Method, device and storage medium for obtaining user disconnection level

Abstract

The present application discloses a method, device and storage medium for obtaining a user's disconnection level, and relates to the field of data processing technology. The specific implementation scheme is: obtaining user variable data, preprocessing and deriving the variable data to obtain a first pending variable; binning the first pending variable, screening the first pending variable according to business logic to obtain a second pending variable; mapping the second pending variable with evidence weights, and screening the second pending variable according to correlation and information value to obtain a model entry variable; inputting the user's model entry variable into a scoring model to obtain the user's disconnection level. The embodiment of the present application can screen out variables that are more closely related to the predicted disconnection level, avoid ineffective work by staff, and more accurately obtain the user's disconnection level.

Description

Technical Field

[0001] The present disclosure relates to the technical field of big data processing, and in particular to a method, device and storage medium for obtaining a user's disconnection level. Background Art

[0002] With the continuous development of financial technology, big data and artificial intelligence technologies are becoming more mature, and the collection business of commercial banks is also transforming towards intelligence, refinement and industrialization. In recent years, affected by internal and external factors, the pressure on retail credit collection has increased and the loss rate has increased significantly. In traditional collection methods, many mobile phone numbers of customers with overdue loans cannot be connected or are in abnormal online status, making it impossible to reach the collection. At the same time, collection operators need to call one by one to confirm that the customer is out of contact, which is time-consuming and laborious. For example, in the telephone collection scenario, the average caller can get a response from the customer within 3 calls. For lost customers, they often cannot be connected after 6 calls or more. This has caused a huge waste of manpower and material costs. Summary of the invention

[0003] The present disclosure provides a method, device and storage medium for obtaining a user's disconnection level, so as to at least solve the problem of avoiding ineffective work of staff in the related art. The technical solution of the present disclosure is as follows:

[0004] According to a first aspect of an embodiment of the present disclosure, a method for obtaining a user disconnection level is provided, comprising:

[0005] Obtain user variable data, and preprocess and derive the variable data to obtain a first pending variable;

[0006] The first pending variable is binned and the first pending variable is screened according to business logic to obtain a second pending variable;

[0007] Performing evidence weight mapping on the second undetermined variable, and screening the second undetermined variable according to relevance and information value to obtain a model entry variable;

[0008] The input variables of the user are input into a scoring model to obtain the disconnection level of the user.

[0009] Optionally, the preprocessing and deriving the variable data to obtain a first undetermined variable includes at least one of the following:

[0010] Delete the variables whose missing rate is greater than or equal to the missing rate threshold, and determine the remaining variables as the first pending variables;

[0011] Variables whose maximum single value ratio is greater than or equal to the single value ratio threshold are deleted, and the remaining variables are determined as the first pending variables.

[0012] Optionally, the binning the first pending variable and screening the first pending variable according to business logic to obtain the second pending variable includes:

[0013] Binning samples corresponding to the first undetermined variable and obtaining a population stability index and information value of the first undetermined variable;

[0014] The first undetermined variable is screened according to the group stability index and the information value to obtain the second undetermined variable.

[0015] Optionally, after the binning process, the number of bad samples in each bin is greater than a bad sample number threshold, and the ratio of the number of samples in each bin to the total number of samples is greater than a ratio threshold.

[0016] Optionally, the step of screening the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable includes:

[0017] The first undetermined variables whose group stability index is less than the group stability threshold are deleted, and the first undetermined variables whose information value is less than the information value threshold are deleted, and the remaining first undetermined variables are determined as second undetermined variables.

[0018] Optionally, performing evidence weight mapping on the second undetermined variable and screening the second undetermined variable according to relevance and information value to obtain a model entry variable includes:

[0019] Obtaining correlation indexes of variable pairs in the second undetermined variable, and calculating variance inflation coefficients and default probabilities of the second undetermined variable;

[0020] Obtaining a second variable pair whose correlation index is greater than a correlation threshold, and retaining the second undetermined variable with the largest information value;

[0021] Delete the second undetermined variable whose variance inflation coefficient is greater than the inflation coefficient threshold, and delete the second undetermined variable whose default probability is greater than the probability threshold, so as to obtain the model input variable.

[0022] According to a second aspect of an embodiment of the present disclosure, a scoring model training method is provided, which is used to train the scoring model described in the first method, and is characterized by comprising:

[0023] Establishing a training data set, wherein the training data set includes the user's model input variables;

[0024] Marking the disconnection level of the user;

[0025] The training data set is input into the scoring model for training to obtain a recommendation scoring model.

[0026] According to a third aspect of an embodiment of the present disclosure, a device for obtaining a user disconnection level is provided, including:

[0027] A preprocessing module, used for obtaining user variable data, preprocessing and deriving the variable data to obtain a first undetermined variable;

[0028] A binning and screening module, used to bin the first pending variable and screen the first pending variable according to business logic to obtain a second pending variable;

[0029] A mapping and screening module, used for performing evidence weight mapping on the second undetermined variable, and screening the second undetermined variable according to relevance and information value to obtain a model input variable;

[0030] The scoring module is used to input the input variables of the user into the scoring model to obtain the loss level of the user.

[0031] Optionally, the preprocessing module includes:

[0032] A first preprocessing submodule deletes variables whose missing rate is greater than or equal to a missing rate threshold, and determines the remaining variables as first pending variables;

[0033] The second preprocessing submodule is used to delete the variables whose maximum single value ratio is greater than or equal to the single value ratio threshold, and determine the remaining variables as the first pending variables.

[0034] Optionally, the binning and screening module includes:

[0035] A binning module, used to bin samples corresponding to the first undetermined variable, and obtain a population stability index and an information value of the first undetermined variable;

[0036] The first screening submodule is used to screen the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable.

[0037] Optionally, after the binning process, the number of bad samples in each bin is greater than a bad sample number threshold, and the ratio of the number of samples in each bin to the total number of samples is greater than a ratio threshold.

[0038] Optionally, the screening submodule includes:

[0039] The screening unit is used to delete the first undetermined variables whose group stability index is less than the group stability threshold, and delete the first undetermined variables whose information value is less than the information value threshold, and determine the remaining first undetermined variables as second undetermined variables.

[0040] Optionally, the mapping screening module includes:

[0041] A second screening submodule, used to obtain the correlation index of the variable pairs in the second undetermined variable, and calculate the variance inflation coefficient and default probability of the second undetermined variable;

[0042] A third screening submodule is used to obtain a second variable pair whose correlation index is greater than a correlation threshold, and retain the second undetermined variable with the largest information value;

[0043] The fourth screening submodule is used to delete the second undetermined variable whose variance inflation coefficient is greater than the inflation coefficient threshold, and delete the second undetermined variable whose default probability is greater than the probability threshold, so as to obtain the input variable.

[0044] According to a fourth aspect of an embodiment of the present disclosure, a scoring training device is provided, for training the scoring model as described in the third aspect, comprising:

[0045] A data set establishment module, used to establish a training data set, wherein the training data set includes the user's input variables;

[0046] A marking module, used for marking the disconnection level of the user;

[0047] The training module is used to input the training data set into the scoring model for training to obtain a recommendation scoring model.

[0048] According to a fifth aspect of an embodiment of the present disclosure, a device for obtaining a user disconnection level is provided, characterized in that it includes:

[0049] processor;

[0050] a memory for storing instructions executable by the processor;

[0051] The processor is configured to execute the instructions to implement the method for obtaining the user disconnection level as described in any one of the first aspects above.

[0052] According to a sixth aspect of an embodiment of the present disclosure, a scoring model training device is provided, characterized in that it includes:

[0053] processor;

[0054] a memory for storing instructions executable by the processor;

[0055] Wherein, the processor is configured to execute the instructions to implement the scoring model training method as described in any one of the second aspects above.

[0056] According to the seventh aspect of an embodiment of the present disclosure, a non-temporary computer-readable storage medium is provided, characterized in that when the instructions in the storage medium are executed by a processor of a user loss level acquisition device, the user loss level acquisition device is enabled to execute a user loss level acquisition method as described in any one of the first aspects above.

[0057] According to an eighth aspect of an embodiment of the present disclosure, a non-temporary computer-readable storage medium is provided, characterized in that when the instructions in the storage medium are executed by a processor of a user loss level acquisition device, the user loss level acquisition device is enabled to execute a scoring model training method as described in any one of the second aspects above.

[0058] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0059] The first pending variable is obtained by preprocessing and deriving the user variable data, and the second pending variable is obtained after screening the first pending variable and screened to obtain the input variable. The user's disconnection level is determined according to the input variable and the scoring model. The embodiment of the present application can screen out variables that are more helpful for prediction, avoid ineffective work of staff, and more accurately predict the user's disconnection level.

[0060] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0062] Figure 1 The figure is a flow chart of a method for obtaining a user disconnection level according to an exemplary embodiment.

[0063] Figure 2 The figure is a flow chart of a method for obtaining a user disconnection level according to an exemplary embodiment.

[0064] Figure 3 The figure is a flow chart of a method for obtaining a user disconnection level according to an exemplary embodiment.

[0065] Figure 4 The figure is a flow chart of a method for obtaining a user disconnection level according to an exemplary embodiment.

[0066] Figure 5 The figure is a flowchart of a scoring model training method according to an exemplary embodiment.

[0067] Figure 6 The figure is an overall flow chart of a scoring model training according to an exemplary embodiment.

[0068] Figure 7 The invention is a flow chart showing a variable screening according to an exemplary embodiment.

[0069] Figure 8 The present invention is a block diagram of a device for obtaining a user disconnection level according to an exemplary embodiment.

[0070] Fig. 9 It is a block diagram of a scoring model training device according to an exemplary embodiment.

[0071] Fig.10 It is a block diagram of a device according to an exemplary embodiment. DETAILED DESCRIPTION

[0072] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0073] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation methods described in the following exemplary embodiments do not represent all implementation methods consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims. The acquisition, storage, use, processing, etc. of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0074] With the continuous development of financial technology, big data and artificial intelligence technologies are becoming more mature, and the collection business of commercial banks is also transforming towards intelligence, refinement and industrialization. In recent years, due to internal and external factors, the pressure on retail credit collection has increased and the loss rate has increased significantly. In traditional collection methods, many customers with overdue loans cannot be connected or have abnormal online status, making it impossible to reach the collection. At the same time, the collection staff need to call one by one to confirm that the customer is out of touch, which is time-consuming and laborious. For example, in the telephone collection scenario, the average caller can get a response from the customer within 3 calls on average. For lost customers, it is often impossible to get through after 6 or more calls. This has caused a huge waste of manpower and material costs. By verifying the customer's identity and mobile phone number status through internal and external data, and predicting the probability of account loss in advance, it will greatly facilitate the collection operation, improve the accuracy of the collection model and strategy, improve the collection contact effect, improve the collection recovery rate, and reduce the ineffective operation time of the collection staff (referred to as "three improvements and one reduction"). In the process of loss of connection repair, a loss of connection prediction model is established, and overdue customers are classified into loss of connection levels before electric reminders are made. Customers with high loss of connection rates are accurately captured for repair, which is conducive to improving repair efficiency and reducing the cost of operator verification.

[0075] How to classify overdue accounts into disconnection levels according to the probability of disconnection before electronic reminder is a technical problem to be solved by this application.

[0076] For the collection of overdue personal loan customers, the current system model is to develop corresponding collection queues for different product lines. In the collection queue, several telephone collection actions will be carried out for different types of overdue customers. From the first day of the customer's overdue payment to the set deadline for telephone collection, the established telephone collection queue will not be changed due to multiple unanswered calls to the account.

[0077] Business personnel put forward the following requirements for debt collection for lost customers: In order to improve debt collection efficiency, customers with different probability of losing contact are divided into lost contact levels, so as to design more targeted plans.

[0078] The method proposed in the present invention for classifying the loss of contact level of overdue accounts is as follows: through the telephone collection reach rate (whether the telephone collection is connected within one month of the overdue period), the number of telephone collection numbers dialed within one month of the overdue period, and the distribution of the number of calls, the current status of telephone collection is analyzed, and the modeling samples and sample label definitions are determined in combination with the business needs of the bank. The characteristics of overdue accounts are constructed through multiple dimensions such as customer basic information, historical repayment information, historical overdue information, and historical collection information. The scoring card model is established using the logistic regression algorithm, and the loss of contact probability of overdue accounts is scored and quantified according to the model. The decision tree algorithm is used to combine the model score and the strong predictive power variables, and finally the loss of contact level is divided.

[0079] In the existing method, within 30 days after the account is overdue, the main collection methods are SMS, intelligent voice, and manual outbound calls. The basic principle is to use SMS and intelligent voice in the early and middle stages of the overdue period, and manual outbound calls in the later stages. If the manual outbound call fails to reach the account on the same day, the account will continue to be pushed outbound calls the next day. If the manual outbound call fails to reach the account on the same day, the account will continue to be pushed outbound calls the next day.

[0080] Figure 1 FIG. 1 is a flow chart showing a method for obtaining a user disconnection level according to an exemplary embodiment. Figure 1 As shown, the method is used in data processing and comprises the following steps:

[0081] Step 101, obtaining user variable data, preprocessing and deriving the variable data to obtain a first pending variable.

[0082] In the embodiment of the present application, the user data includes multiple variable data, and the embodiment of the present application analyzes the disconnection level of each user based on the variable data. In a possible embodiment, the variable data includes the user's current status, repayment record, basic information, PBOC credit, expenditure information, etc.

[0083] The variable data is preprocessed and derived, such as converting some date data into the number of days remaining, whether the ID card is overdue at the end of the account period, etc. Screening is performed based on the missing rate and the maximum single value ratio to obtain the first pending variable. A large missing rate indicates that there is less valid data in the variable, and it is impossible to judge the loss level of all customers based on the variable data. A large maximum single value ratio indicates that the value of the variable data is relatively single, the discrimination is low, and it is not suitable to judge the loss level based on the variable data.

[0084] Regarding the definition of lost contact, according to the internal business definition, plus the limit on the number of days of calls. The final definition is that within 30 days after the overdue, the number of calls made by the intelligent outbound call and the manual outbound call is greater than or equal to 6 times, the number of days of calls is greater than or equal to 5 days, and none of them are connected. Lost contact; within 30 days after the overdue, the intelligent outbound call and the manual outbound call are connected at least once or there is no record of telephone reminders (repayment has been made before the telephone reminder) is not lost contact; within 30 days after the overdue, the intelligent outbound call and the manual outbound call are not connected, but the number of calls and the number of days do not reach the level of lost contact, which is a gray sample. In the statistics of the sample labels of mortgage products, the proportions of non-lost contact, lost contact, and gray samples are 65.16%, 12.06%, and 22.77% respectively. The distribution of the number of overdue days of samples in the three states is counted again. It can be seen that the average number of overdue days for samples without lost contact is 5 days, while that for lost contact samples is 45 days, and that for gray samples is 14 days. The rationality of the sample label definition can be verified by the distribution of overdue days. The lost samples have the largest number of overdue days, followed by the gray samples, and the non-lost samples have the smallest number of overdue days, which is consistent with business logic. The present embodiment of the application will exclude the gray samples, so only the lost and non-lost samples are considered.

[0085] Step 102: binning the first pending variables, and screening the first pending variables according to business logic to obtain second pending variables.

[0086] In the embodiment of the present application, in order to achieve data discretization, the first undetermined variable is binned, and binning is to group the user data corresponding to each first undetermined variable according to a specific rule, and the binning method includes chi-square binning, equidistant binning and equal frequency binning. According to the distribution characteristics of the binned data, it is judged whether the first undetermined variable conforms to the business logic, and the first undetermined variable that conforms to the business logic is screened to obtain the second undetermined variable.

[0087] In one possible embodiment, the binning is performed using a chi-square binning method, which is a bottom-up data discretization method. It relies on the chi-square test: adjacent intervals with the smallest chi-square value are merged together until a certain stopping criterion is met. For accurate discretization, the relative class frequencies should be completely consistent within an interval. Therefore, if two adjacent intervals have very similar class distributions, the two intervals can be merged; otherwise, they should be kept separate. Low chi-square values ​​indicate that they have similar class distributions.

[0088] Step 103, performing weight of evidence (WOE) mapping on the second undetermined variable, and screening the second undetermined variable according to correlation and information value to obtain model input variables.

[0089] In the embodiment of the present application, after the second undetermined variable is screened, WOE mapping is performed, WOE can reflect the correlation of the second undetermined variable, two second undetermined variables are selected to form a variable pair, and the variation or fluctuation of the WOE value of the second undetermined variable is combined with the coefficient fitted by the model to construct the correlation index of each variable pair. Generally, the larger the correlation index, the stronger the correlation of the second undetermined variable in the variable pair. At the same time, the variance inflation coefficient and default probability of the second undetermined variable are calculated for screening.

[0090] Step 104: input the input variables of the user into a scoring model to obtain the disconnection level of the user.

[0091] The final model input variables are screened out through the above steps, and then the scoring model is constructed. The data input into the scoring model is the data corresponding to the model input variables of the user. The scoring model is trained to more accurately determine the loss level of the user.

[0092] The embodiment of the present application obtains the first pending variable by preprocessing and deriving the user variable data, obtains the second pending variable after screening the first pending variable, and screens to obtain the input variable. The user's disconnection level is determined based on the input variable and the scoring model. The embodiment of the present application can screen out variables that are more helpful for prediction, avoid ineffective work by staff, and more accurately predict the user's disconnection level.

[0093] Figure 2 FIG. 1 is a flow chart showing a method for obtaining a user disconnection level according to an exemplary embodiment. Figure 2 As shown, the method is used in data processing and comprises the following steps:

[0094] Step 201, delete the variables whose missing rate is greater than or equal to the missing rate threshold, and determine the remaining variables as the first pending variables.

[0095] In the embodiment of the present application, some variables have a high missing rate and less data, and cannot be used as a scoring standard for the user's disconnection level. It is necessary to delete the variables with a high missing rate to improve the accuracy of the user's disconnection level prediction. In a possible embodiment, the missing rate threshold is 0.8.

[0096] Step 202, delete the variables whose maximum single value ratio is greater than or equal to the single value ratio threshold, and determine the remaining variables as the first pending variables.

[0097] In the embodiment of the present application, the maximum single value of some variables has a high proportion, the data is relatively single, and the discrimination is low, which cannot be used as a scoring standard for the user's disconnection level. It is necessary to delete the variables with a high maximum single value ratio to improve the accuracy of the user's disconnection level prediction. In a possible embodiment, the single value ratio threshold is 0.8. The second undetermined variable is obtained by screening according to the missing rate threshold and the single value ratio threshold.

[0098] By screening the first undetermined variable in the user data, the user's disconnection level is determined according to the first undetermined variable. The embodiment of the present application can screen out variables that are more helpful for prediction, avoid ineffective work of the staff, and more accurately predict the user's disconnection level.

[0099] Figure 3 FIG. 1 is a flow chart showing a method for obtaining a user disconnection level according to an exemplary embodiment. Figure 3 As shown, the method is used in data processing and comprises the following steps:

[0100] Step 301, binning samples corresponding to the first undetermined variable, and obtaining a population stability index and information value of the first undetermined variable;

[0101] In the embodiment of the present application, the second pending variable needs to be screened again, and the data in the second pending variable is first binned. The binning mainly considers the following factors: there are at least 50 bad samples in each box, and each box accounts for at least 2%. The monotonicity of the variables after binning is consistent with business significance, and there must be a certain difference between two adjacent boxes. The binning method of this project is the chi-square binning algorithm, which obtains the cut point of each group and divides the variables into several groups with predictive power.

[0102] Then use the PSI value test method to screen variables. Population Stability Index (PSI) is used to measure the migration of population distribution over time, reflecting the stability of the distribution of validation samples in each score segment and the distribution of modeling samples. The stability of population distribution in different periods is also the key to establishing and using scorecards. If the score distribution and the stability of the scoring model variables change significantly, it may mean that the model is not suitable for new customer groups.

[0103] Stability is calculated in the following way:

[0104]

[0105] Among them, a i is the number of accounts in the scoring interval in the development dataset (population 1); A is the total number of accounts in the development dataset (population 1); c i is the number of accounts in the scoring interval I in the most recent dataset (population 2); C is the total number of accounts in the most recent dataset (population 2).

[0106] The calculation formula of Information Value (IV) is as follows:

[0107]

[0108] When the IV value is too small, it means that the input variable has little effect on the target variable, that is, it does not have enough discrimination for the target variable. When the IV value is too large, this variable may have a decisive impact on the target variable. Although the discrimination is very good, this variable is likely to "deprive" other variables of their role. It may not be suitable for use in the model, but should be analyzed separately as a standard for data classification or an independent business rule, which can be used as a strategy.

[0109] Step 302: Screen the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable.

[0110] In the embodiment of the present application, the first undetermined variable is screened by setting a population stability threshold and an information value threshold to obtain the second undetermined variable.

[0111] Optionally, after the binning process, the number of bad samples in each bin is greater than a bad sample number threshold, and the ratio of the number of samples in each bin to the total number of samples is greater than a ratio threshold.

[0112] Optionally, the step of screening the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable includes:

[0113] The first undetermined variables whose group stability index is less than the group stability threshold are deleted, and the first undetermined variables whose information value is less than the information value threshold are deleted, and the remaining first undetermined variables are determined as second undetermined variables.

[0114] Figure 4 FIG. 1 is a flow chart showing a method for obtaining a user disconnection level according to an exemplary embodiment. Figure 4 As shown, the method is used in data processing and comprises the following steps:

[0115] Step 401, obtaining the correlation index of the variable pairs in the second undetermined variable, and calculating the variance inflation coefficient and default probability of the second undetermined variable;

[0116] In the embodiment of the present application, the weight of evidence WOE mapping is performed on the second undetermined variable, and the WOE binning calculation formula is:

[0117] f G (j) = Good%, f B (j)=Bad% that fall into bin j

[0118]

[0119] The WOE mapping is used for the estimated value of each variable in the regression result, that is, the coefficient of logistic regression, which is the scoring weight corresponding to the variable after conversion. The scoring weight of the variable multiplied by the WOE is the scoring weight corresponding to the variable bin. The sum of the scoring weights of all bins and the constant term is the scoring weight of the account, that is, the scoring result.

[0120] The larger the scoring weight of a bin, the lower the risk and the higher the corresponding model score. Conversely, the lower the scoring weight of a bin, the higher the risk and the lower the corresponding model score. Similarly, the higher the score of an account, the lower the risk of the account becoming a bad account.

[0121] After WOE mapping, two second undetermined variables are selected to form a variable pair, and the correlation index of the variable pair can be calculated. The higher the correlation index, the stronger the correlation of the second undetermined variables in the variable pair.

[0122] At the same time, the variance inflation coefficient and default probability of the second unknown variable are calculated according to the results of WOE mapping.

[0123] Step 402, obtaining a second variable pair whose correlation index is greater than a correlation threshold, and retaining the second undetermined variable with the largest information value.

[0124] In a possible embodiment, the correlation threshold is 0.7. If the correlation index of the variable pair is greater than 0.7, the information value IV of the second undetermined variable in the variable pair is compared, and the second undetermined variable with the largest information value is retained.

[0125] Step 403, delete the second undetermined variable whose variance inflation coefficient is greater than the inflation coefficient threshold, and delete the second undetermined variable whose default probability is greater than the probability threshold, so as to obtain the input variables.

[0126] In the embodiment of the present application, the variance inflation factor (VIF) is a measure of the severity of complex (multi) collinearity in a multivariate linear regression model. It represents the ratio of the variance of the regression coefficient estimate to the variance when the independent variables are assumed to be non-linearly correlated. Set the inflation factor threshold, delete the second undetermined variable whose vif is greater than the inflation factor threshold, and delete the variable whose default probability p value is greater than the probability threshold. The remaining variables are the variables required by the embodiment of the present application, that is, the model input variables.

[0127] After the above variables are screened, the remaining variables that are finally entered into the model are trained for logistic regression. The calibration of the score requires three parameters: standard score, standard odds, and PDO. PDO is the additional score required to double the odds. Standard odds refers to the odds corresponding to the standard score. PDO refers to how many points the score needs to be increased to make the corresponding odds twice the original. The results of the logistic regression are scored according to the score calibration formula. The score calibration formula is: Score = AB*log(odds). According to the standard score of 600 points, the standard odds of 19:1, and PDO equal to 50, the score is scored.

[0128] Figure 5 FIG. 1 is a flowchart of a scoring model training method according to an exemplary embodiment. Figure 5 As shown, the method is used in scoring model training, comprising the following steps:

[0129] Step 501, establishing a training data set, wherein the training data set includes the user's model input variables;

[0130] In an embodiment of the present application, a loss of connection level score of the user is obtained through a scoring model. The modeling algorithm of the scoring model is diverse. In a possible embodiment, the machine learning algorithms that can be selected include but are not limited to SVM, random forest, xgboost, lightgbm, neural network, etc.

[0131] After obtaining the first undetermined variable that is more closely related to the user's disconnection level through screening, a training data set can be constructed according to the first undetermined variable.

[0132] Step 502, marking the disconnection level of the user;

[0133] In the embodiment of the present application, the disconnection level of the user in the training data set is marked.

[0134] Step 503: input the training data set into the scoring model for training to obtain a recommendation scoring model.

[0135] In the embodiment of the present application, the labeled training data set is input into the scoring model for multiple rounds of training, and the hyperparameters in the scoring model are optimized to finally obtain the recommendation scoring model.

[0136] Optionally, inputting the training data set into the scoring model for training includes:

[0137] The scoring model is trained with the goal of maximizing the KS statistic and the area under the curve AUC.

[0138] In the embodiment of the present application, the two main indicators for testing the distinguishing ability of the scoring model are KS and AUC. From the results, the KS of the scoring model training set reached 0.5664, AUC was 0.8592, the KS of the validation set was 0.5633, AUC was 0.8554, and the KS of the cross-time test set was 0.5175, AUC was 0.8345, and the distinguishing effect was very obvious.

[0139] Next, the sorting ability of the scoring model is verified. As the score increases, the loss of contact rate within the group decreases monotonically, indicating that this model has good sorting ability. After the scoring model is verified to be problem-free, the model score can be used to divide the loss of contact level. During the modeling process of this scoring model, it was found that the IV of some historical overdue variables is particularly high. Such variables have a strong control over the model. If it shifts, it will have a serious impact on the entire model. Considering the robustness of the model, variables with particularly high IVs were not used during modeling. In order to further improve the effect of the loss of contact level division, the decision tree algorithm is used to combine the model score with the strong variable, and finally the loss of contact level is divided. This scoring model divides the score card scores into five levels: extremely low, low, medium, high, and extremely high according to a certain proportion.

[0140] Figure 6 FIG. 1 is an overall flow chart of a scoring model training according to an exemplary embodiment. Figure 6 As shown, first prepare modeling tools and user data. The modeling tools include: python3, jupyter notebook. Conduct business interviews and surveys to determine the business goals of modeling; determine the scope of data requirements and how to obtain them, understand the structure of the data, and verify the availability and credibility of relevant data.

[0141] Determine the modeling target. Regarding the definition of lost contact, it is defined according to the internal business definition, plus the limit on the number of days of calls. The final definition is that within 30 days after the overdue period, the number of calls made by the intelligent outbound call and the manual outbound call is greater than or equal to 6 times, the number of days of calls is greater than or equal to 5 days, and none of them are connected. Lost contact; within 30 days after the overdue period, the intelligent outbound call and the manual outbound call are connected at least once or there is no record of telephone reminder (repayment has been made before the telephone reminder) is not lost contact; within 30 days after the overdue period, the intelligent outbound call and the manual outbound call are not connected, but the number of calls and the number of days do not reach the level of lost contact, which is a gray sample. In the sample label statistics of mortgage products, the proportions of non-lost contact, lost contact, and gray samples are 65.16%, 12.06%, and 22.77% respectively. The distribution of the number of overdue days of samples in the three states is statistically analyzed. It can be seen that the average number of overdue days for samples without lost contact is 5 days, while that for lost contact samples is 45 days, and that for gray samples is 14 days. The rationality of the sample label definition can be verified by the distribution of overdue days. The lost samples have the largest number of overdue days, followed by the gray samples, and the non-lost samples have the smallest number of overdue days, which is consistent with business logic. The present embodiment of the application will exclude the gray samples, so only the lost and non-lost samples are considered.

[0142] After determining the user data and the modeling target, the model is developed, the variables are screened, and the scoring model is purchased. After the scoring model is constructed, it is trained with a training data set, and the effect of the scoring model is verified to obtain a recommended scoring model. The user's loss of contact level is divided according to the recommended scoring model.

[0143] Figure 7 FIG. 1 is a flow chart showing a variable screening according to an exemplary embodiment. Figure 7 As shown, first, we need to delete variables with missing value ratio ≥ 0.8 or maximum single value ratio ≥ 0.8, then bin the variables, calculate IV and PSI, delete variables with IV <= 0.02, delete variables with PSI > 0.1, and delete variables that do not conform to business explanations. Then we perform Woe mapping, correlation screening, VIF test, P-value test, stepwise regression and coefficient test. Finally, we adjust the screening variables based on business experience (business significance, future implementation costs).

[0144] Figure 8is a block diagram of a device for obtaining a user disconnection level according to an exemplary embodiment. Fig. 9 The device includes a data preprocessing module 810, a binning and screening module 820, a mapping and screening module 830 and a scoring module 840.

[0145] A preprocessing module 810 is used to obtain user variable data, preprocess and derive the variable data to obtain a first undetermined variable;

[0146] A binning and screening module 820, configured to bin the first pending variables and screen the first pending variables according to business logic to obtain second pending variables;

[0147] A mapping and screening module 830 is used to perform evidence weight mapping on the second undetermined variable, and screen the second undetermined variable according to relevance and information value to obtain a model input variable;

[0148] The scoring module 840 is used to input the input variables of the user into the scoring model to obtain the loss level of the user.

[0149] Fig. 9 is a block diagram of a scoring model training device according to an exemplary embodiment. Fig. 9 The device includes a data set establishment module 910, a labeling module 920 and a training module 930.

[0150] A data set establishment module 910 is used to establish a training data set, wherein the training data set includes the user's input variables;

[0151] A marking module 920, configured to mark the disconnection level of the user;

[0152] The training module 930 is used to input the training data set into the scoring model for training to obtain a recommendation scoring model.

[0153] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0154] Fig.10 FIG. 1 is a block diagram of a device 1000 for implementing the method of acquiring the user disconnection level according to an exemplary embodiment. Fig.10As shown, the device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0155] A number of components in the device 1000 are connected to the I / O interface 1005, including: an input unit 1006, such as a keyboard, a mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a disk, an optical disk, etc.; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0156] The computing unit 1001 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1001 performs the various methods and processes described above, such as a user disconnection level acquisition method. For example, in some embodiments, the user disconnection level acquisition method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the user disconnection level acquisition method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to execute the method for acquiring the user loss level in any other appropriate manner (for example, by means of firmware).

[0157] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0158] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0159] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0161] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0162] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0163] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0164] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for obtaining a user disconnection level, characterized in that: include: Obtain user variable data of the lost-connection sample and the non-lost-connection sample, and preprocess and derive the variable data to obtain a first pending variable; The first pending variable is binned and the first pending variable is screened according to business logic to obtain a second pending variable; Performing evidence weight mapping on the second undetermined variable, and screening the second undetermined variable according to relevance and information value to obtain a model entry variable; Inputting the input variables of the user into a scoring model to obtain the disconnection level of the user; The binning process of the first pending variable and screening the first pending variable according to the business logic to obtain the second pending variable includes: Binning samples corresponding to the first undetermined variable and obtaining a population stability index and information value of the first undetermined variable; The first undetermined variable is screened according to the group stability index and the information value to obtain the second undetermined variable.

2. The method according to claim 1, characterized in that The step of preprocessing and deriving the variable data to obtain a first undetermined variable includes at least one of the following: Delete the variables whose missing rate is greater than or equal to the missing rate threshold, and determine the remaining variables as the first pending variables; Variables whose maximum single value ratio is greater than or equal to the single value ratio threshold are deleted, and the remaining variables are determined as the first pending variables.

3. The method according to claim 1, characterized in that After the binning process, the number of bad samples in each bin is greater than a bad sample number threshold, and the ratio of the number of samples in each bin to the total number of samples is greater than a ratio threshold.

4. The method according to claim 1, characterized in that: The step of screening the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable includes: The first undetermined variables whose group stability index is less than the group stability threshold are deleted, and the first undetermined variables whose information value is less than the information value threshold are deleted, and the remaining first undetermined variables are determined as second undetermined variables.

5. The method according to claim 1, characterized in that The performing evidence weight mapping on the second undetermined variable and screening the second undetermined variable according to the correlation and information value to obtain a model input variable includes: Obtaining correlation indexes of variable pairs in the second undetermined variable, and calculating variance inflation coefficients and default probabilities of the second undetermined variable; Obtaining a second variable pair whose correlation index is greater than a correlation threshold, and retaining the second undetermined variable with the largest information value; Delete the second undetermined variable whose variance inflation coefficient is greater than the inflation coefficient threshold, and delete the second undetermined variable whose default probability is greater than the probability threshold, so as to obtain the model input variable.

6. A scoring model training method, used for training the method for obtaining the user disconnection level according to any one of claims 1 to 5, characterized in that: include: Establishing a training data set, wherein the training data set includes the user's model input variables; Marking the disconnection level of the user; The training data set is input into the scoring model for training to obtain a recommendation scoring model.

7. A device for obtaining a user disconnection level, characterized in that: include: A preprocessing module, used to obtain user variable data of disconnected samples and non-disconnected samples, and preprocess and derive the variable data to obtain a first undetermined variable; A binning and screening module, used to bin the first pending variable and screen the first pending variable according to business logic to obtain a second pending variable; A mapping and screening module, used for performing evidence weight mapping on the second undetermined variable, and screening the second undetermined variable according to relevance and information value to obtain a model input variable; A scoring module, used for inputting the input variables of the user into a scoring model to obtain the disconnection level of the user; The binning and screening module comprises: A binning module, used to bin samples corresponding to the first undetermined variable, and obtain a population stability index and an information value of the first undetermined variable; The first screening submodule is used to screen the first undetermined variable according to the group stability index and the information value to obtain the second undetermined variable.

8. The device according to claim 7, characterized in that The preprocessing module comprises: A first preprocessing submodule deletes variables whose missing rate is greater than or equal to a missing rate threshold, and determines the remaining variables as first pending variables; The second preprocessing submodule is used to delete the variables whose maximum single value ratio is greater than or equal to the single value ratio threshold, and determine the remaining variables as the first pending variables.

9. The device according to claim 7, characterized in that After the binning process, the number of bad samples in each bin is greater than a bad sample number threshold, and the ratio of the number of samples in each bin to the total number of samples is greater than a ratio threshold.

10. The device according to claim 7, characterized in that The screening submodule comprises: The screening unit is used to delete the first undetermined variables whose group stability index is less than the group stability threshold, and delete the first undetermined variables whose information value is less than the information value threshold, and determine the remaining first undetermined variables as second undetermined variables.

11. The device according to claim 7, characterized in that The mapping and screening module comprises: A second screening submodule, used to obtain the correlation index of the variable pairs in the second undetermined variable, and calculate the variance inflation coefficient and default probability of the second undetermined variable; A third screening submodule is used to obtain a second variable pair whose correlation index is greater than a correlation threshold, and retain the second undetermined variable with the largest information value; The fourth screening submodule is used to delete the second undetermined variable whose variance inflation coefficient is greater than the inflation coefficient threshold, and delete the second undetermined variable whose default probability is greater than the probability threshold, so as to obtain the input variable.

12. A scoring model training device, used for training the user disconnection level acquisition device according to any one of claims 7 to 11, characterized in that: include: A data set establishment module, used to establish a training data set, wherein the training data set includes the user's input variables; A marking module, used for marking the disconnection level of the user; The training module is used to input the training data set into the scoring model for training to obtain a recommendation scoring model.

13. A device for obtaining a user disconnection level, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method for obtaining the user loss level as described in any one of claims 1 to 5.

14. A scoring model training device, characterized in that: include: processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the scoring model training method as described in claim 6.

15. A non-transitory computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by the processor of the user loss level acquisition device, the user loss level acquisition device is enabled to execute the user loss level acquisition method according to any one of claims 1 to 5.

16. A non-transitory computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by the processor of the user loss level acquisition device, the user loss level acquisition device is enabled to execute the scoring model training method as described in claim 6.

Images