Model iteration method and device, equipment and storage medium

Abstract

The invention discloses a model iteration method and device, equipment and a storage medium, and the method comprises the steps: scoring a successful credit granting marking sample and a credit granting refusal sample through a preset scoring model, and carrying out the scoring interval division of the successful credit granting marking sample and the credit granting refusal sample based on a scoring result; according to the overdue distribution of the target marked credit granting success sample in each scoring interval, setting a state label for a credit granting rejection sample, and obtaining a marked credit granting rejection sample; and performing iterative training on a preset scoring model through the marked credit granting success sample and the marked credit granting rejection sample until the model converges. Compared with an existing method for increasing samples through a new round of data collection, the method has the advantages that the samples can be expanded through a model expansion method when the number of the samples is small, that is, the iterative samples of the sample expansion model can be directly rejected based on mark credit extension; therefore, while the features of the model iteration samples are increased and the model fitting effect is improved, the iteration time is shortened, and the model iteration efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a model iteration method, device, equipment and storage medium. Background Art

[0002] In the early stage of loan business, user samples are usually small. In order to prevent the small number of samples from causing deviation in the model iteration effect, the rejection inference method can be used to add good and bad user identification to those rejected applicants, thereby increasing the model training samples and improving the model fitting performance. At present, the mature rejection inference methods in the industry include: expansion method, extrapolation method, scattered packaging method, re-weighting method, and re-classification method.

[0003] However, although the expansion method and the extrapolation method make conditional assumptions to a certain extent, such assumptions cannot be actually verified, so it takes a long time to iterate the model; the individual characteristics of the reweighting method are the same as the original sample, and the accuracy is not much improved; the reclassification method requires the use of data from a new cycle, and the data collection process requires a long waiting time; the scattered packaging method relies more on the staff's understanding of the data and business experience to set the threshold for rejecting bad users, which is not only time-consuming and labor-intensive, but also the prediction results are easily affected by human factors, with large deviations and low accuracy. Therefore, there is an urgent need for a model iteration method that is short in time and high in accuracy.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present invention is to provide a model iteration method, device, equipment and storage medium, aiming to solve the technical problems of the existing model iteration method being time-consuming and low in accuracy.

[0006] To achieve the above object, the present invention provides a model iteration method, which comprises:

[0007] Scoring the marked credit success samples and the marked credit rejection samples by using a preset scoring model, and dividing the marked credit success samples and the marked credit rejection samples into scoring intervals based on the scoring results;

[0008] According to the overdue distribution of the target marked credit success samples in each scoring interval, a status label is set for the credit rejection samples to obtain the marked credit rejection samples;

[0009] The preset scoring model is iteratively trained through the marked credit success samples and the marked credit rejection samples until the model converges.

[0010] Optionally, the step of setting a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval to obtain the marked credit rejection sample includes:

[0011] Determine the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target marked credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval;

[0012] A status label is randomly set for the target credit rejection sample based on the predicted overdue rate to obtain a labeled credit rejection sample.

[0013] Optionally, the step of determining the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target marked credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval includes:

[0014] Determining the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval;

[0015] The predicted overdue rate of the target credit rejection sample is determined based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample.

[0016] Optionally, the step of determining the overdue mapping ratio between the target credit success sample and the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval includes:

[0017] According to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval, respectively determine the first network access duration index of the target marked credit success sample and the second network access duration index of the target credit rejection sample;

[0018] Based on the first network access duration indicator and the second network access duration indicator, an overdue mapping ratio between the target marked credit success samples and the target credit rejection samples is determined.

[0019] Optionally, the step of determining the overdue mapping ratio between the target marked credit success samples and the target credit rejection samples based on the first network access time indicator and the second network access time indicator includes:

[0020] Determine a preferred credit granting success sample based on the first network access information and the first network access duration indicator;

[0021] Determine a preferred credit rejection sample based on the second network access information and the second network access duration indicator;

[0022] Determine the preferred credit success ratio and the preferred credit rejection ratio in each scoring interval through the preferred credit success samples and the preferred credit rejection samples;

[0023] According to the preferred credit success ratio and the preferred credit rejection ratio, the overdue mapping ratio between the target marked credit success samples and the target credit rejection samples is determined.

[0024] Optionally, before scoring the marked credit success samples and credit rejection samples by using a preset scoring model, the method further includes:

[0025] Classifying the initial samples according to their credit information to obtain credit success samples and credit rejection samples;

[0026] Marking the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain a marked successful credit granting sample;

[0027] Logistic regression training is performed on the marked credit success samples to obtain a preset scoring model.

[0028] Optionally, the repayment information includes: a repayment schedule and repayment behavior data; the step of marking the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain the marked successful credit granting sample includes:

[0029] Obtaining the repayment schedule and repayment behavior data corresponding to the successful credit granting sample;

[0030] Determine the user actual repayment table corresponding to the successful credit granting sample according to the repayment behavior data of the successful credit granting sample;

[0031] A preset risk analysis is performed on the repayment plan table and the user actual repayment table, and the credit success sample is marked based on the risk analysis result to obtain a marked credit success sample.

[0032] In addition, to achieve the above-mentioned purpose, the present invention further proposes a model iteration device, the model iteration device comprising:

[0033] A scoring module, used to score the marked credit success samples and the credit rejection samples by using a preset scoring model, and divide the marked credit success samples and the credit rejection samples into scoring intervals based on the scoring results;

[0034] A marking module, used for setting a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval, and obtaining the marked credit rejection sample;

[0035] The model iteration module is used to iteratively train the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges.

[0036] In addition, to achieve the above-mentioned purpose, the present invention also proposes a model iteration device, which includes: a memory, a processor, and a model iteration program stored in the memory and executable on the processor, and the model iteration program is configured to implement the steps of the model iteration method described above.

[0037] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a model iteration program is stored, and when the model iteration program is executed by a processor, the steps of the model iteration method described above are implemented.

[0038] The present invention discloses a model iteration method, device, equipment and storage medium. The method includes: classifying the initial samples according to the credit information of the initial samples to obtain the credit success samples and the credit rejection samples; obtaining the repayment plan table and repayment behavior data corresponding to the credit success samples; determining the user actual repayment table corresponding to the credit success samples according to the repayment behavior data of the credit success samples; performing preset risk analysis on the repayment plan table and the user actual repayment table, and marking the credit success samples based on the risk analysis results to obtain the marked credit success samples. Performing logistic regression training on the marked credit success samples to obtain a preset scoring model. Scoring the marked credit success samples and the credit rejection samples through the preset scoring model, and dividing the marked credit success samples and the credit rejection samples into scoring intervals based on the scoring results; setting a status label for the credit rejection samples according to the overdue distribution of the target marked credit success samples in each scoring interval to obtain the marked credit rejection samples; iteratively training the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges. Therefore, the present invention does not require unverifiable conditional assumptions, but directly adds feature labels to credit rejection samples based on the overdue distribution of marked credit success samples. At the same time, compared with the existing method of increasing samples through a new round of data collection, the present invention can expand samples through a model expansion method when the sample size is small, that is, the present application can directly expand the iterative samples of the model based on the marked credit rejection samples, thereby increasing the characteristics of the model iteration samples to improve the model fitting effect while reducing the iteration time and improving the model iteration efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1It is a structural diagram of a model iteration device of a hardware operating environment involved in an embodiment of the present invention;

[0040] Figure 2 It is a first flow chart of the first embodiment of the model iteration method of the present invention;

[0041] Figure 3 This is a schematic diagram of dividing the equidistant scoring intervals according to the first embodiment of the model iteration method of the present invention;

[0042] Figure 4 This is a schematic diagram of non-equidistant scoring interval division according to the first embodiment of the model iteration method of the present invention;

[0043] Figure 5 It is a second flow chart of the first embodiment of the model iteration method of the present invention;

[0044] Figure 6 It is a flow chart of the second embodiment of the model iteration method of the present invention;

[0045] Figure 7 It is a first flow chart of the third embodiment of the model iteration method of the present invention;

[0046] Figure 8 It is a second flow chart of the third embodiment of the model iteration method of the present invention;

[0047] Fig. 9 It is a structural block diagram of the first embodiment of the model iteration device of the present invention.

[0048] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0050] Reference Figure 1 , Figure 1 A schematic diagram of the structure of a model iteration device for a hardware operating environment according to an embodiment of the present invention.

[0051] like Figure 1As shown, the model iteration device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0052] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the model iteration device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0053] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a data storage module, a network communication module, a user interface module and a model iteration program.

[0054] exist Figure 1 In the model iteration device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the model iteration device of the present invention can be set in the model iteration device, and the model iteration device calls the model iteration program stored in the memory 1005 through the processor 1001, and executes the model iteration method provided by the embodiment of the present invention.

[0055] The embodiment of the present invention provides a model iteration method, referring to Figure 2 , Figure 2 It is a first flow chart of the first embodiment of the model iteration method of the present invention.

[0056] like Figure 2 As shown, in this embodiment, the model iteration method includes the following steps:

[0057] Step S10: scoring the marked credit success samples and the credit rejection samples by using a preset scoring model, and dividing the marked credit success samples and the credit rejection samples into scoring intervals based on the scoring results;

[0058] It should be noted that the execution subject of the method of this embodiment can be a computing service device with data processing, network communication, program running and model training functions, such as a tablet computer, a personal computer, etc., and can also be other electronic devices that can achieve the same or similar functions. The model iteration method provided in this embodiment and the following embodiments is specifically described here with the above-mentioned model iteration device.

[0059] It is understandable that the preset scoring model can be a model that has been pre-trained and converged based on the above-mentioned marked credit success samples. Based on the preset scoring model, the marked credit success samples and credit rejection samples can be scored, and the quality of the rejected samples can be judged based on the scoring results.

[0060] It should be understood that the scoring interval to be divided can be the union of the scoring results of the marked credit success samples and the credit rejection samples. For example, if the scoring result of the marked credit success samples is [340, 500] and the scoring result of the credit rejection samples is [450, 660], then the final scoring interval to be divided is [340, 660].

[0061] It is easy to understand that the purpose of dividing the scoring intervals for the marked credit success samples and credit rejection samples based on the scoring results is to improve the accuracy of setting status labels for the credit rejection samples based on the overdue distribution of the target marked credit success samples. The above scoring interval division can be based on the scoring results of the marked credit success samples and credit rejection samples. The specific interval division method and the specific number of intervals can be set according to the actual situation, and this implementation does not impose any restrictions on this. For ease of understanding, Figure 3 and Figure 4 Take this as an example to illustrate. Figure 3 This is a schematic diagram of dividing the equidistant scoring intervals according to the first embodiment of the model iteration method of the present invention. Figure 4 This is a schematic diagram of the non-equidistant scoring interval division of the first embodiment of the model iteration method of the present invention. Figure 3As shown in FIG. 1 , if the union of the scoring results of the marked credit success samples and the credit rejection samples is [340, 660], and when equal distance division is required and the number of interval divisions is 10, the marked credit success samples and the credit rejection samples in this embodiment can be divided into the following according to the scoring intervals: (-inf, 340], [340, 380], [380, 420], [420, 460], [460, 500], [500, 540], [540, 580], [580, 620], [620, 660] and [660, inf); as shown in FIG. Figure 4 As shown, if the union of the scoring results of the marked credit success samples and the credit rejection samples is [320, 500], since the data between [380, 460] is more representative, when non-equidistant division is required and the number of interval divisions is 8, the marked credit success samples and credit rejection samples in this embodiment can be divided into the following according to the scoring intervals: (-inf, 320], [320, 380], [380, 400], [400, 420], [420, 440], [440, 460], [460, 500] and [500, inf).

[0062] Therefore, further, Figure 5 As shown, Figure 5 This is a second flow chart of the first embodiment of the model iteration method of the present invention. In this embodiment, before step S10, it also includes:

[0063] Step S00: classifying the initial samples according to their credit information to obtain credit success samples and credit rejection samples;

[0064] It should be noted that the above-mentioned initial samples can be samples of all users of the loan business to be analyzed. Then, this embodiment can classify the initial samples according to the credit information of the initial samples, which may include successful credit applications and failed credit applications, to obtain successful credit samples (corresponding to users with successful credit applications) and rejected credit samples (corresponding to users with failed credit applications). It can be understood that the successful credit samples and rejected credit samples are not single samples, but sample sets.

[0065] Step S01: marking the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain a marked successful credit granting sample;

[0066] It should be understood that, before building a preset scoring model based on the credit success samples, in order to facilitate model training, the present embodiment may first mark the credit success samples, and the above repayment information may include: a repayment schedule and repayment behavior data. Therefore, further, in the present embodiment, step S01 includes:

[0067] Step S010: Obtain the repayment schedule and repayment behavior data corresponding to the credit successful sample;

[0068] Step S011: determining the user actual repayment table corresponding to the successful credit granting sample according to the repayment behavior data of the successful credit granting sample;

[0069] It can be understood that the above-mentioned repayment plan table can be generated based on the repayment plan of users who have successfully applied for credit and successfully obtained loans (i.e., the above-mentioned successful credit samples), and the above-mentioned repayment behavior data can be the actual post-loan repayment behavior data of the successful credit samples. Then, the model iteration device can generate a user actual repayment table corresponding to the successful credit sample for recording the post-loan repayment behavior data based on the post-loan repayment behavior data of the successful credit sample.

[0070] Step S012: Perform a preset risk analysis on the repayment plan table and the user actual repayment table, and mark the credit success sample based on the risk analysis result to obtain a marked credit success sample.

[0071] It should be understood that based on the post-loan repayment behavior data in the user actual repayment table of the above-mentioned successful credit samples and the generated repayment plan table, the model iteration device in this embodiment can perform aging analysis (Vintage Analysis) and roll rate analysis (Roll Rate Analysis), that is, the above-mentioned preset risk analysis, and based on the definition of "good and bad users", determine good users and bad users from the successful credit application samples, add good labels to good users, add bad labels to bad users, and use all the successful credit samples with corresponding labels as the above-mentioned marked successful credit samples.

[0072] It is understandable that in actual applications, the process of performing vintage and rolling rate analysis based on the post-loan repayment behavior data of the repayment schedule of the successful credit sample, and determining whether each sample in the successful credit sample is a good user or a bad user based on the analysis results can be: the model iteration device can pre-define that users who have been overdue for more than 30 days in 4 performance periods (i.e. MOB4) are bad users, and users who are not overdue in MOB4 are good users. Then, after performing aging analysis and rolling rate analysis based on the post-loan repayment behavior data in the user's actual repayment table and the generated repayment schedule, if the analysis results show that a successful credit sample hits the definition of a bad user, the successful credit sample is determined as a bad sample and a bad label is added to it; if the analysis results show that a successful credit sample hits the definition of a good user, the successful credit sample is determined as a good sample and a good label is added to it.

[0073] Step S02: Performing logistic regression training on the marked credit success samples to obtain a preset scoring model.

[0074] Step S20: setting a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval to obtain the marked credit rejection sample;

[0075] It should be noted that, compared with the unverifiable conditional assumptions of the expansion method and the extrapolation method, and the reclassification method that requires re-sample collection, this embodiment does not require conditional assumptions, nor does it require a new round of data collection. Instead, it can set status labels for rejected samples based on the overdue distribution of the determined target label credit success samples. Therefore, the subsequent adjustment of the model takes less time than the prior art, thereby improving the model iteration efficiency. It is easy to understand that the status label added to the credit rejection sample at this time can be the same as the label added to the credit success sample above, that is, the above status label can include a good label and a bad label to ensure that the analysis of the credit success sample and the credit rejection sample is consistent as much as possible.

[0076] Step S30: iteratively training the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges.

[0077] It should be noted that, compared with the existing re-weighting method, this embodiment can perform sample enhancement based on the characteristics of marked credit rejection samples, that is, merge the above-mentioned marked credit rejection samples and marked credit success samples into a large sample set, and then iteratively train the preset scoring model based on the sample set using the logistic regression method. Therefore, the fitting effect of the model after iteration is better, thereby improving the accuracy of the model.

[0078] The present embodiment discloses a model iteration method, which includes: classifying the initial samples according to the credit information of the initial samples to obtain credit success samples and credit rejection samples; obtaining the repayment plan table and repayment behavior data corresponding to the credit success samples; determining the user actual repayment table corresponding to the credit success samples according to the repayment behavior data of the credit success samples; performing preset risk analysis on the repayment plan table and the user actual repayment table, and marking the credit success samples based on the risk analysis results to obtain the marked credit success samples. Performing logistic regression training on the marked credit success samples to obtain a preset scoring model. Scoring the marked credit success samples and credit rejection samples through the preset scoring model, and dividing the marked credit success samples and credit rejection samples into scoring intervals based on the scoring results; setting a status label for the credit rejection samples according to the overdue distribution of the target marked credit success samples in each scoring interval to obtain the marked credit rejection samples; iteratively training the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges. Therefore, this embodiment does not require unverifiable conditional assumptions, but directly adds feature labels to credit rejection samples based on the overdue distribution of marked credit success samples. At the same time, compared with the existing method of increasing samples through a new round of data collection, the present invention can expand samples through a model expansion method when the sample size is small, that is, the present application can directly expand the iterative samples of the model based on the marked credit rejection samples, thereby increasing the characteristics of the model iteration samples to improve the model fitting effect while reducing the iteration time and improving the model iteration efficiency.

[0079] Reference Figure 6 , Figure 6 This is a flow chart of the second embodiment of the model iteration method of the present invention. Figure 2 The illustrated embodiment provides a second embodiment of the model iteration method of the present invention.

[0080] In this embodiment, step S20 specifically includes:

[0081] Step S21: Determine the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval;

[0082] Step S22: randomly setting a status label for the target credit rejection sample based on the predicted overdue rate to obtain a labeled credit rejection sample.

[0083] It should be noted that compared with the unverifiable conditional assumptions of the existing expansion method and extrapolation method, this embodiment can predict the overdue rate of credit rejection samples based on the overdue distribution of marked credit success samples, and then add corresponding status labels to the credit rejection samples based on the predicted overdue rate of the credit rejection samples.

[0084] It can be understood that in order to avoid dividing the credit rejection samples into good samples and bad samples in a one-size-fits-all manner according to a certain threshold, thereby causing a large difference in the overdue distribution of the credit rejection samples and the overdue distribution of the credit success samples, this embodiment can adopt a segmented correspondence method to improve the mapping accuracy between the target marked credit success samples and the target credit rejection samples.

[0085] It should be understood that the above network access information can be the communication, traffic, package, consumption and other data of the user corresponding to each sample, which can be obtained by querying the preset network access information database. Based on the above analysis, this embodiment can score the marked credit success samples and credit rejection samples respectively according to the preset scoring model, and sort the marked credit success samples and credit rejection samples from low to high according to the scores, so as to obtain different scoring intervals; then this embodiment can calculate the overdue rate of the credit success samples corresponding to each scoring interval according to the overdue behavior data of the user in the repayment behavior data of the credit success samples, and at the same time count the number of credit rejection samples in each scoring interval. It is easy to understand that the above target marked credit success samples and the above target credit rejection samples are marked credit success samples and credit rejection samples located in the same scoring interval. This embodiment can perform separate preset overdue rate calculations in different intervals.

[0086] It should be noted that the above-mentioned status labels may include: good labels and bad labels. After obtaining the predicted overdue rate in each scoring interval, this embodiment can randomly set status labels for the credit rejection samples in each interval so that the ratio between the credit rejection samples with bad labels in each scoring interval and the total target credit rejection samples and the corresponding predicted overdue rate are the same.

[0087] This embodiment determines the predicted overdue rate of the target credit rejection sample based on the overdue distribution and network access information of the target marked credit success sample in each scoring interval and the network access information of the target credit rejection sample in the same scoring interval; randomly sets a status label for the target credit rejection sample according to the predicted overdue rate to obtain the marked credit rejection sample. Therefore, this embodiment can set a status label for the target credit rejection sample in the same scoring interval based on the overdue distribution of the target marked credit success sample, avoiding the problem that the overdue distribution of the credit rejection sample and the business overdue distribution of the credit success sample are greatly different. Therefore, the predicted overdue rate of the credit rejection sample determined by this embodiment is highly accurate, and can also improve the fitting effect of the preset scoring model for subsequent model iteration based on the marked credit rejection sample.

[0088] Reference Figure 7 , Figure 7 This is a first flow chart of the third embodiment of the model iteration method of the present invention, based on the above Figure 2 or Figure 6The embodiment shown in the figure proposes a third embodiment of the model iteration method of the present invention. Figure 7 Based on Figure 1 The illustrated embodiments are presented as examples.

[0089] like Figure 7 As shown, in this embodiment, step S21 includes:

[0090] Step S211: determining the overdue mapping ratio between the target marked credit success samples and the target credit rejection samples according to the first network access information of the target marked credit success samples in each scoring interval and the second network access information of the target credit rejection samples in the same scoring interval;

[0091] Furthermore, if Figure 8 As shown, Figure 8 This is a second flow chart of the third embodiment of the model iteration method of the present invention. In this embodiment, step S211 includes:

[0092] Step S2110: According to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval, respectively determine the first network access duration index of the target marked credit success sample and the second network access duration index of the target credit rejection sample;

[0093] Step S2111: Based on the first network access duration indicator and the second network access duration indicator, determine the overdue mapping ratio between the target marked credit success samples and the target credit rejection samples.

[0094] Step S212: Determine the predicted overdue rate of the target credit rejection sample based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample.

[0095] It should be noted that, according to the first network access information of the target marked credit success sample and the second network access information of the target credit rejection sample, the network access duration corresponding to each sample in the target marked credit success sample and the target credit rejection sample can be obtained, and the network access duration can refer to the length of time of network access, such as a few days, weeks, months, years, etc. In actual applications, according to the mobile phone numbers corresponding to each user in the target marked credit success sample and the target credit rejection sample, by querying the above-mentioned preset network access information database, the network access duration corresponding to each user in the target marked credit success sample and the target credit rejection sample can be queried.

[0096] It should be understood that, in this embodiment, the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample can be determined based on the default rate ratio of the target credit success sample and the target credit rejection sample, that is, the calculation formula of the overdue rate of the target credit rejection sample can be:

[0097] Target credit rejection sample predicted overdue rate = target credit success sample overdue rate * (target credit success sample default rate / target credit rejection sample default rate);

[0098] In the above formula, the target credit success sample default rate / target credit rejection sample default rate can be the above-mentioned overdue mapping ratio. The default rate can be determined based on the default situation in the behavioral data corresponding to each sample. The default situation may include: the user fails to reach the minimum level of promised consumption or fails to pay the phone bill on time, etc.

[0099] It should be understood that, in addition to the above-mentioned methods, after analyzing the network access information such as communication, traffic, packages, and consumption of all mobile network users, this embodiment concludes that the variable "user network access time" has the highest information value (IV, Information value), which can be well used to distinguish between good and bad users (here it can be the quality of the user's credit integrity). In particular, the common feature of the credit successful users and credit rejected users with "network access time of more than 24 months" is that the default rate is about 1.17%, which is much lower than the average default rate of the sample. Therefore, this embodiment can infer that whether it is a credit successful sample or a credit rejected sample, the default rate distribution of users with a network access time of more than 24 months is similar. Therefore, this embodiment can set the variable "network access time of more than 24 months" as the "network access time index value", so as to calculate the proportional relationship between the default rates of the target credit successful sample and the target credit rejected sample based on the network access time index value, and then infer the overdue rate of the target credit rejected sample.

[0100] It can be understood that in actual applications, this embodiment can determine the preferred credit success sample based on the first network access information and the first network access time indicator; determine the preferred credit rejection sample based on the second network access information and the second network access time indicator; determine the preferred credit success ratio and the preferred credit rejection ratio in each scoring interval based on the preferred credit success sample and the preferred credit rejection sample; determine the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample based on the preferred credit success ratio and the preferred credit rejection ratio, the above-mentioned preferred credit success sample can be the user with a network access time of more than 24 months in the target credit success sample, and the preferred credit success ratio can be the ratio of users with a network access time of more than 24 months in the target credit success sample; the preferred credit rejection sample can be the user with a network access time of more than 24 months in the target credit rejection sample, and the preferred credit rejection ratio can be the ratio of users with a network access time of more than 24 months in the target credit rejection sample.

[0101] Therefore, the calculation formula of the target credit rejection sample overdue rate in this embodiment can also be:

[0102] Target credit rejection sample predicted overdue rate = target credit success sample overdue rate * (preferred credit success ratio / preferred credit rejection sample);

[0103] In the formula, the percentage of successful preferred credit granting / samples of rejected preferred credit granting is the above-mentioned overdue mapping ratio.

[0104] This embodiment determines the first network access time index of the target marked credit success sample and the second network access time index of the target credit rejection sample respectively according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval; determines the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample based on the first network access time index and the second network access time index; determines the predicted overdue rate of the target credit rejection sample based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample. Therefore, this embodiment can determine the predicted overdue rate for the target credit rejection sample based on the network access time index, that is, the network access time is more than 24 months, and then add a status label. Therefore, compared with the labeling method based on human business experience, it not only reduces the dependence on human experience, but also improves the accuracy of label setting.

[0105] In addition, an embodiment of the present invention further proposes a storage medium, on which a model iteration program is stored. When the model iteration program is executed by a processor, the steps of the model iteration method described above are implemented.

[0106] refer to Fig. 9 , Fig. 9 It is a structural block diagram of the first embodiment of the model iteration device of the present invention.

[0107] like Fig. 9 As shown, the model iteration device proposed in the embodiment of the present invention includes:

[0108] The scoring module 901 is used to score the marked credit success samples and the credit rejection samples by using a preset scoring model, and divide the marked credit success samples and the credit rejection samples into scoring intervals based on the scoring results;

[0109] A marking module 902, configured to set a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval, and obtain a marked credit rejection sample;

[0110] The model iteration module 903 is used to iteratively train the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges.

[0111] Furthermore, the scoring module 901 is also used to classify the initial samples according to their credit information to obtain credit success samples and credit rejection samples;

[0112] The scoring module 901 is further used to mark the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain a marked successful credit granting sample;

[0113] The scoring module 901 is also used to perform logistic regression training on the marked credit success samples to obtain a preset scoring model.

[0114] It should be understood that the repayment information in this embodiment includes: a repayment schedule and repayment behavior data; further, in this embodiment, the scoring module 901 is also used to obtain the repayment schedule and repayment behavior data corresponding to the credit success sample;

[0115] The scoring module 901 is further used to determine the user actual repayment table corresponding to the successful credit granting sample according to the repayment behavior data of the successful credit granting sample;

[0116] The scoring module 901 is also used to perform a preset risk analysis on the repayment plan table and the user actual repayment table, and mark the credit success sample based on the risk analysis result to obtain a marked credit success sample.

[0117] This embodiment classifies the initial samples according to the credit information of the initial samples to obtain credit success samples and credit rejection samples; obtains the repayment plan table and repayment behavior data corresponding to the credit success samples; determines the user actual repayment table corresponding to the credit success samples according to the repayment behavior data of the credit success samples; performs preset risk analysis on the repayment plan table and the user actual repayment table, and marks the credit success samples based on the risk analysis results to obtain marked credit success samples. Logistic regression training is performed on the marked credit success samples to obtain a preset scoring model. The marked credit success samples and credit rejection samples are scored by the preset scoring model, and the marked credit success samples and credit rejection samples are divided into scoring intervals based on the scoring results; a status label is set for the credit rejection samples according to the overdue distribution of the target marked credit success samples in each scoring interval to obtain a marked credit rejection sample; the preset scoring model is iteratively trained by marking the credit success samples and the marked credit rejection samples until the model converges. Therefore, this embodiment does not require unverifiable conditional assumptions, but directly adds feature labels to credit rejection samples based on the overdue distribution of marked credit success samples. At the same time, compared with the existing method of increasing samples through a new round of data collection, the present invention can expand samples through a model expansion method when the sample size is small, that is, the present application can directly expand the iterative samples of the model based on the marked credit rejection samples, thereby increasing the characteristics of the model iteration samples to improve the model fitting effect while reducing the iteration time and improving the model iteration efficiency.

[0118] Based on the above-mentioned first embodiment of the model iteration device of the present invention, a second embodiment of the model iteration device of the present invention is proposed.

[0119] In this embodiment, the marking module 902 is further used to determine the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target marked credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval;

[0120] The marking module 902 is further configured to randomly set a status label for the target credit rejection sample based on the predicted overdue rate to obtain a marked credit rejection sample.

[0121] This embodiment determines the predicted overdue rate of the target credit rejection sample based on the overdue distribution and network access information of the target marked credit success sample in each scoring interval and the network access information of the target credit rejection sample in the same scoring interval; randomly sets a status label for the target credit rejection sample according to the predicted overdue rate to obtain the marked credit rejection sample. Therefore, this embodiment can set a status label for the target credit rejection sample in the same scoring interval based on the overdue distribution of the target marked credit success sample, avoiding the problem that the overdue distribution of the credit rejection sample and the business overdue distribution of the credit success sample are greatly different. Therefore, the predicted overdue rate of the credit rejection sample determined by this embodiment is highly accurate, and can also improve the fitting effect of the preset scoring model for subsequent model iteration based on the marked credit rejection sample.

[0122] Based on the first embodiment or the second embodiment of the model iteration device of the present invention, a second embodiment of the model iteration device of the present invention is proposed.

[0123] In this embodiment, the marking module 902 is further used to determine the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval;

[0124] The marking module 902 is further configured to determine the predicted overdue rate of the target credit rejection sample based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample.

[0125] Further, as an implementable method, in this embodiment, the marking module 902 is further used to determine the first network access time index of the target marked credit success sample and the second network access time index of the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval;

[0126] The marking module 902 is further configured to determine an overdue mapping ratio between the target marked credit success samples and the target credit rejection samples based on the first network access duration indicator and the second network access duration indicator.

[0127] This embodiment determines the first network access time index of the target marked credit success sample and the second network access time index of the target credit rejection sample respectively according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval; determines the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample based on the first network access time index and the second network access time index; determines the predicted overdue rate of the target credit rejection sample based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample. Therefore, this embodiment can determine the predicted overdue rate for the target credit rejection sample based on the network access time index, that is, the network access time is more than 24 months, and then add a status label. Therefore, compared with the labeling method based on human business experience, it not only reduces the dependence on human experience, but also improves the accuracy of label setting.

[0128] Other embodiments or specific implementations of the model iteration device of the present invention can refer to the above-mentioned method embodiments and will not be described in detail here.

[0129] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0130] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0131] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0132] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A model iteration method, characterized in that: The model iteration method comprises: Scoring the marked credit success samples and the marked credit rejection samples by using a preset scoring model, and dividing the marked credit success samples and the marked credit rejection samples into scoring intervals based on the scoring results; According to the overdue distribution of the target marked credit success samples in each scoring interval, a status label is set for the credit rejection samples to obtain the marked credit rejection samples; The preset scoring model is iteratively trained through the marked credit success samples and the marked credit rejection samples until the model converges.

2. The model iteration method according to claim 1, characterized in that: The step of setting a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval to obtain the marked credit rejection sample comprises: Determine the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target marked credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval; A status label is randomly set for the target credit rejection sample based on the predicted overdue rate to obtain a labeled credit rejection sample.

3. The model iteration method according to claim 2, characterized in that: The step of determining the predicted overdue rate of the target credit rejection samples according to the overdue distribution and network access information of the target marked credit success samples in each scoring interval and the network access information of the target credit rejection samples in the same scoring interval includes: Determining the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval; The predicted overdue rate of the target credit rejection sample is determined based on the overdue mapping ratio and the overdue distribution of the target marked credit success sample.

4. The model iteration method according to claim 3, characterized in that: The step of determining the overdue mapping ratio between the target credit success sample and the target credit rejection sample according to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval includes: According to the first network access information of the target marked credit success sample in each scoring interval and the second network access information of the target credit rejection sample in the same scoring interval, respectively determine the first network access duration index of the target marked credit success sample and the second network access duration index of the target credit rejection sample; Based on the first network access duration indicator and the second network access duration indicator, an overdue mapping ratio between the target marked credit success samples and the target credit rejection samples is determined.

5. The model iteration method according to claim 4, characterized in that: The step of determining the overdue mapping ratio between the target marked credit success sample and the target credit rejection sample based on the first network access time indicator and the second network access time indicator comprises: Determine a preferred credit granting success sample based on the first network access information and the first network access duration indicator; Determine a preferred credit rejection sample based on the second network access information and the second network access duration indicator; Determine the preferred credit success ratio and the preferred credit rejection ratio in each scoring interval through the preferred credit success samples and the preferred credit rejection samples; According to the preferred credit success ratio and the preferred credit rejection ratio, the overdue mapping ratio between the target marked credit success samples and the target credit rejection samples is determined.

6. The model iteration method according to claim 1, characterized in that: Before scoring the marked credit success samples and credit rejection samples by using the preset scoring model, the method further includes: Classifying the initial samples according to their credit information to obtain credit success samples and credit rejection samples; Marking the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain a marked successful credit granting sample; Logistic regression training is performed on the marked credit success samples to obtain a preset scoring model.

7. The model iteration method according to claim 6, characterized in that: The repayment information includes: a repayment schedule and repayment behavior data; the step of marking the successful credit granting sample based on the repayment information of the successful credit granting sample to obtain the marked successful credit granting sample includes: Obtaining the repayment schedule and repayment behavior data corresponding to the successful credit granting sample; Determine the user actual repayment table corresponding to the successful credit granting sample according to the repayment behavior data of the successful credit granting sample; A preset risk analysis is performed on the repayment plan table and the user actual repayment table, and the credit success sample is marked based on the risk analysis result to obtain a marked credit success sample.

8. A model iteration device, comprising: A scoring module, used to score the marked credit success samples and the credit rejection samples by using a preset scoring model, and divide the marked credit success samples and the credit rejection samples into scoring intervals based on the scoring results; A marking module, used for setting a status label for the credit rejection sample according to the overdue distribution of the target marked credit success sample in each scoring interval, and obtaining the marked credit rejection sample; The model iteration module is used to iteratively train the preset scoring model through the marked credit success samples and the marked credit rejection samples until the model converges.

9. A model iteration device, characterized in that: The device comprises: a memory, a processor, and a model iteration program stored in the memory and executable on the processor, wherein the model iteration program is configured to implement the steps of the model iteration method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a model iteration program, and when the model iteration program is executed by the processor, the steps of the model iteration method according to any one of claims 1 to 7 are implemented.

Images