An overdue payment reminder method and device

Abstract

An embodiment of the present application provides an overdue payment reminder method and device. The method includes: obtaining initial features of multiple preset feature fields corresponding to each user data within a preset time period; screening, from the initial features of the multiple preset feature fields, the initial features of the preset feature fields with importance greater than a preset importance threshold as target features; respectively inputting the target features corresponding to each user data into a preset high-probability overdue payment recovery model to obtain the predicted probability of overdue payment recovery for each user; determining a user list of target users according to the user information of the target users whose predicted probability is greater than a preset probability threshold; and sending an overdue payment reminder message to the target users according to the user list of the target users. By applying the technical solution provided by the embodiment of the present application, limited human and material resources can be concentrated on users with a relatively high probability of overdue payment recovery, so as to save resources, improve the success rate of overdue payment collection, and reduce the disturbance to the user side.

Description

Technical Field

[0001] This application relates to the technical fields of big data and artificial intelligence, and particularly to an overdue payment reminder method and device. Background Art

[0002] With the continuous expansion of mobile communication services, the number of mobile communication users is increasing day by day. Consequently, the problem of overdue payments among mobile communication users is becoming increasingly serious. To recover overdue payments from users, mobile communication operators send overdue payment reminder messages to all overdue users. This method of overdue payment reminder consumes a large amount of resources and the effect of overdue payment collection is not ideal. In addition, mobile communication operators send overdue payment reminder messages to all overdue users, that is, repeated reminders, which cause harassment to users. Summary of the Invention

[0003] The purpose of the embodiments of this application is to provide an overdue payment reminder method and device to improve the success rate of overdue payment collection and reduce the disturbance to the user side. The specific technical solutions are as follows:

[0004] In a first aspect, the embodiments of this application provide an overdue payment reminder method, and the method includes:

[0005] Obtain the initial features of multiple preset feature fields corresponding to each user data within a preset time period;

[0006] From the initial features of the multiple preset feature fields, screen out the initial features of the preset feature fields whose importance is greater than a preset importance threshold as target features;

[0007] Input the target features corresponding to each user data into a preset high-probability overdue payment recovery model respectively to obtain the predicted probability of overdue payment recovery for each user. The preset high-probability overdue payment recovery model is a model based on the Light Gradient Boosting Machine (LightGBM);

[0008] Determine the user list of the target users according to the user information of the target users whose predicted probability is greater than a preset probability threshold;

[0009] Send overdue payment reminder messages to the target users according to the user list of the target users.

[0010] Optionally, the step of obtaining the initial features of multiple preset feature fields corresponding to each user data within a preset time period includes:

[0011] Extract the original features of multiple preset feature fields corresponding to each user data within a preset time period;

[0012] Delete the abnormal data in the original features of the multiple preset feature fields extracted, and complete the missing features in the original features of the multiple preset feature fields extracted to obtain the processed features of the multiple preset feature fields;

[0013] Convert the processed features of multiple preset feature fields into numerical data to obtain the initial features of multiple preset feature fields corresponding to each user data.

[0014] Optionally, the step of screening the initial features of the preset feature fields with importance greater than a preset importance threshold from the initial features of the multiple preset feature fields as target features includes:

[0015] Use the random forest algorithm to determine the average value of the reduction in the average error after node classification of each initial feature in the decision tree;

[0016] Take the average value of the reduction in the average error of each initial feature as the importance of this initial feature, and screen the initial features of the preset feature fields with importance greater than the preset importance threshold as target features.

[0017] Optionally, the user information includes multi-dimensional data;

[0018] The step of determining the user list of the target user according to the user information of the target user with a prediction probability greater than a preset probability threshold includes:

[0019] Extract the multi-dimensional data of the target user that matches the preset label caliber from the user information of the target user with a prediction probability greater than the preset probability threshold to obtain the user list of the target user.

[0020] Optionally, the step of sending an overdue reminder message to the target user according to the user list of the target user includes:

[0021] Send the user list of the target user to the contact point that matches the type of the target user, so that when the contact point determines that the target user is in an overdue state, an overdue reminder message is sent to the target user according to the user list of the target user.

[0022] Optionally, the method further includes:

[0023] Divide the target features of the multiple preset feature fields into a training set and a test set;

[0024] The step of inputting the target features corresponding to each user data into a preset high-probability overdue recovery model to obtain the prediction probability of overdue recovery for each user includes:

[0025] Use the target features corresponding to each user data included in the training set to train the preset high-probability overdue recovery model;

[0026] Input the target features corresponding to each user data included in the test set into the pre-trained preset high-probability overdue recovery model respectively, to obtain the predicted probability of overdue recovery for each user.

[0027] In a second aspect, an embodiment of the present application provides an overdue reminder device, and the device includes:

[0028] An acquisition unit, configured to acquire initial features of multiple preset feature fields corresponding to each user data within a preset time period;

[0029] A screening unit, configured to screen out the initial features of the preset feature fields with importance greater than a preset importance threshold from the initial features of the multiple preset feature fields as target features;

[0030] A prediction unit, configured to input the target features corresponding to each user data into a preset high-probability overdue recovery model respectively, to obtain the predicted probability of overdue recovery for each user, and the preset high-probability overdue recovery model is a model based on the weak gradient booster LightGBM;

[0031] A determination unit, configured to determine a user list of the target users according to the user information of the target users with prediction probabilities greater than a preset probability threshold;

[0032] A sending unit, configured to send an overdue reminder message to the target users according to the user list of the target users.

[0033] Optionally, the acquisition unit is specifically configured to:

[0034] Extract the original features of multiple preset feature fields corresponding to each user data within a preset time period;

[0035] Delete abnormal data in the extracted original features of the multiple preset feature fields, and complete missing features in the extracted original features of the multiple preset feature fields, to obtain processed features of the multiple preset feature fields;

[0036] Convert the processed features of the multiple preset feature fields into numerical data, to obtain the initial features of multiple preset feature fields corresponding to each user data.

[0037] Optionally, the screening unit is specifically configured to:

[0038] Use the random forest algorithm to determine the average value of the reduction in the average error after node classification of each initial feature in the decision tree;

[0039] Take the average value of the reduction in the average error of each initial feature as the importance of the initial feature, and screen out the initial features of the preset feature fields with importance greater than a preset importance threshold as target features.

[0040] Optionally, the user information includes multi-dimensional data;

[0041] The determining unit is specifically configured to extract multi-dimensional data of the target user that matches the preset label caliber from the user information of the target user whose prediction probability is greater than the preset probability threshold, so as to obtain the user list of the target user.

[0042] Optionally, the sending unit is specifically configured to:

[0043] Send the user list of the target user to a contact point that matches the type of the target user, so that when the contact point determines that the target user is in an overdue state, an overdue reminder message is sent to the target user according to the user list of the target user.

[0044] Optionally, the device further includes:

[0045] A dividing unit, configured to divide the target features of the multiple preset feature fields into a training set and a test set;

[0046] The predicting unit is specifically configured to:

[0047] Use the target features corresponding to each user data included in the training set to train a preset high-probability overdue recovery model;

[0048] Input the target features corresponding to each user data included in the test set into the trained preset high-probability overdue recovery model respectively, to obtain the prediction probability of overdue recovery for each user.

[0049] In a third aspect, an embodiment of the present application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0050] The memory is used to store a computer program;

[0051] The processor is configured to, when executing the program stored in the memory, implement the steps of any of the above-mentioned overdue reminder methods.

[0052] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the above-mentioned overdue reminder methods are implemented.

[0053] In a fifth aspect, an embodiment of the present application provides a computer program product containing instructions, which when running on a computer, causes the computer to execute any of the above-mentioned overdue reminder methods.

[0054] Advantageous effects of the embodiments of the present application:

[0055] In the technical solution provided by the embodiments of the present application, by using the characteristics of the unique feature fields of users with overdue fees and combining with a preset high-probability overdue fee recovery model, users with a relatively high probability of overdue fee recovery are determined. Then, corresponding reminders are given to the overdue users among these users with a relatively high probability of overdue fee recovery, which improves the success rate of overdue fee collection. This way of overdue fee reminder can achieve precise collection, without having to repeatedly collect overdue fees from all overdue users, reducing the interference to the user side, and concentrating limited human and material resources on users with a relatively high probability of overdue fee recovery, reducing resource consumption.

[0056] Of course, when implementing any product or method of the present application, it is not necessarily required to simultaneously achieve all the above-mentioned advantages. Brief Description of the Drawings

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other embodiments can also be obtained based on these drawings.

[0058] Figure 1 It is the first flowchart of the overdue fee reminder method provided by the embodiments of the present application;

[0059] Figure 2 It is a schematic diagram of the accuracy rate of user collection provided by the embodiments of the present application;

[0060] Figure 3 It is a schematic diagram of the overdue fee recovery situation provided by the embodiments of the present application;

[0061] Figure 4 It is the second flowchart of the overdue fee reminder method provided by the embodiments of the present application;

[0062] Figure 5 It is the third flowchart of the overdue fee reminder method provided by the embodiments of the present application;

[0063] Figure 6 It is a flowchart of the prediction of the overdue fee recovery probability based on the LightGBM-HAR model provided by the embodiments of the present application;

[0064] Figure 7 A schematic structural diagram of the overdue fee reminder device provided by the embodiments of the present application;

[0065] Figure 8 A schematic structural diagram of the electronic device provided by the embodiments of the present application. Detailed Embodiments

[0066] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art based on the present application belong to the scope of protection of the present application.

[0067] With the continuous expansion of mobile communication services, the number of mobile communication users is increasing day by day. As a result, the situation of users' arrears in mobile communication is becoming increasingly serious. To recover the users' arrears, mobile communication operators send arrears reminder messages to all users with arrears. This method of arrears reminder consumes a large amount of resources and the effect of arrears collection is not ideal. In addition, mobile communication operators send arrears reminder messages to all users with arrears, that is, repeated collection, which causes harassment to users.

[0068] To solve the above problems, the embodiments of the present application provide an arrears reminder method, which can be applied to electronic devices such as servers, PCs (Personal Computers), tablets, and mobile terminals. For the convenience of understanding, the following takes the electronic device as the execution subject for explanation, which does not play a limiting role.

[0069] In this method, the electronic device uses the characteristics of the unique feature fields of users with arrears, combines a preset high-probability arrears recovery model, determines the users with a higher probability of arrears recovery, and then gives corresponding reminders to the users with arrears among these users with a higher probability of arrears recovery, improving the success rate of arrears collection. This method of arrears reminder can achieve accurate collection, without repeatedly reminding all users with arrears, reducing the disturbance to the user side, and concentrating limited human and material resources on the users with a higher probability of arrears recovery, reducing resource consumption.

[0070] The following will specifically describe the arrears reminder method provided by the embodiments of the present application through specific embodiments.

[0071] As Figure 1 shown, it is the first flow diagram of the arrears reminder method provided by the embodiments of the present application. The method includes the following steps:

[0072] Step S11, obtain the initial characteristics of multiple preset feature fields corresponding to each user data within a preset time period.

[0073] Among them, the length of the preset time period can be set according to actual needs. For example, the preset time period can be 1 month, 3 months, 6 months, etc. The preset feature fields can include, but are not limited to, account number, user identifier, user age, delinquent account recovery user, local network code, user ID type, the mobile voice minutes of the user in the last three months, the average mobile data traffic of the user in the last three months, the longest delinquent account age of all users of the user, etc. The initial features of each preset feature field can include, but are not limited to, sum, mean, variance, maximum value, minimum value, 25% size, 50% size, and 75% size, etc.

[0074] When performing delinquent account reminders, for each user, the electronic device obtains the initial features of multiple preset feature fields corresponding to the user data within the preset time period. For the convenience of data management, the electronic device can construct a wide table data set based on the obtained initial features of multiple preset feature fields, and use the wide table data set to manage the initial features of multiple preset feature fields corresponding to each user's data.

[0075] In an optional embodiment, the length of the preset time period includes multiple statistical cycles. For example, the statistical cycle is 1 month and the length of the preset time period is 3 months. In this case, for each statistical cycle, the electronic device can obtain the initial features of multiple preset feature fields corresponding to each user's data within the statistical cycle, and construct a wide table data set. For multiple statistical cycles, the electronic device obtains multiple wide table data sets in total. The electronic device can merge these multiple wide table data sets for subsequent processing.

[0076] Step S12: From the initial features of multiple preset feature fields, filter out the initial features of the preset feature fields whose importance is greater than the preset importance threshold as the target features.

[0077] After obtaining the initial features, the electronic device can screen out the feature data with high correlation with the model target according to the feature importance analysis method, merge the feature data with high correlation, and construct new features as needed to achieve data dimensionality reduction. The above feature importance analysis method can be the Random Forests algorithm or other algorithms, which is not limited herein.

[0078] In an alternative embodiment, the feature importance analysis method employs the random forest algorithm. In this case, step S12 above can be: using the random forest algorithm, determine the average value of the reduction in the average error after node classification of each initial feature in the decision tree; use the average value of the reduction in the average error of each initial feature as the importance of that initial feature, and screen the initial features of the preset feature fields whose importance is greater than the preset importance threshold as the target features. The above preset importance threshold can be set according to actual requirements. For example, the preset importance threshold can be 0, 0.1, or 0.2, etc.

[0079] In the embodiments of the present application, the average value of the reduction in the average error is used to measure the importance of the feature. For the average value of the reduction in the average error, it can be calculated through the following formula.

[0080]

[0081]

[0082] In the above formulas (1) and (2), represents the average value of the reduction in the average error of feature j, that is, the importance. M represents the number of decision trees designed according to the random forest algorithm, and L - 1 represents the number of non - leaf nodes; represents the reduction value of the average error after splitting of node t, and 1(V t＝ j) represents that when the feature of node t is j, the value is 1, and when the feature of node t is other features, the value is 0; T i represents the i - th decision tree in the random forest, represents the sum value of the reduction in the average error of feature j in decision tree T.

[0083] After calculating the importance of all initial features, the electronic device can delete the initial features of the preset feature fields whose importance is less than or equal to the preset importance threshold, and retain the initial features of the preset feature fields whose importance is greater than the preset importance threshold. The retained features are the target features.

[0084] Step S13, input the target features corresponding to each user data into the preset high - probability arrears recovery model respectively, and obtain the predicted probability of arrears recovery for each user.

[0085] In the embodiments of the present application, the electronic device has pre - set a preset high - probability arrears recovery model. This preset high - probability arrears recovery model is a HAR (High probability of arrearage recovery) model proposed based on LightGBM (Light Gradient Boosting Machine, weak gradient booster), and can also be called a preset HAR model.

[0086] LightGBM is a classification and prediction framework, and its core is the GBDT (Gradient Boosting Decision Tree) algorithm. The GBDT algorithm is an iterative decision tree algorithm that uses the boosting idea. It takes a series of CART (Classification And Regression Tree) models as weak learners and iteratively optimizes any loss function to generate a strong learner.

[0087] Suppose x represents the features, y represents the label values, the number of samples is m, and the input training samples are D = {(x1, y1), (x2, y2), …, (x m , y m )}, the maximum number of iterations is N, and the loss function is L. Then the algorithm flow of GBDT is as follows:

[0088] 1) Initialize the learner:

[0089]

[0090] where f0(x) represents the output value of the learner, and c represents the mean of all label values.

[0091] When iterating to the n-th round, where n ∈ (1, 2, …, N), we have:

[0092] 2) For sample i ∈ (1, 2, …, m), the negative gradient r ni in the n-th round of iteration can be expressed as:

[0093]

[0094] where f n (x) represents the output value of the learner in the n-th round.

[0095] 3) For sample i ∈ (1, 2, …, m), fit (x i , r ni ) to obtain the decision tree h n (x i ; w) in the n-th round of iteration:

[0096]

[0097] where w represents the decision tree parameters, and w* represents the decision tree parameters that minimize the value of the right - hand formula.

[0098] 4) Define as the generated decision tree h n (x i; the leaf node region corresponding to w), where J is the number of leaf nodes.

[0099] 5) Define each leaf node region in as R nj , and the best fitting value c of its residual can be obtained nj as:

[0100]

[0101] 6) Use the best fitting value c of the residual nj to update the strong learner, and we can get:

[0102]

[0103] where I represents the importance of feature x.

[0104] Combining the above, the strong learner is:

[0105]

[0106] Based on the above preset HAR model, for each user, the electronic device can directly input the target feature corresponding to the user data into the preset high-probability overdue recovery model. The preset high-probability overdue recovery model processes the target feature and outputs the predicted probability of overdue recovery for this user.

[0107] In the embodiments of the present application, the predicted probability output by the preset high-probability overdue recovery model can be expressed as a percentage, such as 70%, 80%, etc. In this case, the preset probability threshold can be 70% or 80%, etc.

[0108] Step S14, determine the user list of the target users according to the user information of the target users whose predicted probability is greater than the preset probability threshold.

[0109] Among them, the user information may include, but is not limited to, multi-dimensional data such as user identification, contact number, overdue amount, product specification identification, sales product specification identification, overdue time, user status, payment method, user type, and whether there is a contract.

[0110] The preset probability threshold can be set according to actual needs. For example, if the predicted probability output by the preset high-probability overdue recovery model is expressed as a percentage, such as 70%, 80%, etc., in this case, the preset probability threshold can be 70% or 80%, etc.; for another example, if the predicted probability output by the preset high-probability overdue recovery model is 0 and 1, in this case, the preset probability threshold can be 0, and at this time, the user corresponding to the predicted probability of 1 is the target user.

[0111] After obtaining the predicted probability of overdue payment recovery for each user, the electronic device screens users with a predicted probability greater than a preset probability threshold from multiple users as target users, and the number of target users can be one or more. The electronic device develops labels based on the user information of the target users and generates a user list of the target users, and the user list may include multiple user labels of the target users.

[0112] In an optional embodiment, the electronic device can analyze multi-dimensional data such as the usage behavior, consumption attributes, product structure, and credit history of overdue users to formulate an overdue payment collection plan and design label specifications. In this case, the above step S14 can be: extracting multi-dimensional data of the target users that match the preset label specifications from the user information of the target users with a predicted probability greater than the preset probability threshold to obtain the user list of the target users.

[0113] Step S15, sending an overdue payment reminder message to the target users according to the user list of the target users.

[0114] After determining the user list of the target users, the electronic device can send an overdue payment reminder message to the target users to collect overdue payments from them. Among them, the form of the overdue payment reminder message can be a telephone voice reminder or a text message reminder, etc., and the embodiments of the present application do not limit this.

[0115] In an optional embodiment, the above step S15 can be: sending the user list of the target users to the contact points that match the types of the target users. When the contact points determine that the target users are in an overdue state, they send overdue payment reminder messages to the target users according to the user list of the target users; when they determine that the target users are not in an overdue state, they no longer give overdue payment reminders to the target users, reducing the interference to users.

[0116] In the embodiments of the present application, the electronic device can send the user list of the target users to the business operation and service coordination center, and the business operation and service coordination center distributes the dispatching requirements and user label data to the contact points that match the types of the target users according to the user list. The user types can be divided into general public user labels and government and enterprise industry user labels. The contact points that match the general public user labels are intelligent call platforms, and the contact points that match the government and enterprise industry user labels are marketing systems. After receiving the user list, the contact points judge and screen the users who are still overdue, and conduct intelligent outbound calls or dispatch orders to front-line employees for collection of the screened users who are still overdue.

[0117] In the embodiments of the present application, corresponding collection is carried out on the target users according to the actual situation, which improves the collection success rate and strengthens the user perception.

[0118] The technical solutions provided by the embodiments of the present application can achieve the following beneficial effects:

[0119] 1) The embodiments of the present application introduce a machine learning algorithm to identify the characteristics of delinquent users. By constructing a HAR model (such as the above-mentioned LightGBM-HAR model), it predicts the users with a high probability of recovery among delinquent users, and then accurately urges these users with a high probability of recovery, thereby improving the success rate of delinquent payment collection. As Figure 2 shown in the schematic diagram of the accuracy of user payment collection, Figure 3 shown in the schematic diagram of the delinquent payment recovery situation, Figure 2 and Figure 3 in, the method provided by the embodiments of the present application was not applied in April and May, and the method provided by the embodiments of the present application was applied in June, July, August, and September. It can be seen from Figure 2 that the success rate of urging delinquent payment recovery users in June, July, August, and September shows an upward trend compared with the previous carpet-like collection method (taking April and May as examples).

[0120] 2) Save labor, material, and time costs, and reduce harassment to users. The embodiments of the present application have clear target users for collection, and can concentrate limited time, labor, and material resources on users with a relatively high probability of delinquent payment recovery, and provide targeted services to them, reducing harassment to users with a relatively low probability of recovery.

[0121] 3) The embodiments of the present application provide a certain income for the operator by reducing the economic losses caused by non-inclusion of delinquent payments, and have a certain positive effect on completing the assessment of relevant indicators for non-inclusion of delinquent payments. As Figure 3 shown, the number of delinquent payment recovery users is gradually increasing in June, July, August, and September, and the amount of delinquent payment recovery also shows an upward trend, which has a positive impact on increasing the operator's income.

[0122] In an embodiment of the present application, the embodiments of the present application also provide a delinquent payment reminder method. As Figure 4 shown, this method may include steps S41-S47. Steps S44-S47 are the same as the above steps S12-S15 and will not be described here again. Steps S41-S43 are an implementable manner of step S11.

[0123] Step S41, extract the original features of multiple preset feature fields corresponding to each user's data within a preset time period.

[0124] When performing delinquent payment reminder, for each user, the electronic device can extract the features of multiple preset feature fields from the user's data within a preset time period as the original features.

[0125] Step S42, delete the abnormal data in the original features of the extracted multiple preset feature fields, and complete the missing features in the original features of the extracted multiple preset feature fields to obtain the processed features of the multiple preset feature fields.

[0126] After extracting the original features of multiple preset feature fields corresponding to each user data, the electronic device performs missing data processing and abnormal data processing on the extracted original features. Specifically, for abnormal data, the replacement method and the deletion method can be adopted to delete the abnormal data in the original features of the multiple preset feature fields extracted; for missing data, according to business understanding and technical methods, the missing data can be filled, that is, the missing features in the original features of the multiple preset feature fields extracted are complemented. The original features after missing data processing and abnormal data processing are the processed features of the multiple preset feature fields.

[0127] Step S43: Convert the processed features of the multiple preset feature fields into numerical data to obtain the initial features of the multiple preset feature fields corresponding to each user data.

[0128] After obtaining the processed features of the multiple preset feature fields, the electronic device can perform a data type conversion operation, that is, convert the processed features of the multiple preset feature fields into numerical data, and the converted features corresponding to each user data are the initial features of the multiple preset feature fields corresponding to the user data.

[0129] In the embodiments of the present application, the execution order of steps S42 and S43 is not limited.

[0130] In the technical solution provided by the embodiments of the present application, the electronic device performs missing data processing, abnormal data processing, and data type conversion on the extracted original features, so that complete and easy-to-process numerical data can be used in subsequent processing, improving the accuracy and efficiency of the overdue payment recovery probability and the accuracy and efficiency of the overdue payment reminder.

[0131] In an embodiment of the present application, the embodiments of the present application also provide an overdue payment reminder method. As Figure 5 shown, the method may include steps S51 - S57. Steps S51, S52, S56, and S57 are the same as steps S11, S12, S14, and S15 above and will not be described herein again. Steps S54 - S55 are an implementable manner of step S13.

[0132] Step S53: Divide the target features of the multiple preset feature fields into a training set and a test set.

[0133] In the embodiments of the present application, after obtaining the target data set composed of target features, the electronic device divides the target data set to obtain a training set and a test set. Among them, the ratio of dividing the target data set into a training set and a test set can be set according to actual needs. For example, the target data set can be divided into a training set and a test set according to a ratio of 7:3 or 8:2, etc., and this is not limited.

[0134] In step S54, the preset high-probability overdue payment recovery model is trained using the target features corresponding to each user data included in the training set.

[0135] The electronic device can train the preset high-probability overdue payment recovery model in combination with the above formulas (3)-(8). After the model training and prediction are completed, the root mean square error is used to evaluate the prediction result, and the root mean square error is denoted as δ RMSE . The root mean square error is the label value y i and the predicted value The square root of the ratio of the square of the deviation to the number of samples m. It can indicate the accuracy of the model prediction result, and its calculation formula is:

[0136]

[0137] After the model training is completed, the output result of the evaluation model (i.e., the preset high-probability overdue payment recovery model) is evaluated. The electronic device can also optimize parameters such as the maximum tree depth, the number of layers of the generated decision tree, the number of leaf nodes, and the learning rate to further improve the prediction accuracy of the model.

[0138] In step S55, the target features corresponding to each user data included in the test set are respectively input into the trained preset high-probability overdue payment recovery model to obtain the predicted probability of overdue payment recovery for each user.

[0139] After the training is completed, the electronic device respectively inputs the target features corresponding to each user data included in the test set into the trained preset high-probability overdue payment recovery model to obtain the predicted probability of overdue payment recovery for each user.

[0140] The following combines Figure 6 The overdue payment recovery probability prediction process shown is taken as an example with the preset high-probability overdue payment recovery model being the LightGBM-HAR model and a 3-month data set as an example for illustration. Figure 6 In, the electronic device splits the 3-month data set into a training set and a test set, performs feature engineering processing on the training set and the test set, screens out the field data relevant to the target, realizes data dimensionality reduction by constructing and deleting features, and uses the dimensionality-reduced training set to construct, train, and optimize the LightGBM-HAR model. Then it judges whether the prediction result of the LightGBM-HAR model meets the prediction conditions. If not, it re-performs feature engineering processing on the training set; if so, it ends the construction, training, and optimization of the LightGBM-HAR model; inputs the test set into the LightGBM-HAR model to predict the users with high-probability overdue payment recovery.

[0141] In the technical solution provided by the embodiment of the present application, the electronic device divides the target features into a training set and a test set. Before predicting the overdue recovery probability of the user each time, the preset high-probability overdue recovery model is trained, so that the preset high-probability overdue recovery model is more in line with the prediction of the current overdue recovery probability, improving the accuracy of the overdue recovery probability prediction, and further improving the success rate of overdue recovery.

[0142] Corresponding to the above-mentioned overdue reminder method, the embodiment of the present application further provides an overdue reminder device, as Figure 7 shown, the device includes:

[0143] An acquisition unit 71, configured to acquire initial features of multiple preset feature fields corresponding to each user data within a preset time period;

[0144] A screening unit 72, configured to screen out initial features of preset feature fields with importance greater than a preset importance threshold from the initial features of multiple preset feature fields as target features;

[0145] A prediction unit 73, configured to input the target features corresponding to each user data into a preset high-probability overdue recovery model respectively to obtain the predicted probability of overdue recovery for each user. The preset high-probability overdue recovery model is a model based on the Light Gradient Boosting Machine (LightGBM);

[0146] A determination unit 74, configured to determine a user list of target users according to the user information of target users whose prediction probability is greater than a preset probability threshold;

[0147] A sending unit 75, configured to send an overdue reminder message to the target users according to the user list of the target users.

[0148] In an optional embodiment, the acquisition unit 71 may specifically be configured to:

[0149] Extract the original features of multiple preset feature fields corresponding to each user data within a preset time period;

[0150] Delete abnormal data in the original features of the extracted multiple preset feature fields, and complete missing features in the original features of the extracted multiple preset feature fields to obtain processed features of the multiple preset feature fields;

[0151] Convert the processed features of the multiple preset feature fields into numerical data to obtain the initial features of the multiple preset feature fields corresponding to each user data.

[0152] In an optional embodiment, the screening unit 72 may specifically be configured to:

[0153] Use the random forest algorithm to determine the average value of the reduction in the average error after node classification of each initial feature in the decision tree;

[0154] The average value of the reduction in the average error of each initial feature is used as the importance of the initial feature, and the initial features of the preset feature fields with importance greater than the preset importance threshold are screened as the target features.

[0155] In an optional embodiment, the user information includes multi-dimensional data;

[0156] The determining unit 74 can specifically be used to extract the multi-dimensional data of the target user that matches the preset label caliber from the user information of the target user with a prediction probability greater than the preset probability threshold, so as to obtain the user list of the target user.

[0157] In an optional embodiment, the sending unit 75 can specifically be used for:

[0158] Send the user list of the target user to the contact point that matches the type of the target user, so that when the contact point determines that the target user is in an overdue state, an overdue reminder message is sent to the target user according to the user list of the target user.

[0159] In an optional embodiment, the above-mentioned overdue reminder device may further include:

[0160] A dividing unit, configured to divide the target features of multiple preset feature fields into a training set and a test set;

[0161] In this case, the prediction unit 73 can specifically be used for:

[0162] Using the target features corresponding to each user data included in the training set to train the preset high-probability overdue recovery model;

[0163] Respectively input the target features corresponding to each user data included in the test set into the trained preset high-probability overdue recovery model to obtain the prediction probability of overdue recovery for each user.

[0164] In the technical solution provided by the embodiments of the present application, by using the features of the unique feature fields of overdue users and combining with the preset high-probability overdue recovery model, users with a relatively high probability of overdue recovery are determined, and then corresponding reminders are made to the overdue users among these users with a relatively high probability of overdue recovery, which improves the success rate of overdue collection. This way of overdue reminder can achieve precise collection, without having to repeatedly remind all overdue users, reducing the interference to the user side, and concentrating the limited human and material resources on users with a relatively high probability of overdue recovery, reducing resource consumption.

[0165] Corresponding to the above-mentioned overdue reminder method, the embodiments of the present application also provide an electronic device, such as Figure 8As shown in the figure, it includes a processor 81, a communication interface 82, a memory 83, and a communication bus 84. Among them, the processor 81, the communication interface 82, and the memory 83 complete mutual communication through the communication bus 84.

[0166] The memory 83 is used to store computer programs.

[0167] When the processor 81 is used to execute the program stored on the memory 83, the steps of the overdue payment reminder method described above are implemented.

[0168] The communication bus mentioned in the electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0169] The communication interface is used for communication between the above-mentioned electronic device and other devices.

[0170] The memory may include a Random Access Memory (RAM), or may also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0171] The processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0172] In another embodiment provided by this application, a computer-readable storage medium is also provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any of the above-mentioned overdue payment reminder methods are implemented.

[0173] In another embodiment provided by the present application, a computer program product including instructions is further provided. When it runs on a computer, it causes the computer to execute the steps of any of the overdue fee reminder methods in the above embodiments.

[0174] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from a website, a computer, a server, or a data center to another website, computer, server, or data center in a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0175] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article, or device including the element.

[0176] Each embodiment in this specification is described in a related manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device, electronic device, storage medium, and computer program product, since they are basically similar to the method embodiments, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiments.

[0177] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are all included in the protection scope of the present application.

Claims

1. A method for overdue payment reminder, characterized in that, The method includes: Obtaining the initial features of multiple preset feature fields corresponding to each user data within a preset time period; Filtering out the initial features of the preset feature fields with importance greater than a preset importance threshold from the initial features of the multiple preset feature fields as target features; Inputting the target features corresponding to each user data into a preset high-probability overdue payment recovery model respectively to obtain the predicted probability of overdue payment recovery for each user, where the preset high-probability overdue payment recovery model is a model based on the Light Gradient Boosting Machine (LightGBM); Determining the user list of the target users according to the user information of the target users with predicted probabilities greater than a preset probability threshold; Sending overdue payment reminder information to the target users according to the user list of the target users.

2. The method according to claim 1, wherein The step of obtaining the initial features of multiple preset feature fields corresponding to each user data within a preset time period includes: Extracting the original features of multiple preset feature fields corresponding to each user data within a preset time period; Deleting the abnormal data in the original features of the multiple preset feature fields extracted and complementing the missing features in the original features of the multiple preset feature fields extracted to obtain the processed features of the multiple preset feature fields; Converting the processed features of the multiple preset feature fields into numerical data to obtain the initial features of multiple preset feature fields corresponding to each user data.

3. The method according to claim 1, wherein The step of filtering out the initial features of the preset feature fields with importance greater than a preset importance threshold from the initial features of the multiple preset feature fields as target features includes: Using the random forest algorithm to determine the average value of the reduction in the average error after node classification in the decision tree for each initial feature; Taking the average value of the reduction in the average error after node classification in the decision tree for each initial feature as the importance of the initial feature, and filtering out the initial features of the preset feature fields with importance greater than a preset importance threshold as target features.

4. The method according to claim 1, characterized in that, The user information includes multi-dimensional data; The step of determining the user list of the target users according to the user information of the target users with predicted probabilities greater than a preset probability threshold includes: Extracting the multi-dimensional data of the target users that match the preset label caliber from the user information of the target users with predicted probabilities greater than a preset probability threshold to obtain the user list of the target users.

5. The method according to claim 1, wherein The step of sending overdue payment reminder information to the target users according to the user list of the target users includes: Sending the user list of the target users to the contacts matching the type of the target users, so that when the contacts determine that the target users are in an overdue payment state, overdue payment reminder information is sent to the target users according to the user list of the target users.

6. The method according to any one of claims 1 to 5, characterized in that The method further includes: Dividing the target features of the multiple preset feature fields into a training set and a test set; The step of inputting the target features corresponding to each user data into a preset high-probability overdue payment recovery model respectively to obtain the predicted probability of overdue payment recovery for each user includes: Training the preset high-probability overdue payment recovery model by using the target features corresponding to each user data included in the training set; Input the target features corresponding to each user data included in the test set into the pre-trained preset high-probability overdue recovery model respectively to obtain the predicted probability of overdue recovery for each user.

7. An overdue payment reminder device, characterized in that, The device includes: An acquisition unit, configured to acquire the initial features of multiple preset feature fields corresponding to each user data within a preset time period; A screening unit, configured to screen out the initial features of the preset feature fields with importance greater than a preset importance threshold from the initial features of the multiple preset feature fields as target features; A prediction unit, configured to input the target features corresponding to each user data into a preset high-probability overdue recovery model respectively to obtain the predicted probability of overdue recovery for each user, where the preset high-probability overdue recovery model is a model based on the Light Gradient Boosting Machine (LightGBM); A determination unit, configured to determine the user list of the target users according to the user information of the target users whose prediction probability is greater than a preset probability threshold; A sending unit, configured to send an overdue reminder message to the target users according to the user list of the target users.

8. The device according to claim 7, characterized in that, The acquisition unit is specifically configured to: Extract the original features of multiple preset feature fields corresponding to each user data within a preset time period; Delete the abnormal data in the original features of the extracted multiple preset feature fields and complete the missing features in the original features of the extracted multiple preset feature fields to obtain the processed features of the multiple preset feature fields; Convert the processed features of the multiple preset feature fields into numerical data to obtain the initial features of multiple preset feature fields corresponding to each user data.

9. The device according to claim 7, characterized in that, The screening unit is specifically configured to: Use the random forest algorithm to determine the average value of the reduction in the average error after node classification of each initial feature in the decision tree; Take the average value of the reduction in the average error after node classification of each initial feature in the decision tree as the importance of the initial feature, and screen out the initial features of the preset feature fields with importance greater than a preset importance threshold as target features.

10. The device according to claim 7, characterized in that, The user information includes multi-dimensional data; The determination unit is specifically configured to extract the multi-dimensional data of the target users that match the preset label caliber from the user information of the target users whose prediction probability is greater than a preset probability threshold to obtain the user list of the target users.

11. The device according to claim 7, characterized in that, The sending unit is specifically configured to: Send the user list of the target users to the contact points that match the type of the target users, so that when the contact points determine that the target users are in an overdue state, they send overdue reminder messages to the target users according to the user list of the target users.

12. The device according to any one of claims 7 to 11, characterized in that, The device further includes: A division unit, configured to divide the target features of the multiple preset feature fields into a training set and a test set; The prediction unit is specifically configured to: Use the target features corresponding to each user data included in the training set to train a preset high-probability overdue recovery model; Input the target features corresponding to each user data included in the test set into the pre-trained preset high-probability overdue recovery model respectively to obtain the predicted probability of overdue recovery for each user.

13. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; The memory is used to store computer programs; The processor is used to implement the method steps described in any one of claims 1-6 when executing the programs stored on the memory.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method steps described in any one of claims 1-6.

Images