Mobile cloud disk introduction period user growth feature importance degree evaluation method and system based on score card model
Through the method based on the score card model, the growth characteristics of mobile cloud disk users are analyzed and evaluated, and the problems of huge data volume and labor-consuming manual analysis are solved, which improves user stickiness and maintenance efficiency, and achieves effective users during the introduction period Analysis and maintenance.
Patent Information
- Application Number
- CN202410875672.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-11-08
AI Technical Summary
In the prior art, the data for mobile cloud disk users' growth characteristics analysis and evaluation during the introduction period is huge. Manual analysis consumes a lot of manpower and is prone to errors, making it difficult to effectively improve user usage stickiness and reduce loss.
Using a scoring card model method, the user data is collected, variable binning, evidence weight transformation, variable screening and logistic regression model construction is constructed, standard scoring cards are generated, and the user growth characteristics is evaluated, which is used to analyze and optimize user growth characteristics. .
It has improved the operator's ability to understand the important characteristics of users and the preferences of promotion, improved the user's use stickiness during the introduction period, reduced user churn, simplified customer service maintenance methods, and improved customer stickiness.
Smart Images

Figure CN118917868A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mobile cloud disk user experience optimization, and in particular to a method and system for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model. Background Art
[0002] In recent years, with the popularization of mobile devices, changes in working methods, and the trend of explosive growth of data, the problems of traditional storage methods have gradually been exposed, and mobile cloud disk technology has been rapidly promoted. Cloud disks can better solve problems such as limited storage space and difficult data synchronization.
[0003] In the existing technology, there are many cloud disk products on the market, and promotion and marketing activities are also flourishing. How to improve the usage stickiness of users in the introduction period and reduce user churn in a highly competitive environment has become one of the important tasks for operators to maintain user activity. To prevent the loss of users in the introduction period, it is necessary to analyze the important features of such users entering the growth and maturity stages, extract the features with high importance, and further optimize the product quality in order to retain the users in the introduction period. The data volume of feature importance evaluation is huge, and manual collection and analysis requires a lot of manpower and is prone to errors.
[0004] Therefore, there is an urgent need to use big data mining technology to assist operators in analyzing the important characteristics of user growth during the introduction period of mobile cloud disks. Summary of the invention
[0005] The purpose of the present invention is to provide a method and system for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model, so as to solve the problems raised in the above-mentioned background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model, the method comprising the following steps:
[0007] S1. Collecting data from cloud disk users for building evaluation models;
[0008] S2. Analyze and process the data collected in step S1, and select feature items according to the needs of establishing a growth feature importance evaluation model;
[0009] S3, based on the data output by S2, performs model training, evaluation and optimization, and outputs a mobile cloud disk user growth feature evaluation model;
[0010] S4. Convert the scoring model obtained in S3 into a standard scoring card, and assess the feature importance score of the cloud disk user according to the standard scoring card.
[0011] Preferably, the data collected in step S1 includes basic user information, terminal information, tag attributes, asset information, package information, embedding behavior, and participation in marketing activities.
[0012] Preferably, step S3 of establishing an importance assessment model includes:
[0013] S31, variable binning, optimal binning of continuous variables based on CART algorithm;
[0014] S32, weight of evidence transformation;
[0015] S33, variable screening, univariate screening based on IV value, multivariate screening based on VIF value analysis of multicollinearity among multiple variables;
[0016] S34, build a logistic regression model, build a scoring model based on the logistic regression algorithm, and use the PSI value to evaluate the stability of the model, and use the KS value and AUC value to evaluate the ability of the model to distinguish good and bad samples;
[0017] S35. Model evaluation.
[0018] Preferably, after the feature variable binning is completed, S32 weight of evidence transformation WOE is performed, and the calculation formula is:
[0019]
[0020] Among them, Bad i is the number of bad samples in the ith bin, Bad T is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples;
[0021] After completing WOE coding, S33 variable selection is performed to select single variables based on information value IV. The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows:
[0022]
[0023] In the formula, N represents the number of variable bins. The variance inflation factor (VIF) is used to measure the severity of multilinearity of the selected variables. When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one. After the variable selection is completed, step S34 is performed to construct a logistic regression model. The logistic regression formula is:
[0024]
[0025] in w Tis the learning parameter, x is the variable after WOE conversion, and p is the bad sample probability, that is, the probability that the user in the introduction period is not converted to the mature growth period.
[0026] Preferably, in step SS4, the probability of a bad sample is expressed as p, and the probability of a good sample is 1-p. According to the basic principle of logistic regression, it can be obtained that:
[0027]
[0028] Define a ratio to express the relative probability of a customer converting:
[0029]
[0030] The score of the scorecard is defined as the linear expression of the logarithm of the ratio:
[0031] Score = AB × log (odds) = AB (w 0 +w 1 x 1 +L+w n x n )
[0032] A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion during the introduction period, the lower the score.
[0033] A user growth feature importance evaluation system for the introduction period of a mobile cloud disk based on a scoring card model, the system comprising a data acquisition module, a data analysis and processing module, a model training module, and a model conversion module;
[0034] Data collection module, which collects data from cloud disk users for building evaluation models;
[0035] The data analysis and processing module analyzes and processes the collected data and selects feature items according to the needs of establishing a growth feature importance assessment model;
[0036] Model training module: performs model training, evaluation and optimization based on the output data, and outputs a mobile cloud disk user growth feature evaluation model;
[0037] The model conversion module converts the acquired scoring model into a standard scoring card, and evaluates the feature importance score of the cloud disk user according to the standard scoring card.
[0038] Preferably, the data collected by the data collection module includes basic user information, terminal information, tag attributes, asset information, package information, embedding behavior, and participation in marketing activities.
[0039] Preferably, the model training module includes variable binning, optimal binning of continuous variables based on the CART algorithm; evidence weight transformation; variable screening, univariate screening based on IV value, multivariate screening based on VIF value analysis of multicollinearity between multiple variables; construction of a logistic regression model, construction of a scoring model based on the logistic regression algorithm, and use of the PSI value to evaluate the stability of the model, and use of the KS value and AUC value to evaluate the ability of the model to distinguish between good and bad samples; model evaluation.
[0040] Preferably, after the feature variable binning is completed, the model training module performs weight of evidence transformation WOE, and the calculation formula is:
[0041] Among them, Bad i is the number of bad samples in the ith bin, Bad T is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples;
[0042] After completing WOE coding, variable selection is performed to select single variables based on information value IV. The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows:
[0043]
[0044] In the formula, N represents the number of variable bins. The variance inflation factor (VIF) is used to measure the severity of the multilinearity of the selected variables. When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one. After the variable selection is completed, the logistic regression model is constructed. The logistic regression formula is:
[0045]
[0046] in w T is the learning parameter, x is the variable after WOE conversion, and p is the bad sample probability, that is, the probability that the user in the introduction period is not converted to the mature growth period.
[0047] Preferably, the model conversion module represents the probability of a bad sample as p, and the probability of a good sample is 1-p, which can be obtained from the basic principle of logistic regression:
[0048]
[0049] Define a ratio to express the relative probability of a customer converting:
[0050]
[0051] The score of the scorecard is defined as the linear expression of the logarithm of the ratio:
[0052] Score = AB × log (odds) = AB (w 0 +w 1 x 1 +L+w n x n )
[0053] A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion during the introduction period, the lower the score.
[0054] Compared with the prior art, the present invention has the following beneficial effects:
[0055] The present invention proposes a method and system for evaluating the importance of growth characteristics of users in the introduction period of a mobile cloud disk based on a scoring card model; by using the basic information, terminal information, tag attributes, asset information, package information, embedding behavior, and participation in marketing activities of cloud disk users, a scoring card is prepared with the help of historical data, and then the importance of the characteristics of the users in the scoring card is evaluated, which can provide a reference basis for operators to understand the important characteristics of users in the introduction period, the focus of promotion, etc., improve the usage stickiness of users in the introduction period, and maintain the number of users; compared with the traditional customer maintenance based on several key business indicators or complaint data, without adding additional equipment or collecting new data, based on the scoring card model and the characteristic distribution trend of users in the introduction period, the important characteristics of the conversion of users in the introduction period to users in the growth and mature period are predicted, which is convenient for the operation side to analyze in advance and carry out targeted customer maintenance measures, so as to conveniently and efficiently improve customer usage stickiness. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0057] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0058] Embodiment 1
[0059] See also Figure 1 The present invention provides a technical solution: a method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model, the method comprising the following steps:
[0060] S1. Collect cloud disk user basic information, terminal information, tag attributes, asset information, package information, embedding behavior, marketing activity participation and other data for building the evaluation model;
[0061] S2. Analyze and process the data collected in step S1, and select feature items according to the needs of establishing a growth feature importance evaluation model;
[0062] S3, based on the data output by S2, performs model training, evaluation and optimization, and outputs a mobile cloud disk user growth feature evaluation model;
[0063] S4. Convert the scoring model obtained in S3 into a standard scoring card, and assess the feature importance score of the cloud disk user according to the standard scoring card.
[0064] The present invention is a method for evaluating the importance of user features in the introduction period of a mobile cloud disk based on a scoring card model. Step S1 is to collect basic information of cloud disk users, terminal information, tag attributes, asset information, package information, embedding behavior, marketing activity participation, etc., for building an evaluation model, as follows:
[0065] Basic information: including age, gender, network access time, user star rating and other indicators;
[0066] Terminal information: including device model, operating system, system version, operator information, storage space, battery information, whether it is a 4 / 5G terminal, etc.
[0067] Tag attributes: including indicators such as life cycle and whether it is active;
[0068] Asset information: including storage space, number of files, number of images, average file size and other indicators;
[0069] Package information: including the name of the traffic package, package fee and other indicators;
[0070] Tracking behavior: including login operations, uploads, downloads, sharing and other indicators;
[0071] Participation in marketing activities: including space collection, speed-up times, birthday benefit collection and other indicators;
[0072] The present invention is a method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model. Step S2 is to analyze and process the data collected in step S1, and select feature items according to the requirements of establishing a growth feature importance evaluation model, specifically as follows:
[0073] This step involves two parts: model target selection and feature item selection. Specifically, for users who are converted from the introduction period to the growth period and maturity period, their feature importance during the survey period will have a certain trend, and there will be certain feature distribution differences from users who have not been converted to the growth period and maturity period. Considering the huge data volume and the inability to guarantee the timeliness of calculation, the selection and prediction of features are still based on the month. Accordingly, the analysis scope is defined based on the end of March, that is, the introduction period users with the 20230331 account period. The sample label is judged based on the life cycle of the life cycle table in April. Users who are converted to the growth and maturity period in April are marked as 1, and the rest are marked as 0. The model is mainly based on the label data of behavior and assets, combined with the usage of competing products and terminal status, the time, price, remaining storage space, receipt status, application authentication status and other data. Each indicator involved in S1 is summarized, and then the missing values of all features are counted. When the missing degree is high, it is eliminated and does not participate in the modeling.
[0074] The present invention is a method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model. Step S3 is to perform model training, evaluation and tuning based on the data output by S2, and output a feature importance evaluation model, which is specifically as follows:
[0075] This step of establishing the evaluation model includes: S31, variable binning; S32, evidence weight transformation; S33, variable screening; S34, building a logistic regression model; S35, model evaluation.
[0076] Among them, S31 variable binning is to discretize the continuous variables in the original data in sections, and merge the multi-state discrete variables to reduce the number of states of the discrete variables. This step can effectively handle the missing values and outliers in the feature variables, simplify the logistic regression model, reduce the risk of model overfitting, and improve the generalization ability of the model. The present invention is based on the CART (Classification and Regression Tree) algorithm to perform optimal binning of continuous variables.
[0077] After the feature variable is binned, the S32 Weight of Evidence (WOE) transformation is performed, which can represent the volume ratio of users who are converted from the introduction period to the mature growth period and those who are not converted in the current bin, and the difference between this ratio in all samples (randomly selected). The larger the WOE, the greater the difference, and the greater the possibility that the users in the introduction period in the current group have not been converted to the mature growth period. Conversely, the smaller the possibility that the users in the current group have not been converted to the mature growth period. The WOE calculation formula is:
[0078]
[0079] Bad i is the number of bad samples in the ith bin, BadT is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples.
[0080] The initially selected variables may have a low correlation with the target variable. In order to ensure the modeling effect, after completing the WOE coding, S33 variable selection is required. To this end, this scheme selects single variables based on information value (IV). The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows:
[0081]
[0082] In the formula, N represents the number of variable bins. In addition to single variable screening, in order to ensure the interpretability of variables when the logistic regression model is converted into a standard scorecard, this solution measures the severity of multilinearity of the selected variables based on the variance inflation factor (VIF). When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one.
[0083] After the variable selection is completed, step S34 needs to be performed to build a logistic regression model. The logistic regression formula is:
[0084]
[0085] in w T is the learning parameter, x is each variable (after WOE conversion), and p is the probability of a bad sample, that is, the probability that the user in the introduction period is not converted to the growth and maturity period. Based on the preprocessed data, grid search and cross-validation are used to obtain the optimal model. In order to ensure the classification effect of the model, model evaluation is also required. This solution selects the AUC (Area Under Curve) that represents the probability that the output result of the model for any bad sample is greater than the output result of the model for any good sample, the KS value (Kolmogorov-Smirnov) that represents the ability of the model to distinguish between good and bad samples, and the PSI value (population stability index) that can represent the stability of the model.
[0086] The present invention is a method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model. The step S4 is to convert the training model into a standard scoring card, and evaluate the importance of user growth features according to the standard scoring card, as follows:
[0087] The probability of a bad sample is expressed as p, and the probability of a good sample is 1-p. According to the basic principle of logistic regression, we can get:
[0088]
[0089] Ratios are defined here to represent the relative probability of a customer converting or not:
[0090]
[0091] The score of the scorecard is defined as the linear expression of the logarithm of the ratio:
[0092] Score = AB × log (odds) = AB (w 0 +w 1 x 1 +L+w n x n )
[0093] A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion in the introduction period, the lower the score. In the present invention, a high score represents users who have been converted from the introduction period to the mature growth period, and a low score represents unconverted users. A and B are unknown in the formula, and can be calculated by substituting two known or assumed scores into the calculation: the expected score under a certain dissatisfaction probability, odds is w 0 The score P 0 ; The score of the doubled probability of dissatisfaction (PDO). The final format is shown in the following table:
[0094]
[0095]
[0096] The scorecard parameters selected by the present invention are: based on the Xgboost4j library, Spark is used to directly run the xgboost model in the cluster for training, where the ratio of the training set to the test set is 8:2, and the trained model is stored in the hive table. After the scorecard is established, the operation staff can perform special maintenance on the features with high importance. When there is a certain maintenance volume, the special maintenance of the features with high importance can improve the usage stickiness of the users in the introduction period to a certain extent.
[0097] Embodiment 2
[0098] Based on the first embodiment, a system for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model is proposed. The system consists of a data acquisition module, a data analysis and processing module, a model training module, and a model conversion module.
[0099] The data collection module collects data from cloud disk users for building evaluation models; the collected data includes basic user information, terminal information, tag attributes, asset information, package information, embedding behavior, and participation in marketing activities.
[0100] The data analysis and processing module analyzes and processes the collected data and selects feature items according to the needs of establishing a growth feature importance assessment model;
[0101] The model training module performs model training, evaluation and optimization based on the output data, and outputs a mobile cloud disk user growth feature evaluation model; including variable binning, optimal binning of continuous variables based on the CART algorithm; evidence weight transformation; variable screening, univariate screening based on IV values, and multivariate screening based on VIF value analysis of multicollinearity between multiple variables; building a logistic regression model, building a scoring model based on the logistic regression algorithm, and using the PSI value to evaluate the stability of the model, and using the KS value and AUC value to evaluate the model's ability to distinguish between good and bad samples; model evaluation.
[0102] After the feature variables are binned, the weight of evidence transformation WOE is performed, and the calculation formula is:
[0103]
[0104] Among them, Bad i is the number of bad samples in the ith bin, Bad T is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples;
[0105] After completing WOE coding, variable selection is performed to select single variables based on information value IV. The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows:
[0106]
[0107] In the formula, N represents the number of variable bins. The variance inflation factor (VIF) is used to measure the severity of the multilinearity of the selected variables. When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one. After the variable selection is completed, the logistic regression model is constructed. The logistic regression formula is:
[0108]
[0109] in w T is the learning parameter, x is the variable after WOE conversion, and p is the bad sample probability, that is, the probability that the user in the introduction period is not converted to the mature growth period.
[0110] The model conversion module converts the acquired scoring model into a standard scoring card, and evaluates the feature importance score of the cloud disk user according to the standard scoring card.
[0111] The probability of a bad sample is expressed as p, and the probability of a good sample is 1-p. According to the basic principle of logistic regression, we can get:
[0112]
[0113] Define a ratio to express the relative probability of a customer converting:
[0114]
[0115] The score of the scorecard is defined as the linear expression of the logarithm of the ratio:
[0116] Score = AB × log (odds) = AB (w 0 +w 1 x 1 +L+w n x n )
[0117] A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion during the introduction period, the lower the score.
[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model, characterized in that: The method comprises the following steps: S1. Collecting data from cloud disk users for building evaluation models; S2. Analyze and process the data collected in step S1, and select feature items according to the needs of establishing a growth feature importance evaluation model; S3, based on the data output by S2, performs model training, evaluation and optimization, and outputs a mobile cloud disk user growth feature evaluation model; S4. Convert the scoring model obtained in S3 into a standard scoring card, and assess the feature importance score of the cloud disk user according to the standard scoring card.
2. According to claim 1, a method for evaluating the importance of user growth characteristics during the introduction period of a mobile cloud disk based on a scoring card model, characterized in that: The data collected in step S1 includes basic user information, terminal information, tag attributes, asset information, package information, embedding behavior, and marketing activity participation.
3. According to claim 1, a method for evaluating the importance of user growth characteristics during the introduction period of a mobile cloud disk based on a scoring card model, characterized in that: Step S3 of establishing an importance evaluation model includes: S31, variable binning, optimal binning of continuous variables based on CART algorithm; S32, weight of evidence transformation; S33, variable screening, univariate screening based on IV value, multivariate screening based on VIF value analysis of multicollinearity among multiple variables; S34, build a logistic regression model, build a scoring model based on the logistic regression algorithm, and use the PSI value to evaluate the stability of the model, and use the KS value and AUC value to evaluate the ability of the model to distinguish good and bad samples; S35. Model evaluation.
4. According to claim 3, a method for evaluating the importance of user growth characteristics during the introduction period of a mobile cloud disk based on a scoring card model is characterized by: After the characteristic variables are binned, the S32 weight of evidence transformation WOE is performed, and the calculation formula is: Among them, Bad i is the number of bad samples in the i-th bin, Bad T is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples; After completing WOE coding, S33 variable selection is performed to select single variables based on information value IV. The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows: In the formula, N represents the number of variable bins. The variance inflation factor (VIF) is used to measure the severity of multilinearity of the selected variables. When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one. After the variable selection is completed, step S34 is performed to construct a logistic regression model. The logistic regression formula is: in w T is the learning parameter, x is the variable after WOE conversion, and p is the bad sample probability, that is, the probability that the user in the introduction period is not converted to the mature growth period.
5. According to claim 1, a method for evaluating the importance of user growth characteristics during the introduction period of a mobile cloud disk based on a scoring card model, characterized in that: In step SS4, the probability of a bad sample is expressed as p, and the probability of a good sample is 1-p. According to the basic principle of logistic regression, we can get: Define a ratio to express the relative probability of a customer converting: The score of the scorecard is defined as the linear expression of the logarithm of the ratio: Score=A-B×log(odds)=A-B(w0+w1x1+L+w n x n ) A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion during the introduction period, the lower the score.
6. A system for evaluating the importance of user growth features during the introduction period of a mobile cloud disk based on a scoring card model according to any one of claims 1 to 5, characterized in that: The system consists of a data acquisition module, a data analysis and processing module, a model training module and a model conversion module; Data collection module, which collects data from cloud disk users for building evaluation models; The data analysis and processing module analyzes and processes the collected data and selects feature items according to the needs of establishing a growth feature importance assessment model; Model training module: performs model training, evaluation and optimization based on the output data, and outputs a mobile cloud disk user growth feature evaluation model; The model conversion module converts the acquired scoring model into a standard scoring card, and evaluates the feature importance score of the cloud disk user according to the standard scoring card.
7. According to claim 6, a user growth feature importance evaluation system based on a scoring card model during the introduction period of a mobile cloud disk, characterized in that: The data collection module collects data including user basic information, terminal information, tag attributes, asset information, package information, embedding behavior, and marketing activity participation.
8. According to claim 6, a user growth feature importance evaluation system based on a scoring card model during the introduction period of a mobile cloud disk, characterized in that: The model training module includes variable binning, optimal binning of continuous variables based on the CART algorithm; evidence weight transformation; variable screening, univariate screening based on IV value, and multivariate screening based on VIF value analysis of multicollinearity between multiple variables; building a logistic regression model, building a scoring model based on the logistic regression algorithm, and using the PSI value to evaluate the stability of the model, and using the KS value and AUC value to evaluate the model's ability to distinguish between good and bad samples; model evaluation.
9. According to claim 6, a user growth feature importance evaluation system based on a scoring card model during the introduction period of a mobile cloud disk, characterized in that: In the model training module, after the feature variable binning is completed, the weight of evidence transformation WOE is performed, and the calculation formula is: Among them, Bad i is the number of bad samples in the i-th bin, Bad T is the total number of bad samples, Good i is the number of good samples in the i-th bin, Good T is the total number of good samples; After completing WOE coding, variable selection is performed to select single variables based on information value IV. The larger the IV value, the stronger the predictive ability of the independent variable. The calculation formula is as follows: In the formula, N represents the number of variable bins. The variance inflation factor (VIF) is used to measure the severity of the multilinearity of the selected variables. When the VIF of a variable is greater than the threshold, the explanatory variables are eliminated one by one. After the variable selection is completed, the logistic regression model is constructed. The logistic regression formula is: in w T is the learning parameter, x is the variable after WOE conversion, and p is the bad sample probability, that is, the probability that the user in the introduction period is not converted to the mature growth period.
10. According to claim 6, a user growth feature importance evaluation system based on a scoring card model during the introduction period of a mobile cloud disk, characterized in that: The model conversion module represents the probability of a bad sample as p, and the probability of a good sample is 1-p. According to the basic principle of logistic regression, we can get: Define a ratio to express the relative probability of a customer converting: The score of the scorecard is defined as the linear expression of the logarithm of the ratio: Score=A-B×log(odds)=A-B(w0+w1x1+L+w n x n ) A and B are constants. The negative sign in front of B represents that the higher the degree of non-conversion during the introduction period, the lower the score.
