which the comprehensive score S of the user is located.
[0014] Further, in step Y, the weighted calculation formula of the comprehensive score S of the user is specifically: S=0.5M+0.2D+0.3R.
[0015] The application further provides a computer program product comprising a computer program which, when executed, implements the user dynamic layering method based on multiple data dimensions as described above.
[0016] The user dynamic layering method provided by the application is based on the three data dimensions of the active data, the login times and the task completion number of the user, evaluates the user by comprehensively considering multiple behavior data, and is more beneficial to mine high-value users, and specifically: on the one hand, S1 adopts a sliding window dynamic attenuation method to calculate the effective active score of the user, and the historical active value and the recent active value of the user are comprehensively considered, the time point of the active data is more remote, the weight coefficient is smaller when the effective active score is calculated, and the contribution to the effective active score is lower, while the weight coefficient of the active data generated recently is relatively larger, and the contribution to the effective active score is higher, so that the recent active user can be effectively identified; on the other hand, S2 determines the login score of the user according to the login times interval segmentation scoring method, the higher the login times interval is, the slower the login score growth rate is, so that when the login times is less, the influence of the login times growth of the user on the user layering is strengthened, and when the login times reaches a certain degree, the influence of the login times growth of the user on the user layering is gradually weakened; on the other hand, S3 determines the task score of the user according to the preset quantile reference of the task completion number of all users in the same period, which can evaluate individual data in combination with the overall data, and can effectively suppress the influence of extreme data on the layering result. Y determines the level of the user according to the effective active score, the login score and the task score of the user, evaluates the user by comprehensively considering multiple dimensions, can distinguish the historical active user and the recent active user, is not easily affected by extreme data, and the layering result is more accurate. BRIEF DESCRIPTION OF DRAWINGS
[0017] Figure 1 is a flowchart of the user dynamic layering method based on multiple data dimensions provided by the application. DETAILED DESCRIPTION
[0018] The application will be further described in detail below in combination with the specific embodiments.
[0019] The present application adopts computer code to program a computer program product, which, when executed by a computer, implements a user dynamic layering method based on multiple data dimensions as shown in Figure 1 The following will take the user dynamic layering of a reading website on the first day of each month as an example to describe the execution process of the method in detail.
[0020] X, obtains the active data, login times and task completion numbers of the user, and then executes the following steps S1-S3 in no particular order.
[0021] For each user of the reading website, the computer obtains the active data, login times and task completion numbers of the user. In order to encourage users to use the website, the reading website rewards users with points for logging in to the reading website, reading books, commenting on books, etc., and the users can use the points to exchange reading coupons or goods. This embodiment takes the point data obtained by the user as the active data of the user. In order to better highlight the users who are active in the recent period, the computer only takes the point data of the past year as the active data, and only takes the data of the past month as the login times and task completion numbers. After the computer obtains the point data of the past year, the login times of the past month and the task completion numbers of the past month of the user, it calculates the score of the user in the three data dimensions through steps S1-S3.
[0022] S1, for the active data of the user, the effective active score of the user is calculated by using a sliding window dynamic attenuation method, specifically, the more distant the time point of the active data, the smaller the weight coefficient when calculating the effective active score.
[0023] The computer calculates the effective active score C of the user for the point data of the past year of the user by using a sliding window dynamic attenuation method, and the specific calculation is as follows:
[0024] Wherein: J i is the active data of the i-th month (i.e. the points of the i-th month, i is counted from the earliest month in time); n is the cumulative active month number of the user, which is 12; λ is the attenuation coefficient, which is 0.2.
[0025] In order to facilitate the comprehensive analysis of the data of the above three data dimensions on the same data dimension, after the computer calculates the effective active score C of the user, it needs to be further converted and processed, specifically, the computer determines the active ranking score M of the user according to the effective active score C of the user, and the determination method is as follows:
[0026] Wherein, P is the ranking of the effective active score of the user among all users, and N is the number of all users.
[0027] S2, according to the login times of the user, the login score of the user is determined according to the login times interval segmentation scoring rule, specifically, the higher the login times interval, the slower the corresponding login score growth rate.
[0028] In this embodiment, the login times is divided into four intervals, and then according to the principle that the higher the login times interval, the slower the corresponding login score growth rate, the login times interval segmentation scoring rule is designed as follows: (1) login times <5: login score D=login times*5; (2) 5≤login times≤15: login score D=25+(login times-5)*4; (3) 16≤login times<30: login score D=65+(login times-15)*2; (4) login times≥30: login score D=100.
[0029] The computer calculates the login score of the user according to the interval of the login times of the user in the past month, and substitutes it into the login score calculation formula of the interval to obtain the login score of the user, for example, the user's login times in the past month is 20 times, which belongs to 16≤login times<30, then its login score D=65+(20-15)*2=75.
[0030] S3, according to the task completion number of the user, the preset quantile number reference is determined according to the task completion number of all users in the same period, and then: if the task completion number of the user exceeds the preset quantile number reference, the task score of the user is set to the highest value, if the task completion number of the user does not exceed the preset quantile number reference, the task score is obtained according to the ratio of the task completion number of the user to the preset quantile number reference, specifically, the larger the task completion number of the user, the higher the task score.
[0031] In this embodiment, the 90% quantile of the task completion number of all users in the same period is taken as the preset quantile number reference. Since the task completion number of all users is different each time the user dynamic layering is performed, the computer calculates the preset quantile number reference of the current period before calculating the task score of each user, specifically, the task completion number of all users in the same period, that is, the task completion number in the past month, is arranged in ascending order, and the task completion number at the 90% position is taken as the preset quantile number reference (the task completion number of 90% of all users is less than this value). Then the computer compares the task completion number of each user with the 90% quantile of the current period, if the task completion number of the user exceeds the 90% quantile of the task completion number of all users in the same period, the task score R of the user is set to 100, otherwise the task score R of the user is calculated according to the following formula:
[0032] wherein L is the number of tasks completed by the user, Q 90 is the 90th percentile of the number of tasks completed by all users at the same time.
[0033] A reading website sometimes launches some activities to encourage users to stay active, such as giving gifts for completing a specified amount of tasks. At this time, some users may continuously complete tasks to obtain the gifts, and the number of tasks completed by these users is abnormally high, but these users are actually not active on the reading website and rarely log in. Due to the abnormally high number of tasks completed, these users are easily misclassified as high-active users. To avoid this situation, the embodiment scales the number of tasks completed by the user by the 90th percentile of the number of tasks completed by all users at the same time to determine the task score, which can effectively reduce the influence of extreme task completion of some users on the user stratification result, so that the stratification result is more accurate.
[0034] Y, determines the level to which the user belongs according to the effective active score, the login score and the task score of the user, and specifically, the level is positively correlated with the effective active score, the login score and the task score.
[0035] After the computer calculates the active ranking score M, the login score D and the task score R of the user, the level to which the user belongs is determined according to the active ranking score M, the login score D and the task score R of the user, and the specific process is as follows: First, the computer calculates the three values of the active ranking score M, the login score D and the task score R of the user by weighting according to the formula: S=0.5M+0.2D+0.3R to obtain the comprehensive score S of the user, then obtains the comprehensive score interval corresponding to each level, and divides the user into the level corresponding to the comprehensive score interval in which the comprehensive score S of the user is located. In this embodiment, five stratification levels of users are set according to the comprehensive score from high to low (the higher the level, the more active the user), which are as follows: 80≤the comprehensive score≤100, the user is a “benchmark user”; 60≤the comprehensive score≤80, the user is a “growth user”; 40≤the comprehensive score≤60, the user is a “seedling user”; 20≤the comprehensive score≤40, the user is a “sprout user”; 0≤the comprehensive score≤20, the user is a “seed user”.
[0036] The computer determines the level to which each user in the reading website belongs according to the above method, realizes dynamic stratification of all users in the reading website, and obtains the user stratification result of the reading website.
[0037] The above description is only an embodiment of the present invention, and does not limit the scope of patent protection. Those skilled in the art can make non-essential changes or substitutions on the basis of the present invention, and still fall within the scope of patent protection.
Claims
1. A method for dynamic user segmentation based on multiple data dimensions, characterized in that, Comprising the following steps: X, obtaining the active data, login times and task completion number of the user, and then executing the following steps S1-S3 in no particular order: -S1, calculating the effective active score of the user by using the sliding window dynamic attenuation method for the active data of the user, the more distant the time point of the active data, the smaller the weight coefficient when calculating the effective active score; -S2, determining the login score of the user according to the login times interval segmentation scoring rule, the higher the login times interval, the slower the login score growth rate corresponding thereto; -S3, determining the preset quantile reference according to the task completion number of all users in the same period, and then: if the task completion number of the user exceeds the preset quantile reference, setting the task score of the user to the highest value, if the task completion number of the user does not exceed the preset quantile reference, deriving the task score according to the ratio of the task completion number of the user to the preset quantile reference, the greater the task completion number of the user, the higher the task score; Y, determining the level to which the user belongs according to the effective active score, login score and task score of the user, specifically, the level level is positively correlated with the effective active score, login score and task score.
2. The method for multi-data dimension based user dynamic layering as claimed in claim 1, wherein the step of In S1, the effective active score C is calculated as follows: Wherein: J i Active data of the ith month; n is the cumulative active month number of the user; λ is the decay coefficient, with a value of 0.
2.
3. The method of claim 2, wherein the step of dynamically layering the users based on the multiple data dimensions further comprises the step of: In step S2, the login times interval segmentation scoring rule is as follows: Login times <5: login score D=login times x 5; 5≤login times≤15: login score D=25+(login times-5) x 4; 16≤login times<30: login score D=65+(login times-15) x 2; Login times≥30: login score D=100.
4. The method of claim 3, wherein the step of dynamically layering the users based on the multiple data dimensions further comprises the step of: In step S3, the preset quantile reference is 90% quantile, if the task completion number of the user exceeds 90% quantile of the task completion number of all users in the same period, the task score R of the user is set to 100, otherwise the task score R of the user is calculated according to the following formula: where L is the number of task completions by the user, Q 90 is the 90th percentile of the number of task completions by all users at the same time.
5. The user dynamic layering method based on multiple data dimensions according to claim 4, characterized in that: In step S1, the active ranking score M of the user is further determined according to the effective active score C of the user; In step Y, the level to which the user belongs is specifically determined according to the active ranking score M, login score D and task score R of the user.
6. The method for multi-dimensional data-based user dynamic layering as claimed in claim 5, wherein, The active ranking score M of the user is determined as follows: Wherein, P is the ranking of the effective active score of the user among all users, and N is the number of all users.
7. The method for multi-data dimension based user dynamic layering as claimed in claim 5 or 6, characterized in that, Step Y specifically: The active ranking score M, login score D and task score R of the user are weighted and calculated to obtain the comprehensive score S of the user; Obtain the comprehensive score interval corresponding to each level, and divide the user into the level corresponding to the comprehensive score interval in which the comprehensive score S of the user is located.
8. The method for multi-data dimension based user dynamic layering as claimed in claim 7, wherein, In step Y, the weighted calculation formula of the comprehensive score S of the user is S=0.5M+0.2D+0.3R.
9. A computer program product comprising a computer program, characterized in that, The computer program is executed to implement the user dynamic layering method based on multiple data dimensions according to any one of claims 1 to 8.