Overdue user management method and device and computer program product

Abstract

The invention discloses an overdue user management method and device and a computer program product. The method comprises the following steps: acquiring a plurality of pieces of first business handling data of a first user in a first time period; a pre-trained overdue user prediction model is used to analyze the plurality of first business handling data to obtain a first prediction overdue label of the first user in a second time period, the first prediction overdue label is used to represent whether the first user has an overdue repayment behavior in the second time period, and the first prediction overdue label is used to represent whether the first user has an overdue repayment behavior in the second time period; the second time period is any time period after the first time period, and the overdue user prediction model comprises at least one tree model and a logic regression device; and executing corresponding management measures on the first user according to the first predicted overdue label. The technical problem that the generalization ability of the overdue user prediction model is poor due to the fact that the number of sample data used by the overdue user prediction model during training is small and a large amount of missing data exists is solved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an overdue user management method and device, and a computer program product. Background Art

[0002] With the continuous development of Internet technology, Internet finance-related technologies are also constantly improving. As an important issue in the financial field, user loan delinquency prediction has also been proposed for many solutions at this stage, such as through statistics and machine learning algorithms (logistic regression algorithm, decision tree, support vector machine, random forest algorithm, etc.). However, this method is only suitable for scenarios with small sample sizes. When faced with massive data, complex features, and nonlinear relationships, the performance of this type of method is generally poor.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide an overdue user management method and device, and a computer program product to at least solve the technical problem that the overdue user prediction model has poor generalization ability due to the small number of sample data used in training the overdue user prediction model and the presence of a large amount of missing data.

[0005] According to one aspect of an embodiment of the present application, a method for managing overdue users is provided, including: obtaining multiple first business processing data of a first user within a first time period; using a pre-trained overdue user prediction model to analyze the multiple first business processing data to obtain a first predicted overdue label of the first user within a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior within the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; executing corresponding management measures on the first user according to the first predicted overdue label.

[0006] Optionally, the types of the first business processing data include: desensitized user data and desensitized transaction data, wherein the desensitized user data includes at least one of the following: user age, user gender, user account status; the desensitized transaction data includes at least one of the following: ordered product name, ordered product type, installment type, repaid amount, and amount to be repaid.

[0007] Optionally, the training process of the overdue user prediction model includes: obtaining multiple groups of sample data, wherein the sample data includes: the second business processing data of the second user in the third time period and the second overdue label of each second user in the fourth time period, wherein the fourth time period is any time period after the third time period; constructing an initial prediction model, wherein the initial prediction model includes at least one tree model and a logistic regressor, and the type of the tree model includes at least one of the following: lightGBM, RandomForest, CatBoost; based on the multiple groups of sample data, and using the k-fold cross-validation method to iteratively train the initial prediction model to obtain the overdue user prediction model.

[0008] Optionally, multiple groups of sample data are obtained, including: obtaining multiple initial business processing data corresponding to multiple second users in a third time period and a second overdue label of each second user in a fourth time period; preprocessing the multiple initial business processing data corresponding to multiple second users in the third time period, wherein the preprocessing includes at least one of the following: standardization processing, data filling, and data cleaning; constructing a target feature matrix based on the preprocessed multiple initial business processing data corresponding to multiple second users in the third time period and a preset Boolean matrix, wherein the Boolean matrix is ​​a matrix composed of 0 and 1 elements, and 1 indicates that the corresponding element is the initial business processing data, and 0 indicates that the corresponding element is the filled business processing data, and the target feature matrix includes the multiple second business processing data corresponding to each second user in the third time period; and the target feature matrix and the second overdue label of each second user in the fourth time period constitute multiple groups of sample data.

[0009] Optionally, a target feature matrix is ​​constructed based on a plurality of pre-processed initial business handling data respectively corresponding to a plurality of second users in a third time period and a preset Boolean matrix, including: forming an initial feature matrix based on a plurality of pre-processed initial business handling data respectively corresponding to a plurality of second users in a third time period, wherein the row dimension of the initial feature matrix represents the plurality of second users, and the column dimension represents the plurality of pre-processed initial business handling data of the second users in the third time period; iteratively updating the initial feature matrix using a gradient descent algorithm, and determining the target feature matrix through the updated initial feature matrix and the Boolean matrix.

[0010] Optionally, the initial feature matrix is ​​iteratively updated using a gradient descent algorithm, and the target feature matrix is ​​determined by the updated initial feature matrix and a preset Boolean matrix, including: determining a mapping matrix and a learning rate for mapping the initial feature matrix to the target feature matrix, and initializing the mapping matrix and the learning rate; in each round of iteration, the following steps are executed cyclically until a preset termination condition is met to obtain the target feature matrix of the last iteration round, including: calculating a descent gradient value of a first loss function in the current iteration round, wherein the first loss function is used to characterize the error between the overdue label and the second overdue label predicted by each row element in the target feature matrix for each second user in the fourth time period; updating the mapping matrix of the current iteration round based on the learning rate and the descent gradient value to obtain the mapping matrix of the next iteration round; performing a dot multiplication of the target feature matrix of the current iteration round with the Boolean matrix to obtain a first matrix, and performing a dot multiplication of the product of the mapping matrix of the current iteration round and the target feature matrix with the complement matrix of the Boolean matrix to obtain a second matrix, and taking the sum of the first matrix and the second matrix as the target feature matrix of the next iteration round.

[0011] Optionally, based on multiple groups of sample data, the initial prediction model is iteratively trained using a k-fold cross-validation method to obtain an overdue user prediction model, including: dividing the multiple groups of sample data into k subsets of equal size, wherein one subset is used to perform one round of iterative training on the initial prediction model; in each round of iteration, the following steps are executed cyclically until the model performance indicator reaches a preset performance indicator threshold, thereby obtaining an overdue user prediction model: using k-1 subsets as training sets, and iteratively training the initial prediction model according to the following method, including: for each group of sample data in the k-1 subsets, inputting multiple second business features in the sample data into each tree model in the initial prediction model, obtaining a second predicted overdue label output by each tree model in the initial prediction model, and inputting the second predicted overdue label output by each tree model into a logistic regression classifier in the initial prediction model, obtaining a first target predicted overdue label output by the logistic regression classifier in the initial prediction model; based on the multiple groups in the k-1 subsets A second loss function is constructed based on the second overdue label in the sample data and the corresponding second target predicted overdue label, and the model parameters of the initial prediction model are adjusted based on the second loss function to obtain an overdue user prediction model; the remaining 1 subset is used as a validation set, and the trained overdue user prediction model is verified according to the following method, including: for each group of sample data in 1 subset, multiple second business features in the sample data are input into each tree model in the overdue user prediction model to obtain a third predicted overdue label output by each tree model in the overdue user prediction model, and the third predicted overdue label output by each tree model is input into the logistic regression classifier in the overdue user prediction model to obtain the second target predicted overdue label output by the logistic regression classifier in the overdue user prediction model; model performance parameters are determined based on the second overdue labels in multiple groups of sample data in 1 subset and the corresponding second target predicted overdue labels, and when the model performance parameters do not reach the performance indicator threshold, the next round of iteration process is continued.

[0012] Optionally, the first predicted overdue label includes: a high-risk overdue label, a medium-risk overdue label or a low-risk overdue label, wherein corresponding management measures are performed on the first user according to the first predicted overdue label, including: when the first predicted overdue label is a high-risk overdue label, overdue reminder information is sent to the first user according to a preset first frequency, and an overdue reminder call is initiated to the first user according to a preset second frequency, wherein the first frequency is greater than the second frequency; when the first predicted overdue label is a medium-risk overdue label or a low-risk overdue label, overdue reminder information is periodically sent to the first user according to a preset third frequency, wherein the third frequency is less than the first frequency.

[0013] According to another aspect of an embodiment of the present application, an overdue user management device is also provided, including: an acquisition module, used to acquire multiple first business processing data of a first user within a first time period; a prediction module, used to analyze the multiple first business processing data using a pre-trained overdue user prediction model, and obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; a management module, used to execute corresponding management measures on the first user according to the first predicted overdue label.

[0014] According to another aspect of an embodiment of the present application, a computer program product is further provided. The computer program product includes a stored computer program, wherein the computer program implements the above-mentioned overdue user management method when executed by a processor.

[0015] In an embodiment of the present application, a plurality of first business processing data of a first user within a first time period are obtained; the plurality of first business processing data are analyzed using a pre-trained overdue user prediction model to obtain a first predicted overdue label of the first user within a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior within the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; corresponding management measures are executed on the first user according to the first predicted overdue label, thereby realizing intelligent and efficient management of the first user with overdue behavior, thereby solving the technical problem of poor generalization ability of the overdue user prediction model due to the small number of sample data used in training the overdue user prediction model and the presence of a large amount of missing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a hardware structure block diagram of an optional computer terminal (or mobile device) for implementing the overdue user management method according to an embodiment of the present application;

[0018] Figure 2 is a flowchart of an optional overdue user management method according to an embodiment of the present application;

[0019] Figure 3 is a structural diagram of an optional overdue user management device according to an embodiment of the present application;

[0020] Figure 4 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application 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 application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In addition, the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0024] Example 1

[0025] According to an embodiment of the present application, a method embodiment of an overdue user management method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1The hardware structure block diagram of a computer terminal (or mobile device) for implementing the overdue user management method is shown. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0027] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0028] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the overdue user management method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the overdue user management method of the above-mentioned application is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0029] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0031] In the above operating environment, Figure 2 is a flow chart of an optional overdue user management method according to an embodiment of the present application, such as Figure 2 As shown, the method at least includes steps S202-S206, wherein:

[0032] Step S202: Acquire a plurality of first service handling data of a first user within a first time period.

[0033] In the technical solution provided in step S202, the above-mentioned first business processing data is used to reflect the repayment status of the first user's installment business during the first time period. Therefore, the types of the first business processing data may include: desensitized user data (i.e., static attribute data) and desensitized transaction data (i.e., transaction attribute data), wherein the desensitized user data refers to the user's basic information and product information associated with the installment business, so the desensitized user data includes but is not limited to: user age, user gender, user account status, etc. The desensitized transaction data refers to the transaction behavior information of the first user during the first time period, so the desensitized user data includes but is not limited to: ordered product name, ordered product type, installment type, repaid amount, pending repayment amount, promotion store, promoter, etc.

[0034] Step S204: Analyze multiple first business processing data using the pre-trained overdue user prediction model to obtain a first predicted overdue label of the first user in the second time period.

[0035] In the technical solution provided in step S204, the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, wherein the second time period is any time period after the first time period. The overdue user prediction model used includes at least one tree model and a logistic regressor.

[0036] As an optional implementation, the training process of the overdue user prediction model may include:

[0037] Step S1, obtaining multiple groups of sample data.

[0038] The sample data includes: the second business handling data (i.e., input data) of the second user in the third time period and the second overdue label (i.e., output label) of each second user in the fourth time period. The second user and the first user are the same type of users, i.e., both are users who handle installment business, and the fourth time period is any time period after the third time period.

[0039] Step S2, constructing an initial prediction model.

[0040] Among them, since the embodiment of the present application is an overdue user prediction model, the sample data collected for it may contain more outliers and / or missing values, and the amount of collected sample data is small, so the performance of the tree model on these sample data will be due to the neural network model. Therefore, the embodiment of the present application adopts a gradient boosting decision tree (GradientBoosting Decision Tree, GBDT) model, so the above-mentioned initial prediction model includes at least one tree model and a logistic regressor, and the type of the tree model includes at least one of the following: lightGBM, RandomForest, CatBoost.

[0041] Step S3, based on multiple groups of sample data, the initial prediction model is iteratively trained using a k-fold cross-validation method to obtain an overdue user prediction model.

[0042] Optionally, in the technical solution provided in step S1 above, multiple groups of sample data may be obtained by the following method, including:

[0043] Step S11, obtaining a plurality of initial business handling data respectively corresponding to a plurality of second users in a third time period and a second overdue label of each second user in a fourth time period.

[0044] Step S12: pre-processing a plurality of initial service handling data corresponding to a plurality of second users in a third time period.

[0045] Specifically, since there may be missing data or abnormal data in the multiple initial business processing data corresponding to each second user in the third time period directly obtained, the collected data can be pre-processed in combination with existing business knowledge, such as filling in missing data, clearing abnormal data, and standardizing data from different units.

[0046] Taking missing value filling as an example, if data such as arrears, payment, ARPU value, etc. are missing, it can be considered that the missing data means the amount is zero; if data such as user account status is missing, it can be considered that the missing data means the status is normal; if data such as installment time, repayment end time, repayment time, etc. that are closely related to whether it is overdue are missing, they can be filled in based on the complementary relationship between time, such as using the payment time or installment time farthest from the current time as the filling field, and then extracting the year, month, day and time features, and performing statistical processing; if data such as age, gender, customer type, etc. are missing, zero values ​​or statistical values ​​can be used to fill in.

[0047] Step S13: constructing a target feature matrix based on the pre-processed multiple initial business handling data corresponding to the multiple second users in the third time period and a preset Boolean matrix.

[0048] The Boolean matrix (denoted as X1) is a matrix composed of 0 and 1 elements, and 1 indicates that the corresponding element is the real initial business processing data, and 0 indicates that the corresponding element is the filled business processing data. In addition, the row dimension of the target feature matrix represents multiple second users, and the column dimension represents multiple second business processing data. Therefore, the target feature matrix includes multiple second business processing data corresponding to each second user in the third time period.

[0049] Step S14, forming a plurality of groups of sample data from the target feature matrix and the second overdue tags of each second user in the fourth time period.

[0050] Furthermore, in the technical solution provided in the above step S13, the target feature matrix can be constructed by the following method, including:

[0051] Step S131, forming an initial feature matrix according to a plurality of pre-processed initial business handling data corresponding to a plurality of second users in a third time period.

[0052] The row dimension of the initial feature matrix (denoted as X0) represents multiple second users, and the column dimension represents multiple pre-processed initial business handling data of the second users in the third time period.

[0053] Step S132, iteratively updating the initial feature matrix using a gradient descent algorithm, and determining the target feature matrix through the updated initial feature matrix and the Boolean matrix.

[0054] It can be understood as: setting a mapping function as y = σ(z) = σ(wX T ), where w∈R k×m represents the mapping matrix (also called weight matrix), And z represents the linear combination of vector and weight (i.e. z = wX T), y indicates whether the second user has overdue repayment behavior in the fourth time period. At the same time, by introducing the cross entropy loss function, and sampling the adaptive iteration of the mapping matrix w by the following gradient descent method, the initial feature matrix is ​​updated; and then the target feature matrix is ​​determined by the updated initial feature matrix and the Boolean matrix.

[0055] Specifically, firstly, a mapping matrix and a learning rate for mapping the initial feature matrix to the target feature matrix are determined, and the mapping matrix and the learning rate are initialized, such as setting the mapping matrix w=0 and the learning rate α=0.1.

[0056] In each round of iteration, the following steps are executed repeatedly until the preset termination condition is met, and the target feature matrix of the last iteration round is obtained. The following takes the tth round of iteration as an example to briefly describe the above iteration process:

[0057] Step 1: Calculate the descent gradient of the first loss function in the current iteration.

[0058] Among them, the first loss function is used to characterize the error between the overdue label and the second overdue label of each second user in the fourth time period predicted by using each row element in the target feature matrix, so its expression can be written as:

[0059]

[0060] Where m is the number of second users in the initial feature matrix (i.e., the number of rows), y i represents the second overdue label of the second user of the i-th in the initial feature matrix, y i Represents the overdue label of the second user of the i-th in the predicted initial feature matrix.

[0061] Therefore, the above descent gradient value can be written as:

[0062] Step 2: Update the mapping matrix of the current iteration round based on the learning rate and the descent gradient value to obtain the mapping matrix of the next iteration round.

[0063] Therefore, the expression of the mapping matrix of the next iteration can be written as:

[0064]

[0065] Among them, w t represents the mapping matrix of the current iteration, α represents the learning rate, and w t+1 Represents the mapping matrix for the next iteration.

[0066] Step 3: Determine whether the current iteration round meets the preset termination condition, that is, calculate the iteration error || w t+1 -w t||2 and In ||w t+1 -w t ||2<0.1 and Stop the iteration when t=t+1 and execute the fourth step; otherwise, update the number of iterations, set t=t+1, and continue to execute the first step.

[0067] Step 4: Perform dot multiplication of the target feature matrix of the current iteration round with the Boolean matrix to obtain the first matrix, and perform dot multiplication of the product of the mapping matrix of the current iteration round and the target feature matrix with the complement matrix of the Boolean matrix to obtain the second matrix. The sum of the first matrix and the second matrix is ​​used as the target feature matrix of the next iteration round.

[0068] Therefore, the expression of the target feature matrix for the next iteration can be written as:

[0069]

[0070] Where mul represents the dot multiplication operation of the corresponding bits of the two matrices. represents the target feature matrix for the next iteration, represents the target feature matrix of the current iteration round, X1 represents the Boolean matrix, logical not X1 represents the complement matrix of the Boolean matrix, which is obtained by performing a logical NOT operation on the Boolean matrix.

[0071] Through the iterative process from the first step to the fourth step mentioned above, the target feature matrix of the last iteration round can be used as the target feature matrix to be determined in step S132. The multiple feature vectors contained in the target feature matrix obtained so far can comprehensively reflect the multiple second business processing data corresponding to each second user in the third time period.

[0072] Furthermore, in the technical solution provided in the above step S14, the initial prediction model can be iteratively trained by the following method, including:

[0073] Step S141, dividing the multiple groups of sample data into k subsets of the same size.

[0074] Each subset is used to perform one round of iterative training on the initial prediction model.

[0075] In each round of iteration, the following training steps and validation steps are executed cyclically until the model performance index reaches the preset performance index threshold, and finally the overdue user prediction model is obtained:

[0076] Step S142, using the k-1 subsets as training sets, and iteratively training the initial prediction model according to the following method, including:

[0077] First, for each set of sample data in the k-1 subsets, multiple second business features in the sample data are input into each tree model in the initial prediction model to obtain a second predicted overdue label output by each tree model in the initial prediction model;

[0078] The second predicted overdue label output by each tree model is input into the logistic regression classifier in the initial prediction model to obtain the first target predicted overdue label output by the logistic regression classifier in the initial prediction model.

[0079] Step S143, constructing a second loss function based on the second overdue labels in multiple groups of sample data in the k-1 subsets and the corresponding second target predicted overdue labels, and adjusting the model parameters of the initial prediction model based on the second loss function to obtain an overdue user prediction model.

[0080] Step S144, using the remaining subset as a validation set, and validating the trained overdue user prediction model according to the following method, including:

[0081] Step 1: for each set of sample data in a subset, multiple second business features in the sample data are input into each tree model in the overdue user prediction model to obtain a third predicted overdue label output by each tree model in the overdue user prediction model;

[0082] Step 2: Input the third predicted overdue label output by each tree model into the logistic regression classifier in the overdue user prediction model to obtain the second target predicted overdue label output by the logistic regression classifier in the overdue user prediction model.

[0083] Step S145, determining the model performance parameters based on the second overdue labels in the multiple groups of sample data in one subset and the corresponding second target predicted overdue labels, and continuing to execute the next round of iteration process when the model performance parameters do not reach the performance indicator threshold.

[0084] Among them, the types of multiple tree models in the above-mentioned initial prediction model select the optimal model combination by setting up parameter tuning, early_stop mechanism, recall rate, accuracy rate and other indicators. Usually, three common tree types, lightGBM, RandomForest and CatBoost, can be selected to integrate machine learning algorithms. Accordingly, the machine learning libraries LGBMClassifier, XGBClassifier and CatBoostClassifier can be called to select default parameters as a set of regular hyperparameters.

[0085] It should be noted that in step S3, the overdue user prediction model can be loaded into the memory, for example, the raw data of the overdue user prediction model can be loaded from the non-volatile memory into the volatile memory, so that the processor runs the overdue user prediction model. The raw data of the overdue user prediction model refers to unprocessed data, usually including parameters and structural data of the overdue user prediction model. The structural data can be a calculation relationship based on the parameters, such as the forward propagation calculation relationship between the intermediate layers and between the neurons. Specifically, the structural data can include the code related to the structure of the first neural network, such as the code for performing related calculations between the intermediate layers and between the neurons.

[0086] In one embodiment, an area for loading the overdue user prediction model can be divided in the memory, which can include a structure data storage area and a parameter storage area. The structure data storage area is used to store structure-related codes, and the parameters referenced by it can point to the address of a specific parameter in the parameter storage area through a pointer. During the training process of the overdue user prediction model, it may be necessary to frequently update the parameters, and the parameter values ​​in the parameter storage area can be updated.

[0087] The following comparative experimental results in Table 1 illustrate the advantages of the overdue user prediction model provided by the embodiment of the present application in terms of model accuracy and generalization ability.

[0088] First, the sample data obtained in the above steps after being filled with the adaptive feature matrix (i.e., multiple groups of sample data consisting of the target feature matrix and the second overdue labels of each second user in the fourth time period) are divided into 5 equal parts, and in each round of training, 4 of them are selected as sample data subsets for model training, and the remaining 1 sample data subset is used for model verification. Then, different prediction models, such as LightGBM, random forest model and the overdue user prediction model provided in the embodiment of the present application, are used to predict the 4 sample data subsets to obtain the F1 scores of each prediction method, as shown in Table 1 below.

[0089]

[0090] In addition, the overdue user prediction model provided in the embodiment of the present application is tested using the remaining 1 piece of data and the sample data filled with business knowledge (the number of sample data is the same as the number of samples contained in the remaining 1 piece of sample data subset) to obtain the F1 score of each prediction method, as shown in Table 2 below.

[0091] Table 2

[0092]

[0093] It is not difficult to see from the above Tables 1 and 2 that the overdue user prediction model provided in the embodiment of the present application has achieved good results in terms of performance indicators on the validation set and the test set. In other words, the overdue user prediction model provided in the embodiment of the present application has better accuracy and generalization performance.

[0094] Step S206: executing corresponding management measures on the first user according to the first predicted overdue label.

[0095] Optionally, in the technical solution provided in step S206, the method may include:

[0096] When the first predicted overdue label is a high-risk overdue label, overdue reminder information is sent to the first user at a preset first frequency, and an overdue reminder call is initiated to the first user at a preset second frequency, wherein the first frequency is greater than the second frequency.

[0097] For example, SMS messages can be pushed to users with high-risk overdue labels on the 2nd and 15th of each month, with the SMS content template being "Hello user xx! Please pay off the overdue fees for the month before the 5th (or 18th)." At the same time, intelligent voice calls are made to users who are still labeled as high-risk on the 8th of each month to guide them to pay off their outstanding phone bills.

[0098] When the first predicted overdue label is a medium-risk overdue label or a low-risk overdue label, overdue reminder information is periodically sent to the first user according to a preset third frequency, wherein the third frequency is lower than the first frequency.

[0099] For example, set up SMS push to users with high-risk overdue labels on the 2nd of each month, and the SMS content template is "Hello user xx! Please pay the overdue fees for the month before the 5th."

[0100] Based on the scheme defined by the above steps S202 to S206, it can be known that in the embodiment, based on multiple groups of sample data of the target feature matrix, the initial prediction model is iteratively trained using the k-fold cross-validation method to obtain an overdue user prediction model, and the overdue user prediction model is used to analyze multiple first business handling data to obtain the first predicted overdue label of the first user in the second time period, and the corresponding management measures are performed on the first user according to the first predicted overdue label. Among them, the target feature matrix is ​​constructed by using the multiple initial business handling data corresponding to the multiple second users in the third time period and the second overdue labels of each second user in the fourth time period obtained in advance and combined with the Boolean matrix, so the generalization ability and model prediction accuracy of the trained overdue user prediction model are relatively high. Therefore, the overdue user prediction model provided in the embodiment of the present application can effectively solve the technical problem that the generalization ability of the overdue user prediction model is poor due to the small number of sample data used in the training of the overdue user prediction model and the presence of a large amount of missing data, thereby realizing intelligent and efficient management of the first user with overdue behavior.

[0101] Example 2

[0102] Based on Embodiment 1 of the present application, an embodiment of an overdue user management device is also provided, and the overdue user management method of the above embodiment is executed when the device is running. Figure 3 is a schematic diagram of the structure of an optional overdue user management device according to an embodiment of the present application, such as Figure 3 As shown, the overdue user management device at least includes an acquisition module 32, a prediction module 34 and a management module 36, wherein:

[0103] An acquisition module 32, configured to acquire a plurality of first service handling data of a first user within a first time period;

[0104] The prediction module 34 is used to analyze the plurality of first business handling data using the pre-trained overdue user prediction model to obtain a first predicted overdue label of the first user in the second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor;

[0105] The management module 36 is used to execute corresponding management measures on the first user according to the first predicted overdue label.

[0106] It should be noted that the various modules in the above-mentioned overdue user management device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0107] Example 3

[0108] According to an embodiment of the present application, a non-volatile storage medium is further provided, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the overdue user management method in Example 1.

[0109] Optionally, the device where the non-volatile storage medium is located implements the following steps by running the program: obtaining multiple first business processing data of the first user in a first time period; using a pre-trained overdue user prediction model to analyze the multiple first business processing data to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; executing corresponding management measures on the first user according to the first predicted overdue label.

[0110] According to an embodiment of the present application, a computer program product is also provided, which includes a stored computer program, wherein when the computer program is executed by a processor, the overdue user management method in Example 1 is implemented.

[0111] Optionally, the computer program executes the following steps: obtaining multiple first business processing data of the first user in a first time period; using a pre-trained overdue user prediction model to analyze the multiple first business processing data to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; executing corresponding management measures on the first user according to the first predicted overdue label.

[0112] According to an embodiment of the present application, a processor is also provided, which is used to run a program, wherein the overdue user management method in Example 1 is executed when the program is running.

[0113] Optionally, when the program is running, the following steps are executed: obtaining multiple first business processing data of the first user in a first time period; using a pre-trained overdue user prediction model to analyze the multiple first business processing data to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; executing corresponding management measures on the first user according to the first predicted overdue label.

[0114] According to an embodiment of the present application, an electronic device is also provided, wherein: Figure 4 is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application, such as Figure 4 As shown, the electronic device includes one or more processors; a memory for storing one or more programs, which, when the one or more programs are executed by the one or more processors, enables the one or more processors to run the programs, wherein the programs are configured to execute the overdue user management method in the above-mentioned embodiment 1 when running.

[0115] Optionally, the processor is configured to execute the following steps through a computer program: obtaining multiple first business processing data of the first user in a first time period; using a pre-trained overdue user prediction model to analyze the multiple first business processing data to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period is any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; executing corresponding management measures on the first user according to the first predicted overdue label.

[0116] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0117] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0119] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0120] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or CD-ROM and other media that can store program codes.

[0122] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for managing overdue users, characterized in that: include: Acquire multiple first business handling data of a first user in a first time period; Analyzing the plurality of first business handling data using a pre-trained overdue user prediction model to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period being any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; Execute corresponding management measures on the first user according to the first predicted overdue label.

2. The method according to claim 1, characterized in that The types of the first business processing data include: desensitized user data and desensitized transaction data, wherein: The desensitized user data includes at least one of the following: user age, user gender, user account status; The desensitized transaction data includes at least one of the following: the name of the ordered product, the type of ordered product, the type of installment payment, the amount repaid, and the amount to be repaid.

3. The method according to claim 1, characterized in that The training process of the overdue user prediction model includes: Acquire multiple groups of sample data, wherein the sample data includes: second business handling data of the second user in a third time period and second overdue labels of each of the second users in a fourth time period, wherein the fourth time period is any time period after the third time period; Constructing an initial prediction model, wherein the initial prediction model includes at least one tree model and a logistic regressor, and the type of the tree model includes at least one of the following: lightGBM, RandomForest, CatBoost; Based on the multiple groups of sample data, the initial prediction model is iteratively trained using a k-fold cross-validation method to obtain the overdue user prediction model.

4. The method according to claim 3, characterized in that Get multiple sets of sample data, including: Acquire a plurality of initial business handling data respectively corresponding to a plurality of the second users in a third time period and a second overdue label of each of the second users in a fourth time period; Preprocessing a plurality of initial business handling data respectively corresponding to a plurality of the second users in a third time period, wherein the preprocessing includes at least one of the following: standardization processing, data filling, and data cleaning; Constructing a target feature matrix based on the pre-processed multiple initial business handling data corresponding to the multiple second users in the third time period and a preset Boolean matrix, wherein the Boolean matrix is ​​a matrix composed of 0 and 1 elements, and 1 indicates that the corresponding element is the initial business handling data, and 0 indicates that the corresponding element is the filled business handling data, and the target feature matrix includes the multiple second business handling data corresponding to each of the second users in the third time period; The multiple groups of sample data are composed of the target feature matrix and the second overdue tags of each of the second users in a fourth time period.

5. The method according to claim 4, characterized in that Constructing a target feature matrix based on a plurality of pre-processed initial business handling data respectively corresponding to a plurality of the second users in a third time period and a preset Boolean matrix, including: An initial feature matrix is ​​formed according to the pre-processed initial business handling data respectively corresponding to the plurality of second users in the third time period, wherein the row dimension of the initial feature matrix represents the plurality of second users, and the column dimension represents the pre-processed initial business handling data of the second users in the third time period; The initial feature matrix is ​​iteratively updated using a gradient descent algorithm, and the target feature matrix is ​​determined by using the updated initial feature matrix and the Boolean matrix.

6. The method according to claim 5, characterized in that The initial feature matrix is ​​iteratively updated using a gradient descent algorithm, and the target feature matrix is ​​determined by using the updated initial feature matrix and a preset Boolean matrix, including: Determining a mapping matrix and a learning rate for mapping the initial feature matrix to a target feature matrix, and initializing the mapping matrix and the learning rate; In each round of iteration, the following steps are executed cyclically until the preset termination condition is met, and the target feature matrix of the last iteration round is obtained, including: Calculating a descending gradient value of a first loss function in a current iteration round, wherein the first loss function is used to characterize an error between an overdue label predicted by using each row element in the target feature matrix for each second user in a fourth time period and the second overdue label; The mapping matrix of the current iteration round is updated based on the learning rate and the descent gradient value to obtain the mapping matrix of the next iteration round; The target feature matrix of the current iteration round is point-multiplied with the Boolean matrix to obtain a first matrix, and the product of the mapping matrix of the current iteration round and the target feature matrix is ​​point-multiplied with the complement matrix of the Boolean matrix to obtain a second matrix, and the sum of the first matrix and the second matrix is ​​used as the target feature matrix of the next iteration round.

7. The method according to claim 3, characterized in that Based on the multiple groups of sample data, the initial prediction model is iteratively trained using a k-fold cross validation method to obtain the overdue user prediction model, including: Dividing the multiple groups of sample data into k subsets of equal size, wherein one subset is used to perform one round of iterative training on the initial prediction model; In each round of iteration, the following steps are executed cyclically until the model performance index reaches the preset performance index threshold, thereby obtaining the overdue user prediction model: The k-1 subsets are used as training sets, and the initial prediction model is iteratively trained according to the following method, including: for each group of sample data in the k-1 subsets, multiple second business features in the sample data are input into each tree model in the initial prediction model to obtain a second predicted overdue label output by each tree model in the initial prediction model, and the second predicted overdue label output by each tree model is input into a logistic regression classifier in the initial prediction model to obtain a first target predicted overdue label output by the logistic regression classifier in the initial prediction model; Constructing a second loss function based on the second overdue labels in the multiple groups of sample data in the k-1 subsets and the corresponding second target predicted overdue labels, and adjusting the model parameters of the initial prediction model based on the second loss function to obtain the overdue user prediction model; The remaining 1 subset is used as a validation set, and the overdue user prediction model obtained through training is validated according to the following method, including: for each group of sample data in the 1 subset, a plurality of second business features in the sample data are input into each tree model in the overdue user prediction model to obtain a third predicted overdue label output by each tree model in the overdue user prediction model, and the third predicted overdue label output by each tree model is input into a logistic regression classifier in the overdue user prediction model to obtain a second target predicted overdue label output by the logistic regression classifier in the overdue user prediction model; The model performance parameters are determined based on the second overdue labels in the multiple groups of sample data in the one subset and the corresponding second target predicted overdue labels, and when the model performance parameters do not reach the performance indicator threshold, the next round of iteration process is continued.

8. The method according to claim 1, characterized in that The first predicted overdue label includes: a high-risk overdue label, a medium-risk overdue label, or a low-risk overdue label, wherein executing corresponding management measures on the first user according to the first predicted overdue label includes: When the first predicted overdue label is the high-risk overdue label, sending overdue reminder information to the first user at a preset first frequency, and initiating an overdue reminder call to the first user at a preset second frequency, wherein the first frequency is greater than the second frequency; When the first predicted overdue label is the medium-risk overdue label or the low-risk overdue label, overdue reminder information is periodically sent to the first user according to a preset third frequency, wherein the third frequency is lower than the first frequency.

9. An overdue user management device, characterized in that: include: An acquisition module, used to acquire a plurality of first business handling data of a first user within a first time period; A prediction module, used to analyze the plurality of first business handling data using a pre-trained overdue user prediction model to obtain a first predicted overdue label of the first user in a second time period, wherein the first predicted overdue label is used to characterize whether the first user has overdue repayment behavior in the second time period, the second time period being any time period after the first time period, and the overdue user prediction model includes at least one tree model and a logistic regressor; A management module is used to execute corresponding management measures on the first user according to the first predicted overdue label.

10. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, the overdue user management method described in any one of claims 1 to 8 is implemented.

Images