Collection scoring method and device

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the application with reference to the drawings in the embodiments of the application. Apparently, the described embodiments are only some of the embodiments of the application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

[0070] The application is applicable to numerous general purpose or special purpose computing device environments or configurations. For example: personal computer, server computer, handheld or portable device, tablet type device, multiprocessor device, distributed computing environment including any of the above devices or devices, etc.

[0071] The embodiment of the present application provides a collection scoring method, which can be applied to various system platforms, and its execution body can run on a computer terminal or a processor of various mobile devices. The method flow chart of the collection collection scoring method is as follows figure 1 shown, including:

[0072] S101. Acquire user information of a target user.

[0073] In this embodiment, the user information of the target user is acquired in response to the collection scoring request, wherein the target user is the user to be scored for collection, and the user information includes but not limited to account information, statement information, transaction flow information, and credit card repayment information and consumer information.

[0074] In this embodiment, the collection scoring request is a request sent by bank staff when it is necessary to perform collection scoring on the target user.

[0075] S102. Analyze the user information, and determine the range of overdue repayment days to which the target user's overdue repayment days belong.

[0076] In this embodiment, the range of the days of overdue repayment is preset. Optionally, the preset range of days of overdue repayment is 0-30 days of overdue repayment, and 31-60 days of overdue repayment.

[0077] In this embodiment, the user information of the overdue user is analyzed, and the overdue date of the overdue user included in the user information is obtained. Based on the overdue date, the overdue repayment days of the overdue user are calculated, and based on the preset range of overdue repayment days, determine The range of overdue repayment days to which the target user's repayment overdue days belong.

[0078] S103. Among the pre-built collection scoring models, the collection scoring model corresponding to the range of overdue repayment days is determined as the target collection scoring model.

[0079] In this embodiment, collection scoring models corresponding to different ranges of overdue days are constructed in advance, and different collection scoring models correspond to different ranges of overdue days.

[0080] In this implementation, determine the range of overdue repayment days corresponding to the number of overdue repayment days, that is, determine whether the range of overdue repayment days is 1 to 30 days, or the range of overdue repayment days is 31 days to 60 days, and then The collection scoring model corresponding to the overdue repayment days is determined, and the collection collection scoring model corresponding to the overdue repayment days among the various pre-built collection scoring models is determined as the target model.

[0081] In this example, see figure 2 , the construction process of each collection scoring model includes the following steps:

[0082] S201. Collect sample data of multiple overdue users, and each sample data includes multiple variable data.

[0083] To collect sample data of multiple overdue users, it should be noted that the sample data is data generated by the history of overdue users, and each sample data includes multiple variables.

[0084] S202. Classify each sample data according to the number of overdue repayment days to obtain multiple sample sets.

[0085] In this embodiment, the number of overdue repayment days corresponding to each sample data is determined, and each sample data is classified according to the overdue repayment days to obtain multiple sample sets. Optionally, each sample data can be classified into two categories , two sample sets are obtained, namely the first sample set and the second sample set. The number of overdue repayment days corresponding to the sample data belongs to the second overdue repayment days range. Optionally, the first overdue repayment days ranges from 1 to 30 days, and the second overdue repayment days ranges from 31 to 60 days; that is to say , different sample sets correspond to different ranges of overdue repayment days.

[0086] S203. For each sample set, according to the preset variable screening strategy corresponding to the sample set, perform variable screening on each sample data in the sample set, obtain the target sample data corresponding to each sample data, and form each target sample data The target sample set corresponding to the sample set.

[0087] In this embodiment, multiple variable screening strategies are preset, and different variable screening strategies correspond to different ranges of overdue repayment days.

[0088] In this embodiment, for each sample set, a preset variable screening strategy corresponding to the sample set is obtained. Specifically, according to the range of overdue repayment days corresponding to the sample set, the preset number of overdue repayment days is determined. The variable filtering strategy corresponding to the scope.

[0089] In this embodiment, for each sample set, according to the variable screening strategy corresponding to the sample set, variable screening is performed on each sample data in the sample set, and the result of variable screening in each sample data is determined as the result of the sample data. Target sample data.

[0090] It should be noted that each sample data in the same sample set performs variable screening according to the same variable screening strategy, and sample data in different sample sets performs variable screening according to different variable screening strategies.

[0091] refer to image 3 According to the preset variable screening strategy corresponding to the sample set, the variable screening is performed on each sample data in the sample set, and the process of obtaining the target sample data corresponding to each sample data includes the following steps:

[0092] S301. For each variable in each sample data in the sample set, calculate the missing rate and information value of the variable, and if the missing rate is less than the preset missing threshold and the information value is within the preset range, determine the variable as the first variable .

[0093] In this embodiment, for each variable in each sample data in the sample set, the missing rate and information value of the variable are calculated. For the specific calculation methods of the missing rate and information value, please refer to the prior art.

[0094] In this embodiment, for each variable in each sample data in the sample set, it is judged whether the missing rate of the variable is less than the preset missing threshold, and whether the information value is within the preset range. Optionally, the missing threshold can be 80 %, the preset range may be greater than 0.01 and less than 10.

[0095] In this embodiment, for each variable in each sample data in the sample set, if the missing rate of the variable is less than the preset missing threshold and the information value is within the preset range, then the variable is determined as the first variable; otherwise, The variable is not determined as the first variable, that is, when the missing rate of the variable is not less than the preset missing threshold, or the information value is not within the preset range, the variable is not determined as the first variable.

[0096] S302. For each first variable, judge whether the first variable conforms to a preset first business logic rule corresponding to the first variable, and if so, determine the first variable as a second variable.

[0097] In this embodiment, first business logic rules corresponding to different variables are preset.

[0098]In this embodiment, for each first variable, the first business logic rule corresponding to the first variable is obtained, and it is judged whether the first variable conforms to its corresponding first business logic rule. If the first variable conforms to its corresponding The first business logic rule determines the first variable as the second variable, and if the first variable does not comply with the corresponding first business logic rule, the first variable is not determined as the second variable.

[0099] S303. Binning each second variable to obtain a plurality of variable bins.

[0100] Carry out boxing processing on each second variable, specifically, perform boxing processing on each second variable according to the data type of the second variable to obtain multiple variable boxes, wherein the data type includes but not limited to numerical type and text type .

[0101] S304. For each variable box, if the proportion result of the variable box is less than the preset proportion threshold, determine each second variable in the variable box as a third variable.

[0102] In this embodiment, for each variable box, the proportion result of the variable box is calculated. Specifically, the quantity of the second variable included in the variable box is divided by the sum of the quantities of all second variables to obtain the variable , that is, the proportion result of the variable box is used to indicate the proportion between the quantity of the second variable included in the variable box and the sum of the quantities of all the second variables.

[0103] In this embodiment, for each variable box, it is judged whether the proportion result of the variable box is less than the preset proportion threshold, and if the proportion result of the variable box is less than the preset proportion threshold, then each One second variable is determined as the third variable, and if the proportion result of the variable box is not less than the preset proportion threshold, then no one of the second variables in the variable box is determined as the third variable.

[0104] S305. For each third variable, calculate the multicollinearity index VIF between the third variable and each other third variable, and remove other third variables whose VIF is greater than a preset threshold, wherein the other variables are each third variable A third variable of the three variables other than the third variable.

[0105] In this embodiment, for each third variable, the multicollinearity index VIF between the third variable and each other third variable is calculated. For the specific calculation method of the multicollinearity index VIF, please refer to the prior art, here No longer. Wherein, other third variables are third variables except the third variable among the third variables.

[0106] In this embodiment, for each third variable, it is judged whether the multicollinearity index VIF between the third variable and each other third variable is greater than a preset threshold, and each other variable whose VIF is greater than a preset threshold is evaluated. remove. Optionally, the preset threshold may be 10.

[0107] In this embodiment, in order to control the multicollinearity among the variables, linear regression is used to calculate the variance inflation factor (VIF). VIF is calculated by regressing each tertiary variable on the other tertiary variables, calculating R 2 , R 2 The larger the , the greater the possibility that the variability of a variable can be explained by the combination of other variables in the model.

[0108] Among them, VIF=1/(1-R^2), the greater the VIF, the greater the possibility of multicollinearity between the third variable and other third variables.

[0109] It should be noted that the third variable that has been eliminated is no longer used in the calculation of the multicollinearity index VIF.

[0110] For each third variable mentioned above, the process of calculating the multicollinearity index VIF between the third variable and each other third variable, and eliminating other third variables whose VIF is greater than the preset threshold is given as an example as follows:

[0111] The third variable is A, B, C, D, and E. For the third variable A, calculate the multicollinear VIF between AB, AC, AD, and AE, wherein the VIF between AB is greater than the preset threshold, The VIF between AE is greater than the preset threshold, then the third variables B and E are eliminated, and then for the third variable C, calculate the multicollinearity VIF between CDs, if the VIF between CDs is greater than the preset threshold, then Eliminate the third variable D.

[0112] S306. Among the remaining third variables, a third variable that satisfies a preset second business logic rule corresponding to the sample set is determined as a target variable.

[0113] In this embodiment, for each of the remaining third variables, it is judged whether the third variable satisfies the preset second business logic rule corresponding to the sample set to which the third variable belongs, and if the third variable satisfies the preset and For the second business logic rule corresponding to the sample set to which the third variable belongs, the third variable is determined as the target variable.

[0114] In this embodiment, the remaining third variables are further screened through the second business logic rule, and the variables that meet the business meaning and have significant characteristics are screened out.

[0115] S307. Compose each target variable into a target sample set corresponding to the sample set.

[0116] In this embodiment, each target variable is formed into a target sample set corresponding to the sample set.

[0117] Optionally, for each sample data in the first sample set, according to the variable screening rules corresponding to the first sample set, 16 variables are selected from each variable included in each sample data; For each sample data, according to the variable screening rules corresponding to the second sample set, 9 variables are screened out from each variable included in each second sample data.

[0118] In this embodiment, for each variable in each sample data in the sample set, the missing rate judgment, information value judgment, first business logic rule judgment, binning processing, multicollinearity index VIF judgment and second business logic are performed Judgment by rules to realize the screening of each variable to screen out the target variable.

[0119] S204. For each target sample set, select a plurality of target sample data from the target sample set to form a training data set, and select a plurality of target sample data from the target sample set to form a test data set, according to each target sample data included in the training data set , train the pre-built logistic regression model, and test the trained logistic regression model according to each target sample data included in the test data set. After the trained logistic regression model passes the test, the logistic regression that passes the test will be The model is determined as the initial collection scoring model for the target sample set.

[0120] In this embodiment, for each target sample set, multiple target sample data are selected from the target sample set to form a training data set, and multiple target sample data are selected from the target sample set to form a test data set; Select 70% of the target sample data in the sample set to make the training data set, and 30% of the target sample data can be selected from the target sample set to form the test data set; it should be noted that the number of target sample data included in the training data set and the test data The sum of the quantity of the target sample data included in the set is equal to the quantity of the target sample data included in the target sample set.

[0121] In this embodiment, for each target sample set, the pre-built logistic regression model is trained according to each target sample data in the training data set corresponding to the target sample set. The regression model is trained. After the training is completed, the trained logistic regression model is tested according to each target sample data in the test dataset corresponding to the target sample dataset. Specifically, the ROC curve of the trained logistic regression model is obtained. , KS value, where, ROC curve: Receiver Operating Characteristic (Receiver Operating Characteristic) curve is a useful visualization tool for comparing two classification models. The ROC curve shows the trade-off between true positive rate (TPR) and false positive rate (FPR) for a given model. KS (Kolmogorov-Smirnov): The indicator measures the difference between the cumulative divisions of good and bad samples. The KS value ranges from 0% to 100%. The criteria for discrimination are as follows:

[0122] KS value: <20%; Poor;

[0123] KS value: 20%-40%; general;

[0124] KS value: 41%-50%; good;

[0125] KS value: 51%-75%; very good;

[0126] KS value: >75%; too high, need to verify the model carefully.

[0127] In this embodiment, based on the ROC curve and the KS value, it is judged that the trained logistic regression models all pass the test, and if they pass the test, the logistic regression model that passes the test is determined as the initial collection scoring model of the target sample set.

[0128] S205. For each initial collection scoring model, collect a cross-time verification sample set of the initial collection scoring model, verify the initial collection scoring model according to each cross-time verification sample data included in the cross-time verification sample set, and verify the initial collection scoring model after the initial collection scoring model passes After verification, the initial collection scoring model is determined as the collection scoring model.

[0129] In this embodiment, for each initial collection scoring model, the cross-time verification sample set of the initial collection scoring model is collected. The cross-time verification set is the repayment behavior information of overdue users during the performance period, and the performance period is N after the observation point. Days, the observation point is the collection time point for collecting sample data of overdue users. It should be noted that, for different initial collection models, N is different; for the range of overdue repayment days is 1 to 30 days corresponding to the initial collection score model, N can be set to 90, and for the initial collection scoring model corresponding to the range of overdue repayment days from 31 to 60 days, N can be set to 60.

[0130] Wherein, the repayment behavior information includes the performance of the overdue user during the performance period. Optionally, it can reflect the performance of the overdue user during the performance period in terms of good or bad. In this embodiment, for different overdue repayment days, a first performance classification table and a second performance classification table are set to define the user's performance during the performance period, and the first performance classification table is used to define the range of overdue repayment days as The performance classification of overdue users corresponding to 1 to 30 days in the performance period, as shown in Table 1, the second performance classification table is used to define the performance of overdue users in the performance period corresponding to the range of overdue repayment days ranging from 31 to 60 days Classification, as shown in Table 2.

[0131] Table 1 First performance classification table

[0132]

[0133] Table 2 Second Performance Classification Table

[0134]

[0135] S104. Input the user information into the target collection scoring model to obtain the collection scoring result of the target user.

[0136] In this embodiment, the user information is input into the target collection scoring model, and the collection scoring result of the target user output by the target collection scoring model is obtained, so as to determine the corresponding collection strategy based on the collection scoring result and realize accurate collection. Wherein, the collection scoring result can be represented by "good" or "bad", that is, the target collection scoring model outputs a good or bad collection scoring result.

[0137] refer to Figure 4 , the process of inputting user information into the target collection scoring model to obtain the collection scoring result of the target user, including:

[0138] S401. Obtain multiple variable attributes of the target collection scoring model.

[0139]In this embodiment, the variable attributes corresponding to each collection scoring model are determined, that is to say, which variables need to be screened out as the input of the collection scoring model are determined.

[0140] In this embodiment, multiple variable attributes of the target collection scoring model are obtained, and the variable attributes are used to indicate the type of variables to be input into the model.

[0141] S402. Extract information corresponding to each variable attribute from the user information.

[0142] Analyze the user information, determine the type of each variable included in the user information, and extract the information corresponding to each variable attribute from the user information based on the type of each variable included in the user information, that is, for each Variable attribute, extract the variable whose type corresponds to the variable attribute from the user information.

[0143] S403. Input the information corresponding to each variable attribute into the target collection scoring model to obtain the collection scoring result of the target user.

[0144] In this embodiment, the extracted information corresponding to each variable attribute is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0145] Specifically, the process of inputting the information corresponding to each variable attribute into the target collection scoring model to obtain the collection scoring result of the target user specifically includes the following steps:

[0146] Calculate the eigenvector of the information corresponding to each variable attribute;

[0147] The characteristic vector of the information corresponding to each variable attribute is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0148] In this embodiment, the eigenvectors of the information corresponding to each variable attribute are calculated to obtain the eigenvectors of the information corresponding to each variable attribute, and the eigenvectors of the information corresponding to each variable attribute are input into the target collection score In the model, the collection scoring results of the target users are obtained.

[0149] The collection scoring method provided by the embodiment of this application obtains the user information of the target user, analyzes the user information, determines the range of the overdue payment days to which the target user's overdue repayment days belong, and combines the pre-built collection scoring models with The collection scoring model corresponding to the range of overdue repayment days is determined as the target collection scoring model, and the user information is input into the target collection scoring model to obtain the collection scoring result of the target user. Apply the collection scoring method provided by the embodiment of this application to pre-construct collection scoring models corresponding to different ranges of overdue repayment days, based on the range of overdue repayment days to which the target user's overdue repayment days belong, from each pre-built collection scoring model Determine the target collection scoring model, and based on the target collection scoring model, carry out collection scoring on target users, realize the use of different collection models for collection scoring for different ranges of overdue repayment days, improve the accuracy of collection scoring, and thus can be based on collection scoring As a result, the applicable collection strategy was determined to achieve precise collection.

[0150] and figure 1 Corresponding to the method described, the embodiment of the present application also provides a recovery scoring device for figure 1 The specific implementation of the method in the method, its structural diagram is as follows Figure 5 shown, including:

[0151] An acquisition unit 501, configured to acquire user information of a target user; the target user is a user to be scored for collection;

[0152] The first determining unit 502 is configured to analyze the user information and determine the range of overdue repayment days to which the overdue repayment days of the overdue user belongs;

[0153] The second determining unit 503 is configured to determine the collection scoring model corresponding to the overdue repayment days range among the pre-built collection scoring models as the target collection scoring model;

[0154] The input unit 504 is configured to input the user information into the target collection scoring model to obtain a collection scoring result of the target user.

[0155] In the debt collection scoring device provided in the embodiment of the present application, the range of overdue repayment days to which the repayment days belong determines the target collection scoring model from each pre-built collection scoring model, and based on the target collection scoring model, collects the target user. Realize In order to improve the accuracy of collection scoring by using different collection models for different ranges of overdue repayment days, it is possible to determine the applicable collection strategy based on the collection scoring results to achieve precise collection.

[0156] In one embodiment of the present application, based on the aforementioned solution, the input unit 504 is used to input the user information into the target collection scoring model to obtain the collection scoring result of the target user, including the input unit 504 being specifically used to :

[0157] Obtain multiple variable attributes of the target collection scoring model;

[0158] Extracting information corresponding to each variable attribute from the user information;

[0159] The information corresponding to each variable attribute is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0160] In one embodiment of the present application, based on the foregoing solution, the input unit 504 is used to input the information corresponding to each variable attribute into the target collection scoring model to obtain the collection scoring result of the target user, including the The input unit 504 is specifically used for:

[0161] Calculate the eigenvector of the information corresponding to each variable attribute;

[0162] The characteristic vector of the information corresponding to each variable attribute is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0163] In one embodiment of the present application, based on the foregoing solution, it can also be configured as:

[0164] A collection unit, configured to collect sample data of multiple overdue users, each sample data includes multiple variables;

[0165] The classification unit is used to classify each sample data according to the number of overdue repayment days to obtain multiple sample sets;

[0166] A screening unit is configured to, for each sample set, perform variable screening on each sample data in the sample set according to a preset variable screening strategy corresponding to the sample set, to obtain target sample data corresponding to each sample data , forming each target sample data into a target sample set corresponding to the sample set;

[0167] A training unit, for each target sample set, selecting a plurality of target sample data from the target sample set to form a training data set, and selecting a plurality of target sample data from the target sample set to form a test data set, and according to the Each target sample data included in the training data set is used to train the pre-built logistic regression model, and according to each target sample data included in the test data set, the corresponding trained logistic regression model is tested. After the logistic regression model passes the test, the logistic regression model passing the test is determined as the initial collection scoring model of the target sample set;

[0168] The verification unit is configured to, for each initial collection scoring model, collect a cross-time verification sample set of the initial collection scoring model, and verify the initial collection scoring model according to each cross-time verification sample data included in the cross-time verification sample set Verification is performed, and after the initial collection scoring model passes the verification, the initial collection scoring model is determined as the collection scoring model.

[0169] In one embodiment of the present application, based on the aforementioned solution, the training unit is used to perform variable screening on each sample data in the sample set according to the preset variable screening strategy corresponding to the sample set, to obtain each sample The target sample data corresponding to the data, including the training unit is specifically used for:

[0170] For each variable in each sample data in the sample set, calculate the missing rate and information value of the variable, if the missing rate is less than the preset missing threshold, and the information value is within the preset range, then the said variable is determined as a first variable;

[0171] For each first variable, judging whether the first variable conforms to a preset first business logic rule corresponding to the first variable, and if so, determining the first variable as a second variable;

[0172] Perform binning processing on each second variable to obtain multiple variable bins;

[0173] For each variable box, if the proportion result of the variable box is less than the preset proportion threshold, each second variable in the variable box is determined as a third variable; wherein, the proportion of the variable box The result is used to indicate the ratio between the quantity of the second variable included in the variable box and the sum of the quantities of all the second variables;

[0174] For each third variable, calculate the multicollinearity index VIF between the third variable and each other third variable, and remove other third variables whose VIF is greater than a preset threshold, wherein the other variables are each third variable. A third variable of the three variables other than said third variable;

[0175] Determining, among the remaining third variables, a third variable that satisfies a preset second business logic rule corresponding to the sample set as a target variable;

[0176] Composing each target variable into a target sample set corresponding to the sample set.

[0177] The embodiment of the application also provides a storage medium, the storage medium includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to perform the following operations:

[0178] Obtain the user information of the target user; the target user is the user who is to be collected and scored;

[0179] Analyzing the user information to determine the range of overdue repayment days to which the overdue repayment days of the overdue user belong;

[0180] Determining the collection scoring model corresponding to the range of overdue repayment days among the pre-built collection scoring models as the target collection scoring model;

[0181] The user information is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0182] The embodiment of the present application also provides an electronic device, the schematic diagram of which is shown in Image 6 As shown, it specifically includes a memory 601, and one or more instructions 602, wherein one or more instructions 602 are stored in the memory 601, and are configured to be executed by one or more processors 603 602 performs the following operations:

[0183] Obtain the user information of the target user; the target user is the user who is to be collected and scored;

[0184] Analyzing the user information to determine the range of overdue repayment days to which the overdue repayment days of the overdue user belong;

[0185] Determining the collection scoring model corresponding to the range of overdue repayment days among the pre-built collection scoring models as the target collection scoring model;

[0186] The user information is input into the target collection scoring model to obtain the collection scoring result of the target user.

[0187]It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. For the same and similar parts in each embodiment, refer to each other, that is, Can. As for the device-type embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to part of the description of the method embodiments.

[0188] Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

[0189] For the convenience of description, when describing the above devices, functions are divided into various units and described separately. Of course, when implementing the present application, the functions of each unit can be implemented in one or more pieces of software and/or hardware.

[0190] It can be known from the above description of the implementation manners that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the essence of the technical solution of this application or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in storage media, such as ROM/RAM, disk , optical disc, etc., including several instructions to enable a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0191] The above is a detailed introduction to a collection scoring method and device provided by this application. In this paper, specific examples are used to illustrate the principle and implementation of this application. The description of the above examples is only used to help understand the method of this application. and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of ​​this application, there will be changes in the specific implementation and application scope. limits.