CN113627653B - Method and device for determining activity prediction strategy of mobile banking user - Google Patents

Method and device for determining activity prediction strategy of mobile banking user Download PDF

Info

Publication number
CN113627653B
CN113627653B CN202110798024.6A CN202110798024A CN113627653B CN 113627653 B CN113627653 B CN 113627653B CN 202110798024 A CN202110798024 A CN 202110798024A CN 113627653 B CN113627653 B CN 113627653B
Authority
CN
China
Prior art keywords
user
activity
mobile phone
model
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110798024.6A
Other languages
Chinese (zh)
Other versions
CN113627653A (en
Inventor
李阳强
肖慧英
陈嘉敏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Suoxinda Data Technology Co ltd
Original Assignee
Shenzhen Suoxinda Data Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Suoxinda Data Technology Co ltd filed Critical Shenzhen Suoxinda Data Technology Co ltd
Priority to CN202110798024.6A priority Critical patent/CN113627653B/en
Publication of CN113627653A publication Critical patent/CN113627653A/en
Application granted granted Critical
Publication of CN113627653B publication Critical patent/CN113627653B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/02—Banking, e.g. interest calculation or account maintenance

Abstract

The embodiment of the invention discloses a method and a device for determining a mobile banking user activity prediction strategy, wherein the method comprises the following steps: equidistant grouping is carried out on the target sequences obtained by sequencing the D mobile banking users from high to low according to the first user activity according to a preset K grouping method, so that a plurality of user groups are obtained; based on a preset LightGBM algorithm, the first isTraining and verifying an activity prediction model by using data of mobile banking users contained in each user group to obtain a trained first stepA liveness prediction model; determining the model comprehensive value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the use data of the mobile banking users in the nth group; and taking the activity prediction model of a plurality of user groups corresponding to the target grouping method of the maximum model comprehensive value degree and the classification model as target prediction strategies to predict the activity of the mobile banking users.

Description

Method and device for determining activity prediction strategy of mobile banking user
Technical Field
The invention relates to the technical field of liveness prediction, in particular to a method and a device for determining liveness prediction strategies of mobile banking users.
Background
The activity prediction of a mobile banking user is a classification problem, namely, predicting whether the mobile banking user will be active. The two-classification model is a model commonly used in the field of intelligent recognition of machine learning. In the two classification models, the LightGBM is an algorithm with a good prediction effect, is an integrated learning type, is realized efficiently through gradient lifting decision trees, and has the characteristics of high prediction precision and high running speed.
In predicting a classification model by using the LightGBM algorithm, in order to improve the accuracy of model prediction, the following methods are generally used in the industry: firstly, optimizing through algorithm parameters, such as limiting the depth, the leaf number or the number of trees and the like of each tree; and secondly, adding sample characteristics with stronger predictive force in the process of characteristic engineering. The prediction precision in application of the two classification models trained by the LightGBM algorithm based on the method can be effectively improved. However, the first method requires continuous adjustment of the model structure, which is time consuming and energy consuming, and the second method relies on the sample characteristics of the collected samples, which are predicted to be strong during sample collection, and thus lacks a method for improving the accuracy of user activity prediction.
Disclosure of Invention
The invention mainly aims to provide a method, a device, computer equipment and a storage medium for determining an activity prediction strategy of a mobile banking user, which can solve the problem that a method for improving the accuracy of user activity prediction is lacking in the prior art.
In order to achieve the above object, a first aspect of the present invention provides a method for determining an activity prediction policy of a mobile banking user, the method comprising:
collecting using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to the sequence from high to low by utilizing the first user activity of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
equidistant grouping is carried out on the user sequences according to preset K grouping methods and the first user liveness, so that a plurality of user groups respectively corresponding to the K grouping methods are obtained, and the number of the user groups divided by the K grouping methods is different;
Based on a preset LightGBM algorithm, the first isTraining and verifying an activity prediction model by using data of mobile banking users contained in the individual user groups to obtain the trained +.>A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer,/and a plurality of groups are selected from the group consisting of>Represents the ith in the nth group;
determining the model comprehensive value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the use data of the mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
and determining a grouping method with the maximum model comprehensive value degree in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity degree of a mobile phone bank to be predicted.
In order to achieve the above object, a second aspect of the present invention provides a method for predicting activity of a mobile banking user, the method comprising:
acquiring use data of a mobile phone banking user to be predicted on the mobile phone banking, wherein the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of the mobile phone banking user to be predicted after desensitization;
inputting the usage data into a classification model included in a target prediction strategy, and determining second user liveness corresponding to the mobile banking user to be predicted, wherein the target prediction strategy is determined by using the method for determining the liveness prediction strategy of the mobile banking user according to the first aspect;
determining a target liveness prediction model according to the second user liveness and liveness prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
inputting the usage data into the target liveness prediction model, and determining the third liveness corresponding to the mobile banking user to be predicted;
and carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
In order to achieve the above object, a third aspect of the present invention provides a device for determining an activity prediction policy of a mobile banking user, the device including:
the characteristic acquisition module is used for: the method comprises the steps of acquiring using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to a sequence from high to low by utilizing first user liveness of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
and a user grouping module: the user sequence is equidistantly grouped according to preset K grouping methods and the first user activity level to obtain a plurality of user groups corresponding to the K grouping methods respectively, and the number of the user groups divided by the K grouping methods is different;
model training module: for basing the preset LightGBM algorithm on the firstTraining and verifying an activity prediction model by using data of mobile banking users contained in the individual user groups to obtain the trained +. >A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer,/and a plurality of groups are selected from the group consisting of>Represents the ith in the nth group;
the value calculation module: the method comprises the steps of determining the model comprehensive value of an nth group according to the model value of an activity prediction model corresponding to each user group in the nth group and the use data of mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
policy access module: the method is used for determining a grouping method with the largest model comprehensive value in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity of a mobile phone bank to be predicted.
In order to achieve the above object, a fourth aspect of the present invention provides a device for predicting activity of a mobile banking user, the device comprising:
And a data acquisition module: the method comprises the steps of acquiring use data of a mobile phone bank user to be predicted on the mobile phone bank, wherein the use data at least comprise basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the mobile phone bank user to be predicted after desensitization;
a first prediction module: the method is used for inputting the usage data into a classification model included in a target prediction strategy, determining the activity of a second user corresponding to the mobile banking user to be predicted, wherein the target prediction strategy is determined by using the method for determining the activity prediction strategy of the mobile banking user according to any one of the first aspect;
model determination module: the target activity prediction model is determined according to the second user activity and activity prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
a second prediction module: the target activity degree prediction model is used for inputting the use data to determine the third user activity degree corresponding to the mobile banking user to be predicted;
probability determination module: and the average value calculation is used for carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
To achieve the above object, a fifth aspect of the present invention provides a computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps as set forth in any one of the first or second aspects.
To achieve the above object, a sixth aspect of the present invention provides a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps as in either the first or the second aspect.
The embodiment of the invention has the following beneficial effects:
the invention provides a method and a device for determining a mobile banking user activity prediction strategy, wherein the method comprises the following steps: predicting the use data of each mobile banking user based on the classification model to obtain the first user activity of each mobile banking user; the first user activity is sequenced from high to low to obtain a user sequence formed by each mobile banking user; equidistant grouping of the preset K grouping methods is carried out on the user sequence, and a plurality of user groups corresponding to the grouping methods are obtained; and training and verifying the liveness prediction model based on a plurality of user groups corresponding to each grouping method, selecting a target grouping method corresponding to the maximum model comprehensive value of all liveness prediction models in each grouping method, and taking the target grouping method, each liveness prediction model corresponding to the target grouping method and the two classification models as target prediction strategies to predict the user liveness of the mobile banking user to be predicted. Firstly, the model parameters do not need to be continuously adjusted, time and energy are reduced, secondly, sample characteristics with strong prediction force do not need to be collected when the samples are collected, the accuracy of the user activity of actual prediction can be improved by adopting the target prediction strategy, and the model prediction is stable.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Wherein:
fig. 1 is a flow chart of a method for determining an activity prediction policy of a mobile banking user in an embodiment of the present invention;
fig. 2 is another flow chart of a method for determining an activity prediction policy of a mobile banking user according to an embodiment of the present invention;
FIG. 3 is a flow chart of a method for predicting the activity of a mobile banking user in an embodiment of the invention;
fig. 4 is a block diagram of a device for determining an activity prediction policy of a mobile banking user according to an embodiment of the present invention;
FIG. 5 is a block diagram of a device for predicting the activity of a mobile banking user in an embodiment of the invention;
fig. 6 is a block diagram of a computer device in an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, fig. 1 is a flowchart illustrating a method for determining an activity prediction policy of a mobile banking user according to an embodiment of the present invention. The method shown in fig. 1 comprises the following steps:
101. collecting using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to the sequence from high to low by utilizing the first user activity of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
in the embodiment of the invention, the collected usage data of the mobile banking user on the mobile banking is subjected to activity prediction of a classification model to obtain the first user activity indicating the activity degree of the mobile banking user, and the first user activity is sequenced from high to low to finish sequencing of the mobile banking users to determine the user sequence.
The collection mode may be, for example, reading historical data of a server side of a mobile banking, where the collection frequency may be that only usage data of a mobile banking user in a certain period is obtained or usage data in all periods is obtained, and according to different activity prediction purposes, different data collection modes and data statistics time may be selected, which is not limited herein. The range of the mobile banking users can be all mobile banking users who open accounts in the mobile banking.
In one possible implementation, the usage data includes, but is not limited to, usage footprint data of a mobile banking user at a mobile banking, such as: the method comprises the steps of carrying out desensitization on basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of a mobile phone banking user to obtain a data set, wherein the desensitization refers to filtering or shielding sensitive information of the mobile phone banking user from information related to privacy parts of the mobile phone banking user. Further, the basic attribute features included in the usage data include, but are not limited to, personal basic condition data of the mobile banking user after desensitization of gender, age, academic, occupation, etc.; the asset status characteristics included in the usage data include, but are not limited to, mobile banking user personal asset data after desensitization of month-average and day-average assets, demand deposit balance, regular deposit balance, fund balance, financial product balance, insurance balance, national debt balance, and the like; the fund flow characteristics included in the use data comprise the total amount of money flowing in and total amount of money flowing out, total number of times of flowing in and out of money flowing out of a deposit card and transaction flow record data of mobile banking users such as various fund items or financial products in the period of about 1, 3, 6 and 12 months, and further, the fund items comprise fund incoming and outgoing items of consumption, financial products, funds, insurance, transfer accounts, third party payment, payouts, payments and the like of the mobile banking users; the mobile phone banking use characteristics included in the use data comprise the total amount and total times of mobile phone banking transactions of mobile phone banking users through mobile phone banking transactions and the amount and times of each fund item in the time of nearly 1, 3, 6 and 12 months. In addition, the use data also comprises the number of months before the user of the mobile phone bank opens an account, the number of days before the mobile phone bank is active, and the recent active characteristics of the mobile phone bank are included. The financial product features of the bank opening included in the usage data comprise short message notification of the mobile phone banking user opening at the bank, mobile phone banking opening, fund opening, financial opening, insurance opening, national debt opening, credit card opening, purchase opening and other business products. And counting the use data of all the account opening users, so that the use footprint of each mobile phone banking user on the mobile phone banking is realized, and the use data of each mobile phone banking user is acquired.
Taking the recent active feature of the mobile phone bank as an example, the calculation mode of the recent active feature of the mobile phone bank can be determined by the following calculation modes:
recent active feature of mobile banking= 8*M 1 +4*M 2 +2*M 3 +1*M 4 ，
Wherein the method comprises the steps of
The user sequence is an ordered queue, and is obtained by ordering the D mobile banking users from high to low based on the first user activity of the D mobile banking users from high to low, and it can be understood that the ordering is performed on the mobile banking users, and meanwhile, the use data corresponding to each mobile banking user can be changed correspondingly, so that the user sequence comprises the first activity probability of the D mobile banking users arranged from high to low according to the first activity and the use data corresponding to each mobile banking user. Further, D represents the number of mobile banking users, and D is a positive integer.
The two classification models are used for predicting probability values corresponding to the user liveness, and further obtaining first user liveness of each user in the user sequence. The first user activity of each mobile banking user can be obtained by inputting the usage data into the two classification models, and it can be understood that the first user activity can be used for reflecting the activity degree of the mobile banking user on the mobile banking.
102. Equidistant grouping is carried out on the user sequences according to preset K grouping methods and the first user liveness, so that a plurality of user groups respectively corresponding to the K grouping methods are obtained, and the number of the user groups divided by the K grouping methods is different;
in the embodiment of the invention, the grouping method indicates how many user groups are divided into mobile banking users, wherein K grouping methods represent K (i.e. multiple) grouping methods, the value of K is greater than or equal to 2, further, if the value of K is 2, it indicates that two grouping methods exist at present, and if the value of K is 3, it indicates that three grouping methods exist at present, and the number of user groups of each grouping method is different, in the two grouping methods, the number of grouping method 1 is 10 groups, and the number of user groups of grouping method 2 is 12 groups (i.e. the number of groups of user groups corresponding to each grouping method is different).
Further, the number of users of each user group included in the same grouping method is also different, and the user groups of each grouping method are divided based on equidistant grouping, so that the variation of the flag values of each user group included in the same grouping method is limited to the same range, the flag values of each group can be understood as the first activity upper limit value and the first activity lower limit value in each user group, and thus the variation range of the flag values of each user group, that is, the variation range between the first activity upper limit value and the first activity lower limit value is the same, and therefore the number of mobile banking users of the user groups equally divided by the K grouping methods is not necessarily the same, but the variation range of the user activity probability critical value of each user group included in the same grouping method is consistent.
103. Based on a preset LightGBM algorithm, the first isTraining and verifying an activity prediction model by using data of mobile banking users contained in the individual user groups to obtain the trained +.>N is the group corresponding to the grouping method KOther number, n is a positive integer, +.>Represents the ith in the nth group;
in the embodiment of the invention, based on a preset LightGBM algorithm, the use data of each user group is input into a corresponding liveness prediction model for training and verification until the trained first is obtainedAnd (5) an liveness prediction model.
For example, if k=2, the preset K grouping methods are grouping method 1 and grouping method 2, and if the corresponding preset group numbers of the user groups are 2 and 3, the grouping method 1=2 corresponds to P 2 1 P 2 2 Two user groups; grouping method 2=3 presence P 3 1 、P 3 2 P 3 3 Three user groups; and then need to add the P 2 1 P. Th 2 2 P. Th 3 1 P. Th 3 2 P (th) 3 3 And inputting corresponding activity prediction models by using data corresponding to the user groups for training and verification. Thereby obtaining the P 2 1 P. Th 2 2 P. Th 3 1 P. Th 3 2 P (th) 3 3 Five corresponding liveness prediction models, and grouping method 1=2 includes P < th ] 2 1 P. Th 2 2 Two activity prediction models formed by the two models; grouping method 2=3 includes P 3 1 P. Th 3 2 P (th) 3 3 Three liveness prediction models are formed.
It is understood that the usage data of D mobile banking users on the mobile banking may further include a first tag that is active or a second tag that is inactive. Based on a preset LightGBM algorithm, training and verifying the liveness model of user data corresponding to each user group respectively, and further obtaining each liveness prediction model after training, and further training the model: one or more of the usage data is used as input and then a predicted value (y ') of the user's activity level is returned as output, and the purpose of model training is to obtain the best parameters of the calculation model by using the data so as to achieve that the predicted value of the model is as close to the actual real value as possible. Model verification: the method is to predict new use by using a trained model, and compare the predicted result with the predicted result of the model in training to evaluate whether the model is reliable. Further verification may be performed by sample cross-validation, and the like, as examples and are not limited thereto.
For example, if k=4, the preset K grouping methods are grouping method 1, grouping method 2, grouping method 3 and grouping method 4, and if the number of the corresponding preset user groups is 4, 5, 6 and 7, the number of the corresponding user groups is 4, 5, 6 and 7, respectively, the four equidistant divisions of the mobile banking users D according to the number of the corresponding user groups of each grouping method 1, grouping method 2, grouping method 3 and grouping method 4 are obtained, so that the grouping method 1 corresponds to 4 user groups, the grouping method 2 corresponds to 5 user groups, and so on. Further, 22 user groups are obtained at this time, and the usage data included in the 22 user groups are respectively input into the corresponding liveness prediction models to train and verify the liveness prediction models. I.e. the firstThe user data of the mobile banking user contained in the individual user group is input as an input variable to the user data input unit>Training the activity prediction model corresponding to each user group, and carrying out +.>User active labels or user inactive labels corresponding to mobile banking users contained in individual user groups are used as target variables, training is performed based on a preset LightGBM algorithm, andand after training, verifying the trained liveness prediction model by utilizing each user group. Finally, 22 trained liveness prediction models corresponding to the user groups are obtained.
104. Determining the model comprehensive value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the use data of the mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
the model value refers to a classification model evaluation index auc (Area Under Curve), and the prediction performance of the liveness prediction model is reflected by auc.
For example, if auc =1, the liveness prediction model is a perfect classifier; if auc = [0.85,0.95], the liveness prediction model is a classifier with good prediction effect; if auc = [0.7,0.85], the liveness prediction model is a classifier with a general prediction effect; by analogy, if auc =0.5, the predictive effect of the liveness predictive model is similar to a random guess (e.g., copper loss), i.e., the model has no predictive value. When auc <0.5, the prediction effect is worse than random guess.
The model comprehensive value degree refers to a comprehensive evaluation index of a plurality of classification models. In the embodiment of the present invention, the model integrated value is the model integrated value of the nth group, that is, in each of When n takes the same value, the corresponding individual +.>To->Individual user group corresponding individual->To->The comprehensive evaluation indexes of n models of the individual liveness prediction models (namely, the comprehensive value degree of each liveness prediction model corresponding to each user group included in the same grouping method). For example, if the preset K grouping methods are 4, n is respectively valued according to the group numbers corresponding to the grouping method 1, the grouping method 2, the grouping method 3 and the grouping method 4, so as to obtain the model comprehensive value of the nth group.
105. And determining a grouping method with the maximum model comprehensive value degree in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity degree of a mobile phone bank to be predicted.
The target grouping method is a grouping method corresponding to the maximum value of the model comprehensive value in the model comprehensive value, and the grouping method K corresponding to the maximum value is selected as the target grouping method. And further obtaining a target prediction strategy, wherein the target prediction strategy comprises a classification model, a target grouping method and activity prediction models of a plurality of user groups corresponding to the target grouping method.
The invention provides a method for determining a mobile banking user activity prediction strategy, which comprises the following steps: predicting the use data of each mobile banking user based on the classification model to obtain the first user activity of each mobile banking user; the first user activity is sequenced from high to low to obtain a user sequence formed by each mobile banking user; equidistant grouping of the preset K grouping methods is carried out on the user sequence, and a plurality of user groups corresponding to the grouping methods are obtained; and training and verifying the liveness prediction model based on a plurality of user groups corresponding to each grouping method, selecting a target grouping method corresponding to the maximum model comprehensive value of all liveness prediction models in each grouping method, and taking the target grouping method, each liveness prediction model corresponding to the target grouping method and the two classification models as target prediction strategies to predict the user liveness of the mobile banking user to be predicted. Firstly, the model parameters do not need to be continuously adjusted, time and energy are reduced, secondly, sample characteristics with strong prediction force do not need to be collected when the samples are collected, the accuracy of the user activity of actual prediction can be improved by adopting the target prediction strategy, and the model prediction is stable.
Referring to fig. 2, fig. 2 is another flow chart of a method for determining an activity prediction policy of a mobile banking user according to an embodiment of the present invention. The method as shown in fig. 2 comprises the following steps:
201. obtaining using sample data of D mobile phone bank sample users on the mobile phone banks and user active labels corresponding to the mobile phone bank sample users, wherein the using sample data at least comprises basic attribute characteristics, asset condition characteristics, fund running line characteristics, mobile phone bank using characteristics and opened financial product characteristics of the mobile phone bank sample users after desensitization, and the user active labels comprise first labels corresponding to active users or second labels corresponding to inactive users;
it should be noted that, the step 201 is similar to the step 101, and the repeated content is not described herein, and reference may be made to the content shown in the foregoing step 101.
It should be noted that, the data collection dimensions of the usage sample data corresponding to the mobile phone banking sample user and the usage data corresponding to the mobile phone banking user are the same, including, but not limited to, the basic attribute data, the asset status data, the funds flow data, the mobile phone banking usage data, the opened financial product data, and the like, which are desensitized, corresponding to each mobile phone banking sample user, and the usage data can be specifically expressed in step 101.
202. Based on a preset LightGBM algorithm, performing iterative training on the two classification models by using the sample data of the D mobile phone bank sample users on the mobile phone banks until the two classification models are determined to converge according to the user active labels corresponding to the D mobile phone bank sample users;
it should be noted that, the content of step 201 is similar to that of step 103, and the repeated content will not be described herein, and reference may be made to the content shown in step 103.
It can be understood that by performing iterative training and verification on the classification model by using a mobile banking sample user carrying an activity label, until the output activity prediction result is the same as or approaches to the first label or the second label corresponding to the user activity label, further determining that the classification model converges, wherein the convergence can be understood that the influence of parameter optimization on the prediction result is less, the accuracy of the prediction result cannot be further improved again, further obtaining the classification model capable of realizing activity prediction, and further predicting the first user activity of the mobile banking user.
203. Inputting the use data included by the D mobile banking users into a trained two-class model, and determining the first user activity corresponding to each mobile banking user;
The number of the first user activity probabilities is the same as the number of the mobile banking users and is used for indicating the activity degree of each mobile banking user using the mobile banking, and it can be understood that the user activities among different users are different, so that the first user activities of each mobile banking user are different.
204. Collecting using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to the sequence from high to low by utilizing the first user activity of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
205. equidistant grouping is carried out on the user sequences according to preset K grouping methods and the first user liveness, so that a plurality of user groups respectively corresponding to the K grouping methods are obtained, and the number of the user groups divided by the K grouping methods is different;
206. Based on a preset LightGBM algorithm, the first isTraining and verifying an activity prediction model by using data of mobile banking users contained in the individual user groups to obtain the trained +.>A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer,/and a plurality of groups are selected from the group consisting of>Represents the ith in the nth group;
207. determining the model comprehensive value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the use data of the mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
it should be noted that steps 204 to 207 shown in fig. 2 are similar to steps 101 to 104 shown in fig. 1, and reference should be made to the foregoing steps 101 to 104 shown in fig. 1 for avoiding repetition of the description.
Wherein, step 207 further comprises: by using the firstDetermining the +.f. of each mobile banking user using data and a preset model value algorithm contained in each user group >And the model value degree corresponding to each liveness prediction model.
The preset model value algorithm is used for calculating the model value of each liveness prediction model. By way of example, by separately combining the respective ones ofThe method comprises the steps of sorting first user liveness corresponding to using data of each mobile phone banking user on the mobile phone banking, which is included in each user group, according to a sequence from high to low, halving the using data of each arranged D mobile phone banking users on the mobile phone banking to obtain two user groups, taking one V user group as active sample data of V=D/2 mobile phone banking users on the mobile phone banking, taking the other H user group as inactive sample data of H=D/2 mobile phone banking users on the mobile phone banking, arranging and combining the two groups to obtain H×V arrangement and combination groups, accumulating the number Q of the first user liveness corresponding to the active sample data in each arrangement and combination group to be larger than the number Q of the first user liveness corresponding to the inactive sample data, and finally passing through auc i Determining each model value auc =q/(h×v) i 。
Further, step 207 specifically includes:
A. the number Di of the mobile phone banking users, the total number D of the mobile phone banking users and the model value of each liveness prediction model corresponding to each user group are obtained;
B. And determining the model comprehensive value of the nth group by using the number Di of mobile banking users included in each user group corresponding to the nth group, the model value of the activity prediction model corresponding to each user group in the nth group and the total number D of mobile banking users.
The model comprehensive value of the nth group according to the different values of K is also different, and when the values of K are multiple, the model comprehensive value of the nth group corresponding to K can be obtained. And respectively calculating the model comprehensive value degree of each liveness prediction model corresponding to the grouping method K to obtain the model comprehensive value degree of the nth group.
Furthermore, the model comprehensive value, that is, the comprehensive evaluation index, can be obtained through a preset model comprehensive value algorithm, and the preset model comprehensive value algorithm formula can be expressed as follows:
wherein, K represents a grouping method, n represents the number of groups corresponding to the grouping method K, and i represents the group of each user group in n user groups in the grouping method K; auc i (i e {1,2,., n }) is the model value of the liveness prediction model corresponding to the i-th user group; d (D) i (i e {1,2,., n }) represents the number of users in the i-th group, the n value of which is different for different grouping methods, i.e. the number of groups of user groups comprised by different grouping methods is different; d represents the total number of users.
208. And determining a grouping method with the maximum model comprehensive value degree in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity degree of a mobile phone bank to be predicted.
It should be noted that, the content of step 208 is similar to that of step 105, and for avoiding repetition, reference is made to the content of step 105.
In a feasible implementation manner, a plurality of user groups of each grouping method are obtained through K grouping methods, a plurality of model comprehensive value degrees corresponding to K can be obtained through the preset model comprehensive value degree algorithm formula, the grouping method with the largest model comprehensive value degree is selected to be determined as a target grouping method through the model comprehensive value degree comparison of each grouping method, the subsequent activity degree prediction of the mobile banking user to be predicted is carried out, and the polarity activity degree prediction of the mobile banking user to be predicted with the largest prediction accuracy can be guaranteed.
The invention provides a method for determining a mobile banking user activity prediction strategy, which comprises the following steps: predicting the use data of each mobile banking user based on the classification model to obtain the first user activity of each mobile banking user; the first user activity is sequenced from high to low to obtain a user sequence formed by each mobile banking user; equidistant grouping of the preset K grouping methods is carried out on the user sequence, and a plurality of user groups corresponding to the grouping methods are obtained; and training and verifying the liveness prediction model based on a plurality of user groups corresponding to each grouping method, selecting a target grouping method corresponding to the maximum model comprehensive value of all liveness prediction models in each grouping method, and taking the target grouping method, each liveness prediction model corresponding to the target grouping method and the two classification models as target prediction strategies to predict the user liveness of the mobile banking user to be predicted. Firstly, the model parameters do not need to be continuously adjusted, time and energy are reduced, secondly, sample characteristics with strong prediction force do not need to be collected when the samples are collected, the accuracy of the user activity of actual prediction can be improved by adopting the target prediction strategy, and the model prediction is stable.
Referring to fig. 3, fig. 3 is a flowchart illustrating a method for predicting the activity of a mobile banking user according to an embodiment of the present invention. The method shown in fig. 3 comprises the following steps:
301. acquiring use data of a mobile phone banking user to be predicted on the mobile phone banking, wherein the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of the mobile phone banking user to be predicted after desensitization;
it should be noted that, the content of step 301 is similar to that of step 201, and the repeated content will not be described herein, and reference may be made to the content shown in step 201.
It should be noted that, the usage data corresponding to the mobile banking user to be predicted is the same as the data collection dimension of the usage sample data corresponding to the mobile banking sample user, including, but not limited to, basic attribute data, asset status data, fund flow data, mobile banking usage data, and opened financial product data after desensitization corresponding to each mobile banking user to be predicted.
302. Inputting the usage data into a classification model included in a target prediction strategy, and determining second user liveness corresponding to the mobile banking user to be predicted, wherein the target prediction strategy is determined by using a determination method of the liveness prediction strategy of the mobile banking user shown in any one of fig. 1 or fig. 2;
It will be appreciated that steps 301 and 302 are similar to steps 201 and 202, and reference is specifically made to steps 201 and 202.
Further, the second user activity degree is obtained by inputting the use data of the user to be predicted into a two-classification model, and the obtained model prediction result is used for reflecting the activity degree of the user to be predicted on the mobile phone bank.
303. Determining a target liveness prediction model according to the second user liveness and liveness prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
and searching an activity prediction model corresponding to each user group in a target grouping method included in the target prediction strategy, and further obtaining an activity prediction model of the user group corresponding to the second user activity so as to determine the target activity prediction model.
It can be understood that each user group in the target grouping method is obtained by equidistant grouping through the first user activity, so that each user group corresponds to a variation range of user activity, further, through the second user activity, the first user activity identical to the second user activity in which user group exists in each user group in the target grouping method can be obtained, and further, an activity prediction model of the user group corresponding to the first user activity identical to the second user activity exists is used as a target activity prediction model, so as to predict the third user activity corresponding to the mobile banking user to be predicted in step 304.
304. Inputting the usage data into the target liveness prediction model, and determining the third liveness corresponding to the mobile banking user to be predicted;
it can be understood that the third user activity degree is the activity degree of the mobile banking user to be predicted on the mobile banking, which is obtained by carrying out activity degree prediction on the usage data by using the target activity degree prediction model.
305. And carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
It can be understood that the same usage data to be predicted refers to usage data to be predicted corresponding to the same user, and the final user activity probability is determined by performing average calculation on the second user activity probability p1 and the third user activity probability p2 corresponding to the same user, so that the activity prediction of the user is more accurate and reliable. The method can truly respond to the activity degree of the user to be predicted in using the mobile phone bank.
The invention provides a method for predicting the activity of a mobile banking user, which comprises the following steps: acquiring use data of a mobile phone banking user to be predicted on the mobile phone banking, wherein the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of the mobile phone banking user to be predicted after desensitization; inputting the usage data into a classification model included in a target prediction strategy, and determining second user liveness corresponding to a mobile banking user to be predicted, wherein the target prediction strategy is determined by using a determination method of the liveness prediction strategy of the mobile banking user shown in any one of fig. 1 or fig. 2; determining a target liveness prediction model according to the second user liveness and liveness prediction models of a plurality of user groups corresponding to a target grouping method included in a target prediction strategy; inputting the use data into a target liveness prediction model, and determining the third user liveness corresponding to the mobile banking user to be predicted; and carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted. The target liveness prediction model comprising the two classification models and the target grouping method is utilized to predict the liveness of the user to be predicted twice, and the target grouping method is determined by the grouping method with the largest model comprehensive value degree in the K grouping methods, so that the prediction accuracy can be ensured by the liveness prediction model comprising the largest model comprehensive value degree during the prediction of the second liveness, and finally, the average value calculation of the prediction results of the two liveness is carried out to determine the final user liveness probability, the prediction result is accurate, and the real assessment of the liveness of the user of the mobile banking to be predicted on the mobile banking is facilitated.
Further, taking the case of performing activity prediction on N mobile banking users to be predicted at the same time as performing the descriptions of steps 301-305, it can be understood that when performing activity prediction on N mobile banking users to be predicted at the same time, steps 301 and 305 are similar to performing activity prediction on one mobile banking user to be predicted, so that details are not repeated, and the description of the foregoing embodiment can be referred to, where when N users to be detected are N, the second user activities of N to be predicted usage data corresponding to the mobile banking users are sequenced from high to low to obtain a to-be-predicted user sequence, and equidistant grouping of the to-be-predicted user sequence is performed according to a target grouping method, and each to-be-predicted user group of the target grouping method is determined, where the target grouping method is a grouping method corresponding to the greatest model comprehensive value in the model comprehensive value, and then each user group is input into an activity prediction model corresponding to the target grouping method to obtain the third user activities of N users;
further, assuming that the number of groups of the target grouping method k=3 is taken as an example, there are three liveness prediction models i, and each group is input into the corresponding liveness prediction model i, wherein the correspondence here means correspondence according to the user group.
In a feasible implementation manner, the target activity prediction model corresponding to each user can be obtained according to the second user activities corresponding to n mobile banking users to be predicted and the matching manner of the first user activities of each user in the user group divided by the target grouping method in the model training stage, and the third user activities corresponding to each user can be obtained by utilizing the target activity prediction model. Therefore, the above description of steps 301-305, which is performed by taking the simultaneous prediction of the liveness of N mobile banking users to be predicted as an example, is not limited to the specific example.
In a possible implementation manner, the method for predicting the activity of the mobile banking user further includes:
a. if the final user liveness is greater than or equal to a preset first liveness threshold value, determining that the mobile banking user to be predicted is a highly active guest group, and sending first recommendation information to a preset terminal of the mobile banking user to be predicted according to a first sending frequency, wherein the first sending frequency comprises daily, and the first recommendation information comprises warm prompt information;
b. if the final user activity is smaller than a preset first activity threshold and the final user activity is greater than or equal to a second activity threshold, determining that the mobile banking user to be predicted is a moderately active guest group, and sending second recommendation information to a preset terminal of the mobile banking user to be predicted according to a second sending frequency, wherein the second sending frequency comprises daily, and the second recommendation information comprises daily corresponding preferential activities; the preset first activity threshold is greater than the preset second activity threshold;
c. If the final user activity is smaller than a preset second activity threshold, determining that the mobile banking user to be predicted is a low-activity guest group, and sending third recommendation information to a preset terminal of the mobile banking user to be predicted according to a third sending frequency, wherein the third sending frequency comprises a preset special date, and the third recommendation information comprises preferential activities corresponding to the special date.
It should be noted that, the first sending frequency, the second sending frequency and the third sending frequency are different information pushing frequencies adopted for different guest groups, so as to meet the requirements of different guest groups, and meanwhile, the activity of the low-activity guest group can be improved, and the activity of the high-activity guest group can be maintained.
For example, the preset first liveness threshold may be 0.8, the preset second liveness threshold may be 0.2, and further, the highly active guest group is a user whose end user liveness probability is greater than or equal to 0.8; the moderately active guest group is a user with the final user activity probability being more than or equal to 0.2 and less than 0.8; the low liveness group is a user whose end user liveness probability is less than 0.2. Recommendation information is sent to the high-activity guest group, the medium-activity guest group and the low-activity guest group according to different recommendation strategies respectively to strengthen the user activity, wherein the different recommendation strategies comprise a first recommendation strategy, a second recommendation strategy and a third recommendation strategy; the first recommendation strategy is to adopt a shielding contact marketing strategy for the low-activity guest group; the second recommendation strategy adopts a positive marketing strategy for the medium activity client; the third recommendation strategy is to adopt a non-marketing strategy for the high-activity client; the shielding contact strategy is used for sending first recommendation information related to the recommendation preferential activity corresponding to the special date on the special date; the active marketing strategy is to send second recommendation information related to the recommendation activities corresponding to the daily on the daily basis; the non-marketing strategy includes sending a third recommendation of a warm reminder corresponding to a day of the day.
In one possible implementation, the shielding contact marketing strategy may be to notify the user corresponding to the low liveness guest group only when the activities with great preference force such as small and long holidays, annual end and the like; the active marketing strategy can inform the user corresponding to the moderate activity guest group of daily activities and offers, and the user corresponding to the moderate activity guest group is taken as a daily key marketing guest group; the non-marketing strategy may be to send warm cues only at regular times to maintain contact with highly active demographics.
Furthermore, the customers whose activity probability in two months is reduced beyond the preset loss threshold value can be defined as loss early warning customer groups based on the activity change of the mobile banking users for nearly 2 months, wherein the preset loss threshold value can be 0.2, the loss early warning customer groups adopt a large-amount stimulation marketing strategy, the user activity degree is timely improved while loss is prevented through activities with larger preferential amount, the risk problem that account exists due to the fact that mobile banking users corresponding to the loss early warning customer groups do not operate the mobile banking for a long time can be prevented, and illegal personnel can be prevented from stealing information and the like.
Referring to fig. 4, fig. 4 is a block diagram illustrating a determination apparatus for activity prediction policy of a mobile banking user according to an embodiment of the present invention. The apparatus shown in fig. 4 includes:
Feature acquisition module 401: the method comprises the steps of acquiring using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to a sequence from high to low by utilizing first user liveness of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
user grouping module 402: the user sequence is equidistantly grouped according to preset K grouping methods and the first user activity level to obtain a plurality of user groups corresponding to the K grouping methods respectively, and the number of the user groups divided by the K grouping methods is different;
model training module 403: for setting the P-th based on a preset LightGBM algorithm n i Training and verifying an liveness prediction model by using data of mobile banking users contained in each user group to obtain a trained P-th n i A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer, P n i Represents the ith in the nth group;
the value calculation module 404: the method comprises the steps of determining the model comprehensive value of an nth group according to the model value of an activity prediction model corresponding to each user group in the nth group and the use data of mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
policy access module 405: the method is used for determining a grouping method with the largest model comprehensive value in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity of a mobile phone bank to be predicted.
It will be appreciated that the functions of the modules shown in fig. 4 are similar to those of the method shown in fig. 1, and reference may be made to the foregoing related contents of the steps of the method shown in fig. 1 for avoiding repetition.
The invention provides a device for determining a mobile banking user activity prediction strategy. Predicting the use data of each mobile banking user based on the classification model to obtain the first user activity of each mobile banking user; the first user activity is sequenced from high to low to obtain a user sequence formed by each mobile banking user; equidistant grouping of the preset K grouping methods is carried out on the user sequence, and a plurality of user groups corresponding to the grouping methods are obtained; and training and verifying the liveness prediction model based on a plurality of user groups corresponding to each grouping method, selecting a target grouping method corresponding to the maximum model comprehensive value of all liveness prediction models in each grouping method, and taking the target grouping method, each liveness prediction model corresponding to the target grouping method and the two classification models as target prediction strategies to predict the user liveness of the mobile banking user to be predicted. Firstly, the model parameters do not need to be continuously adjusted, time and energy are reduced, secondly, sample characteristics with strong prediction force do not need to be collected when the samples are collected, the accuracy of the user activity of actual prediction can be improved by adopting the target prediction strategy, and the model prediction is stable.
Referring to fig. 5, fig. 5 is a block diagram illustrating a device for predicting activity of a mobile banking user according to an embodiment of the present invention. The apparatus shown in fig. 5 includes:
the data acquisition module 501: the method comprises the steps of acquiring use data of a mobile phone bank user to be predicted on the mobile phone bank, wherein the use data at least comprise basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the mobile phone bank user to be predicted after desensitization;
the first prediction module 502: the target prediction strategy is determined by using a determination method of the activity prediction strategy of the mobile banking user as shown in any one of the figure 1 or the figure 2;
model determination module 503: the target activity prediction model is determined according to the second user activity and activity prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
the second prediction module 504: the target activity degree prediction model is used for inputting the use data to determine the third user activity degree corresponding to the mobile banking user to be predicted;
Probability determination module 505: and the average value calculation is used for carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
It will be appreciated that the functions of the modules shown in fig. 5 are similar to those of the method shown in fig. 3, and reference may be made to the foregoing related contents of the steps of the method shown in fig. 3 for avoiding repetition.
The invention provides a prediction device for mobile banking user liveness, which comprises: and a data acquisition module: the method comprises the steps of acquiring use data of a mobile phone banking user to be predicted on the mobile phone banking, wherein the use data at least comprise basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of the mobile phone banking user to be predicted after desensitization; a first prediction module: the method comprises the steps of inputting usage data into a classification model included in a target prediction strategy, and determining second user liveness corresponding to a mobile banking user to be predicted, wherein the target prediction strategy is determined by using a determination method of the liveness prediction strategy of the mobile banking user as shown in any one of fig. 1 or 2; model determination module: the target activity prediction model is determined according to the second user activity and activity prediction models of a plurality of user groups corresponding to a target grouping method included in a target prediction strategy; a second prediction module: the method comprises the steps of inputting usage data into a target liveness prediction model, and determining third user liveness corresponding to a mobile banking user to be predicted; probability determination module: and the average value calculation is used for carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted. The target liveness prediction model comprising the two classification models and the target grouping method is utilized to predict the liveness of the user to be predicted twice, and the target grouping method is determined by the grouping method with the largest model comprehensive value degree in the K grouping methods, so that the prediction accuracy can be ensured by the liveness prediction model comprising the largest model comprehensive value degree during the prediction of the second liveness, and finally, the average value calculation of the prediction results of the two liveness is carried out to determine the final user liveness probability, the prediction result is accurate, and the real assessment of the liveness of the user of the mobile banking to be predicted on the mobile banking is facilitated.
In one possible implementation, the apparatus shown in fig. 5 further includes:
the customer group recommendation module is used for determining that the mobile banking user to be predicted is a highly active customer group if the activity degree of the end user is greater than or equal to a preset first activity threshold value, and sending first recommendation information to a preset terminal of the mobile banking user to be predicted according to a first sending frequency, wherein the first sending frequency comprises daily, and the first recommendation information comprises warm prompt information; if the final user activity is smaller than a preset first activity threshold and the final user activity is greater than or equal to a second activity threshold, determining that the mobile banking user to be predicted is a moderately active guest group, and sending second recommendation information to a preset terminal of the mobile banking user to be predicted according to a second sending frequency, wherein the second sending frequency comprises daily, and the second recommendation information comprises daily corresponding preferential activities; the preset first activity threshold is greater than the preset second activity threshold; if the final user activity is smaller than a preset second activity threshold, determining that the mobile banking user to be predicted is a low-activity guest group, and sending third recommendation information to a preset terminal of the mobile banking user to be predicted according to a third sending frequency, wherein the third sending frequency comprises a preset special date, and the third recommendation information comprises preferential activities corresponding to the special date.
FIG. 6 illustrates an internal block diagram of a computer device in one embodiment. The computer device may specifically be a terminal or a server. As shown in fig. 6, the computer device includes a processor, a memory, and a network interface connected by a system bus. The memory includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and may also store a computer program which, when executed by a processor, causes the processor to implement the method described above. The internal memory may also have stored therein a computer program which, when executed by a processor, causes the processor to perform the method described above. It will be appreciated by those skilled in the art that the structure shown in FIG. 6 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.
In one embodiment, a computer device is presented comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps as shown in any one of fig. 1, 2 or 3.
In one embodiment, a computer-readable storage medium is provided, storing a computer program that, when executed by a processor, causes the processor to perform the steps shown in any one of fig. 1, 2 or 3.
Those skilled in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, where the program may be stored in a non-volatile computer readable storage medium, and where the program, when executed, may include processes in the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The non-volatile memory may include read only memory (ROD), programmable ROD (PROD), electrically Programmable ROD (EPROD), electrically erasable programmable ROD (eepro), or flash memory. Volatile memory can include random access memory (RAD) or external cache memory. By way of illustration and not limitation, RADs are available in a variety of forms such as Static RAD (SRAD), dynamic RAD (DRAD), synchronous DRAD (SDRAD), double Data Rate SDRAD (DDRSDRAD), enhanced SDRAD (ESDRAD), synchronous link (syncldink) DRAD (SLDRAD), memory bus (RaDbus) direct RAD (RDRAD), direct memory bus dynamic RAD (DRDRAD), and memory bus dynamic RAD (RDRAD).
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.

Claims (9)

1. The method for determining the activity prediction strategy of the mobile banking user is characterized by comprising the following steps of:
collecting using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to the sequence from high to low by utilizing the first user activity of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
Equidistant grouping is carried out on the user sequences according to preset K grouping methods and the first user liveness, so that a plurality of user groups respectively corresponding to the K grouping methods are obtained, and the number of the user groups divided by the K grouping methods is different;
based on a preset LightGBM algorithm, the P-th algorithm is performed n i Training and verifying an liveness prediction model by using data of mobile banking users contained in each user group to obtain a trained P-th n i A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer, P n i Represents the ith in the nth group;
determining the model comprehensive value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the use data of the mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
determining a grouping method with the largest model comprehensive value degree in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification models as target prediction strategies, wherein the target prediction strategies are used for predicting the user activity degree of a mobile phone bank to be predicted;
The method comprises the steps of using the first user liveness of the D mobile banking users to sort the D mobile banking users according to the sequence from high to low to obtain a user sequence, and further comprising the following steps:
obtaining using sample data of D mobile phone bank sample users on the mobile phone banks and user active labels corresponding to the mobile phone bank sample users, wherein the using sample data at least comprises basic attribute characteristics, asset condition characteristics, fund running line characteristics, mobile phone bank using characteristics and opened financial product characteristics of the mobile phone bank sample users after desensitization, and the user active labels comprise first labels corresponding to active users or second labels corresponding to inactive users;
based on a preset LightGBM algorithm, performing iterative training on the two classification models by using the sample data of the D mobile phone bank sample users on the mobile phone banks until the two classification models are determined to converge according to the user active labels corresponding to the D mobile phone bank sample users;
and inputting the use data included by the D mobile banking users into the trained two classification models, and determining the first user activity corresponding to each mobile banking user.
2. The method according to claim 1, wherein determining the model integrated value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the usage data of the mobile banking users in the nth group further comprises:
by using the P n i Determining the P-th value algorithm of the use data and the preset model value algorithm of the mobile banking users contained in the user groups n i And the model value degree corresponding to each liveness prediction model.
3. The method according to claim 1, wherein the determining the model integrated value of the nth group according to the model value of the activity prediction model corresponding to each user group in the nth group and the usage data of the mobile banking users in the nth group includes:
the number Di of the mobile phone banking users, the total number D of the mobile phone banking users and the model value of each liveness prediction model corresponding to each user group are obtained;
and determining the model comprehensive value of the nth group by using the number Di of mobile banking users included in each user group corresponding to the nth group, the model value of the activity prediction model corresponding to each user group in the nth group and the total number D of mobile banking users.
4. The method for predicting the activity of the mobile banking user is characterized by comprising the following steps of:
acquiring use data of a mobile phone banking user to be predicted on the mobile phone banking, wherein the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone banking use characteristics and opened financial product characteristics of the mobile phone banking user to be predicted after desensitization;
inputting the usage data into a classification model included in a target prediction strategy, and determining second user liveness corresponding to the mobile banking user to be predicted, wherein the target prediction strategy is determined by using the method for determining the liveness prediction strategy of the mobile banking user according to any one of claims 1-3;
determining a target liveness prediction model according to the second user liveness and liveness prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
inputting the usage data into the target liveness prediction model, and determining the third liveness corresponding to the mobile banking user to be predicted;
and carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
5. The method of claim 4, wherein the method further comprises:
if the final user liveness is greater than or equal to a preset first liveness threshold value, determining that the mobile banking user to be predicted is a highly active guest group, and sending first recommendation information to a preset terminal of the mobile banking user to be predicted according to a first sending frequency, wherein the first sending frequency comprises daily, and the first recommendation information comprises warm prompt information;
if the final user activity is smaller than a preset first activity threshold and the final user activity is greater than or equal to a second activity threshold, determining that the mobile banking user to be predicted is a moderately active guest group, and sending second recommendation information to a preset terminal of the mobile banking user to be predicted according to a second sending frequency, wherein the second sending frequency comprises daily, and the second recommendation information comprises daily corresponding preferential activities; the preset first activity threshold is greater than the preset second activity threshold;
if the final user activity is smaller than a preset second activity threshold, determining that the mobile banking user to be predicted is a low-activity guest group, and sending third recommendation information to a preset terminal of the mobile banking user to be predicted according to a third sending frequency, wherein the third sending frequency comprises a preset special date, and the third recommendation information comprises preferential activities corresponding to the special date.
6. A device for determining an activity prediction strategy of a mobile banking user, the device comprising:
the characteristic acquisition module is used for: the method comprises the steps of acquiring using data of D mobile phone banking users on the mobile phone banking, and sequencing the D mobile phone banking users according to a sequence from high to low by utilizing first user liveness of the D mobile phone banking users to obtain a user sequence; the first user activity is obtained by predicting the use data by using a classification model, and the use data at least comprises basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the desensitized mobile phone bank user;
and a user grouping module: the user sequence is equidistantly grouped according to preset K grouping methods and the first user activity level to obtain a plurality of user groups corresponding to the K grouping methods respectively, and the number of the user groups divided by the K grouping methods is different;
model training module: for setting the P-th based on a preset LightGBM algorithm n i Training and verifying an liveness prediction model by using data of mobile banking users contained in each user group to obtain a trained P-th n i A liveness prediction model, n is the number of groups corresponding to the grouping method K, n is a positive integer, P n i Represents the ith in the nth group;
the value calculation module: the method comprises the steps of determining the model comprehensive value of an nth group according to the model value of an activity prediction model corresponding to each user group in the nth group and the use data of mobile banking users in the nth group, wherein the model comprehensive value is used for reflecting the actual degree of the obtained user activity when the user activity prediction of the mobile banking users is carried out by utilizing each activity prediction model corresponding to the nth group;
policy access module: the method comprises the steps of determining a grouping method with the largest model comprehensive value in the K grouping methods as a target grouping method, and taking the target grouping method, activity prediction models of a plurality of user groups corresponding to the target grouping method and the classification model as a target prediction strategy, wherein the target prediction strategy is used for predicting the user activity of a mobile phone bank to be predicted;
the method comprises the steps of using the first user liveness of the D mobile banking users to sort the D mobile banking users according to the sequence from high to low to obtain a user sequence, and further comprising the following steps: obtaining using sample data of D mobile phone bank sample users on the mobile phone banks and user active labels corresponding to the mobile phone bank sample users, wherein the using sample data at least comprises basic attribute characteristics, asset condition characteristics, fund running line characteristics, mobile phone bank using characteristics and opened financial product characteristics of the mobile phone bank sample users after desensitization, and the user active labels comprise first labels corresponding to active users or second labels corresponding to inactive users; based on a preset LightGBM algorithm, performing iterative training on the two classification models by using the sample data of the D mobile phone bank sample users on the mobile phone banks until the two classification models are determined to converge according to the user active labels corresponding to the D mobile phone bank sample users; and inputting the use data included by the D mobile banking users into the trained two classification models, and determining the first user activity corresponding to each mobile banking user.
7. A device for predicting activity of a mobile banking user, the device comprising:
and a data acquisition module: the method comprises the steps of acquiring use data of a mobile phone bank user to be predicted on the mobile phone bank, wherein the use data at least comprise basic attribute characteristics, asset condition characteristics, fund running characteristics, mobile phone bank use characteristics and opened financial product characteristics of the mobile phone bank user to be predicted after desensitization;
a first prediction module: the method is used for inputting the usage data into a classification model included in a target prediction strategy, determining the activity of a second user corresponding to the mobile banking user to be predicted, wherein the target prediction strategy is determined by using the method for determining the activity prediction strategy of the mobile banking user according to any one of claims 1-3;
model determination module: the target activity prediction model is determined according to the second user activity and activity prediction models of a plurality of user groups corresponding to a target grouping method included in the target prediction strategy;
a second prediction module: the target activity degree prediction model is used for inputting the use data to determine the third user activity degree corresponding to the mobile banking user to be predicted;
Probability determination module: and the average value calculation is used for carrying out average value calculation on the second user activity level and the third user activity level, and determining the final user activity level of the mobile banking user to be predicted.
8. A computer readable storage medium storing a computer program, which when executed by a processor causes the processor to perform the steps of the method of any one of claims 1 to 3 or 4 to 5.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program which, when executed by the processor, causes the processor to perform the steps of the method of any of claims 1 to 3 or 4 to 5.
CN202110798024.6A 2021-07-14 2021-07-14 Method and device for determining activity prediction strategy of mobile banking user Active CN113627653B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110798024.6A CN113627653B (en) 2021-07-14 2021-07-14 Method and device for determining activity prediction strategy of mobile banking user

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110798024.6A CN113627653B (en) 2021-07-14 2021-07-14 Method and device for determining activity prediction strategy of mobile banking user

Publications (2)

Publication Number Publication Date
CN113627653A CN113627653A (en) 2021-11-09
CN113627653B true CN113627653B (en) 2023-10-20

Family

ID=78379810

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110798024.6A Active CN113627653B (en) 2021-07-14 2021-07-14 Method and device for determining activity prediction strategy of mobile banking user

Country Status (1)

Country Link
CN (1) CN113627653B (en)

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105869047A (en) * 2016-03-28 2016-08-17 湖北经济学院 Multidimensional method for evaluating mobile banking user credibility
CN106780008A (en) * 2016-12-05 2017-05-31 山东浪潮商用系统有限公司 A kind of integration managing method and apparatus
CN109727070A (en) * 2018-12-28 2019-05-07 中国银行股份有限公司 A kind of determination method and device of potential any active ues
CN111626766A (en) * 2020-04-23 2020-09-04 深圳索信达数据技术有限公司 Mobile banking marketing customer screening method integrating multiple machine learning models
CN111695042A (en) * 2020-06-10 2020-09-22 湖南湖大金科科技发展有限公司 User behavior prediction method and system based on deep walking and ensemble learning
CN111709826A (en) * 2020-06-11 2020-09-25 中国建设银行股份有限公司 Target information determination method and device
CN111882113A (en) * 2020-07-01 2020-11-03 中国银行股份有限公司 Enterprise mobile banking user prediction method and device
CN111967671A (en) * 2020-08-18 2020-11-20 中国银行股份有限公司 Cross-border active user identification method and device based on support vector data field description
CN111967971A (en) * 2020-08-18 2020-11-20 中国银行股份有限公司 Bank client data processing method and device
CN111988294A (en) * 2020-08-10 2020-11-24 中国平安人寿保险股份有限公司 User identity recognition method, device, terminal and medium based on artificial intelligence
CN112148995A (en) * 2020-11-02 2020-12-29 深圳壹账通智能科技有限公司 Product recommendation method and device, electronic equipment and readable storage medium
CN113112339A (en) * 2021-05-17 2021-07-13 中国银行股份有限公司 Recommendation method and device for mobile banking products

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10055726B2 (en) * 2014-07-14 2018-08-21 Jpmorgan Chase Bank, N.A. Systems and methods for management of mobile banking resources
RU2635275C1 (en) * 2016-07-29 2017-11-09 Акционерное общество "Лаборатория Касперского" System and method of identifying user's suspicious activity in user's interaction with various banking services

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105869047A (en) * 2016-03-28 2016-08-17 湖北经济学院 Multidimensional method for evaluating mobile banking user credibility
CN106780008A (en) * 2016-12-05 2017-05-31 山东浪潮商用系统有限公司 A kind of integration managing method and apparatus
CN109727070A (en) * 2018-12-28 2019-05-07 中国银行股份有限公司 A kind of determination method and device of potential any active ues
CN111626766A (en) * 2020-04-23 2020-09-04 深圳索信达数据技术有限公司 Mobile banking marketing customer screening method integrating multiple machine learning models
CN111695042A (en) * 2020-06-10 2020-09-22 湖南湖大金科科技发展有限公司 User behavior prediction method and system based on deep walking and ensemble learning
CN111709826A (en) * 2020-06-11 2020-09-25 中国建设银行股份有限公司 Target information determination method and device
CN111882113A (en) * 2020-07-01 2020-11-03 中国银行股份有限公司 Enterprise mobile banking user prediction method and device
CN111988294A (en) * 2020-08-10 2020-11-24 中国平安人寿保险股份有限公司 User identity recognition method, device, terminal and medium based on artificial intelligence
CN111967671A (en) * 2020-08-18 2020-11-20 中国银行股份有限公司 Cross-border active user identification method and device based on support vector data field description
CN111967971A (en) * 2020-08-18 2020-11-20 中国银行股份有限公司 Bank client data processing method and device
CN112148995A (en) * 2020-11-02 2020-12-29 深圳壹账通智能科技有限公司 Product recommendation method and device, electronic equipment and readable storage medium
CN113112339A (en) * 2021-05-17 2021-07-13 中国银行股份有限公司 Recommendation method and device for mobile banking products

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Fraud Detection in Banking Data by Machine Learning Techniques;Hashemi, SK;IEEE ACCESS;第11卷;3034-3043 *
客户离网预警研究报告;陈志伟;钟方正;孔莹;梁晓;叶林;张鹏;;电信科学(第S1期);281-282 *
陈志伟 ; 钟方正 ; 孔莹 ; 梁晓 ; 叶林 ; 张鹏 ; .客户离网预警研究报告.电信科学.2020,(第S1期),281-282. *

Also Published As

Publication number Publication date
CN113627653A (en) 2021-11-09

Similar Documents

Publication Publication Date Title
Ghodselahi et al. Application of artificial intelligence techniques for credit risk evaluation
CN106600369A (en) Real-time recommendation system and method of financial products of banks based on Naive Bayesian classification
CN110378786B (en) Model training method, default transmission risk identification method, device and storage medium
CN110599336A (en) Financial product purchase prediction method and system
Sharaf et al. StockPred: a framework for stock Price prediction
RU2723448C1 (en) Method of calculating client credit rating
Sharma et al. Machine learning model for credit card fraud detection-a comparative analysis.
CN115545886A (en) Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium
Diwate et al. Loan Approval Prediction Using Machine Learning
Jurgovsky Context-aware credit card fraud detection
CN113627653B (en) Method and device for determining activity prediction strategy of mobile banking user
Chen et al. Feature selection on credit risk prediction for peer-to-peer lending
El-Berawi et al. Adaptive deep learning based cryptocurrency price fluctuation classification
Vahid et al. Modeling corporate customers’ credit risk considering the ensemble approaches in multiclass classification: evidence from Iranian corporate credits
Ullah et al. Predicting Default Payment of Credit Card Users: Applying Data Mining Techniques
Huang Credit card transaction fraud using machine learning algorithms
CN113706258A (en) Product recommendation method, device, equipment and storage medium based on combined model
Wu et al. CDGAT: a graph attention network method for credit card defaulters prediction
Dzelihodzic et al. Data Mining Techniques for Credit Risk Assessment Task
Keswani et al. Adapting machine learning techniques for credit card fraud detection
Ivan Bankruptcy Prediction for Innovative Companies
Gao et al. Forecasting of Credit Card Default Based on Incremental Random Forest
Liu et al. Research on the identification and prediction of default risk of online lending platform customers
Yang et al. A novel hybrid data mining framework for credit evaluation
Carvalho Modelling Real Entangled Systems Using Projections in Flat Abstract Spaces

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant