Automatic model framework and risk management and control method based on quasi-real-time effect

Abstract

The invention provides an automatic model framework based on a quasi-real-time effect and a risk management and control method, and belongs to the technical field of risk management, and the method specifically comprises the steps: a variable pool module is responsible for carrying out the extraction of credit features of credit granting application information of credit users to obtain a variable pool, and a quasi-real-time model module is responsible for carrying out the extraction of credit granting application information according to a preset period; and the data monitoring module is responsible for carrying out quasi real-time updating processing on the samples based on the variable pool, carrying out incremental training on the basic model by utilizing the samples subjected to confrontation processing based on the basic model, carrying out updating processing on the credit model, and carrying out online processing after updating is completed, and the data monitoring module is responsible for adding timestamp features into the credit model. And the change trend of the credit model and the customer group is analyzed by using the timestamp feature, so that the timeliness of update processing of the credit model is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk management, and particularly relates to an automated model framework based on quasi-real-time effect and a risk control method. Background Art

[0002] In the field of credit risk control, user risks often have a lag. To solve the above technical problems, existing technical solutions often determine the credit risk by the changing credit characteristics of users and the original risk models. Specifically, in the invention patent application CN202311379683.1 "A Multi-objective Risk Control Strategy Optimization Method and System", a similar technical solution is given. However, the above technical solution has the following technical problems: Since the risk control model is built based on the risk labels of historical users, when the portrait of the new customer group changes, the model cannot effectively capture user risks in a timely manner. At the same time, in the same model application scenario, it is necessary to consider building models with multiple risk labels or evaluating the effects of multiple data features. Therefore, how to automatically and real-time build a model to solve the technical knowledge, background, time-consuming and repetitive steps required in the actual application of machine learning methods is a relatively important topic.

[0003] To solve the above technical problems, this application provides an automated model framework based on quasi-real-time effect and a risk control method. Summary of the Invention

[0004] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: Specifically, in the first aspect, an automated model framework based on quasi-real-time effect of this application specifically includes: A variable pool module, a quasi-real-time model module, and a data monitoring module; The variable pool module is responsible for extracting the credit characteristics of the credit application information of credit users to obtain a variable pool; The quasi-real-time model module is responsible for performing quasi-real-time update processing of samples based on the variable pool at a preset cycle, and performing incremental training of the basic model using the samples after adversarial processing, and performing update processing of the credit model. After the update is completed, online processing is performed; The data monitoring module is responsible for adding a timestamp feature to the credit model and analyzing the change trends of the credit model and the customer group using the timestamp feature.

[0005] A further technical solution lies in that the basic model is constructed by building an AutoML framework.

[0006] In the second aspect, this application provides a risk control method, which is applied to the above-mentioned automated model framework based on quasi-real-time effect, and specifically includes: S1 classifies the credit users into different customer groups based on the credit application information of the credit users, and determines the adversarial processing customer groups in the customer groups according to the distribution of the customer groups in the training data of the credit model and the distribution data of the credit users on different dates; S2 takes the adversarial processing customer groups on the current date as the matching processing customer groups, and when it is determined that the training processing risk of the credit model does not meet the requirements according to the changes in the augmented data of different matching processing customer groups, proceeds to the next step; S3 determines the target impact customer groups in the customer groups and the impact correlation coefficients with different matching processing customer groups according to the changes in the credit user data of different customer groups, the credit user data of different matching processing customer groups, and the changes in the augmented data; S4 determines the verification processing strategies for different target impact customer groups during the training processing of different matching processing customer groups according to the impact correlation coefficients between the target impact customer groups and different matching processing customer groups and the data augmentation situation on the current date, determines the training processing order of different matching processing customer groups based on the verification processing strategies, and conducts credit risk control processing based on the updated credit model.

[0007] A further technical solution lies in that classifying the credit users into different customer groups specifically includes: Determining the number of identical information items in the credit application information of different credit users based on the similarity of the credit application information of the credit users; Classifying the credit users into different customer groups according to the number of identical information items.

[0008] A further technical solution lies in that classifying the credit users with the number of identical information items greater than the preset threshold of the number of identical information items into the same customer group.

[0009] A further technical solution lies in that the distribution of the customer groups in the training data of the credit model includes the number of credit users in the customer groups in the credit model and the proportion of the number of credit users in all credit users in the credit model.

[0010] A further technical solution lies in that the method for determining the adversarial processing customer groups in the customer groups is: Determining the proportion of the number of credit users in the customer groups in the historical training data of the credit model based on the distribution of the customer groups in the training data of the credit model, and using it as the proportion of customer group users; Determining the dates when there are credit users in the customer groups according to the distribution data of the credit users on different dates, and using them as the matching credit dates; Determine the confrontation processing requirement coefficient of the customer group based on the ratio of the quantity proportion of the matched credit dates to the quantity proportion of the customer group users, and determine whether the customer group belongs to the confrontation processing customer group based on the confrontation processing requirement coefficient.

[0011] A further technical solution is that when the confrontation processing requirement coefficient is greater than the preset processing requirement coefficient threshold, it is determined that the customer group belongs to the confrontation processing customer group.

[0012] A further technical solution is that the method for determining the training processing order of the matched processing customer group is as follows: With different verification processing strategies for the matched processing customer group, determine the target impact customer groups that need to be verified for different matched processing customer groups, and use them as the verification impact customer groups; According to different verification impact customer groups in different verification processing strategies of the matched processing customer group, determine the quantity proportion of the matched processing customer groups with verification processing in the matched processing customer group, and use it as the verification requirement coefficient; Based on the sum of the verification requirement coefficients of different verification impact customer groups of different matched processing customer groups, determine the verification requirement values of different matched processing customer groups, and determine the training processing order of different matched processing customer groups based on the verification requirement values.

[0013] A further technical solution is that determining the training processing order of different matched processing customer groups based on the verification requirement values specifically includes: Determine the training processing order of different matched processing customer groups from largest to smallest according to the verification requirement values, and after all the matched processing customer groups are trained in accordance with the training processing order, perform the training processing of the credit model according to all the customer groups except the matched processing customer groups.

[0014] A further technical solution is that performing credit risk control processing based on the updated credit model specifically includes: Based on the updated credit model, and combining the credit application information of different credit users as the input quantity, determine the credit risk of the credit users based on the output quantity of the updated credit model.

[0015] The beneficial effects of the present invention are as follows: According to the influence correlation coefficients between the target impact customer groups and different matched processing customer groups and the data amplification situation on the current date, determine the verification processing strategies of different target impact customer groups during the training processing of different matched processing customer groups, which not only ensures the reliability of the verification processing of the target impact customer groups with a higher degree of influence by the matched processing customer groups, but also avoids the technical problem that the verification processing efficiency is difficult to meet the requirements caused by performing verification processing during the training processing of different matched processing customer groups.

[0016] Based on the verification processing strategy, determine the training processing order for different matching processing customer groups, fully considering the number of target impact customer groups that need to be verified during training processing for different matching processing customer groups and the verification association situation between the target impact customer groups and other matching processing customer groups. This realizes the screening of matching processing customer groups with a relatively large number of target impact customer groups that need to be verified and a relatively large number of verification associations between the target impact customer groups and other matching processing customer groups, and promptly conducts verification processing, improving the efficiency and reliability of the positioning processing when the verification processing result is abnormal.

[0017] Other features and advantages will be described in the following specification. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification and the accompanying drawings.

[0018] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specifically provides preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By referring to the accompanying drawings and describing its exemplary embodiments in detail, the above and other features and advantages of the present invention will become more obvious.

[0020] Figure 1 is a framework diagram of an automated model framework based on a quasi-real-time effect; Figure 2 is a flowchart of a multi-source data verification and reach method for uninsured persons; Figure 3 is a flowchart of a method for determining visit-difficult persons among uninsured persons; Figure 4 is a flowchart of determining that the visit processing difficulty of the target area meets the requirements; Figure 5 is a flowchart of determining that optimization processing of the reach processing strategy is required. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To enable those skilled in the art of this technology to better understand the technical solutions in this specification, the following will, in conjunction with the accompanying drawings in the embodiments of this specification, clearly and completely describe the technical solutions in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0022] In the field of credit risk control, user risks often have a lag. Since the risk control model is built based on the risk labels of historical users, when the portrait of the new customer group changes, the model cannot effectively capture user risks in a timely manner. At the same time, in the same model application scenario, it is necessary to consider building models with multiple risk labels or evaluating the effects of multiple data features. Therefore, how to automatically and real-time build a model to solve the technical knowledge, background, time-consuming and repetitive steps required by machine learning methods in practical applications is a relatively important topic. The present invention proposes a method for updating a model based on an automated modeling framework with quasi-real-time effects, so as to quickly respond to the impact of customer group fluctuations on the online model and ensure the stability of the model effect.

[0023] The user data and other data involved in this application have been obtained with full consent and authorization, and the collection, use and processing of relevant information comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0024] 1) Implementation solution: 1. Variable pool module: Establish an alternative feature pool as a benchmark for variable screening, and on this basis, build an index screening system including stability, discrimination, and generality, and finally form an automated modeling standard variable pool; 2. Quasi-real-time model module: a. Automatic sample update: Adopt a sliding window + incremental data update mode to achieve quasi-real-time sample update; b. Automatic model update: First, build an AutoML framework to train the base model, and then introduce real-time sample information in the form of adversarial weights to perform incremental training on the basis of the base model, so as to achieve quasi-real-time model update; c. Automatic model online: Automatically convert the offline model file into an online format file for output, so as to realize the automation of the modeling process; d. Time series dimension data monitoring: Add a timestamp feature to the features of the tree model. When the traffic changes, the timestamp feature has a high importance. By parsing the tree structure, it can be detected how and when time affects the data. In addition, the differences in traffic before and after the split created by the timestamp can be analyzed.

[0025] Embodiment 1 As Figure 1 shown, this application provides an automated model framework based on quasi-real-time effects, specifically including: Variable pool module, quasi-real-time model module, data monitoring module; The variable pool module is responsible for extracting the credit characteristics of the credit application information of credit users to obtain a variable pool; The quasi-real-time model module is responsible for performing quasi-real-time update processing of samples based on the variable pool according to a preset period, and performing incremental training of the basic model using the adversarially processed samples based on the basic model, so as to perform update processing of the credit model. After the update is completed, online processing is performed; The data monitoring module is responsible for adding timestamp features to the credit model and analyzing the changing trends of the credit model and customer groups using the timestamp features.

[0026] Furthermore, the basic model is constructed by building an AutoML framework.

[0027] Embodiment 2 In a second aspect, as Figure 2 shown, the present application provides a risk control method, which is applied to the above-mentioned automated model framework based on quasi-real-time effects, and specifically includes: S1 Based on the credit application information of credit users, the credit users are divided into different customer groups, and the adversarial processing customer groups in the customer groups are determined according to the distribution of the training data of the credit model in the customer groups and the distribution data of credit users on different dates; The adversarial processing demand coefficient of the customer group is determined by the ratio of the proportion of the number of dates with credit users in the customer group to the proportion of the number of credit users in the historical training data of the credit model of the customer group. When the adversarial processing demand coefficient is greater than the preset processing demand coefficient threshold, it is determined that the customer group belongs to the adversarial processing customer group.

[0028] S2 Taking the adversarial processing customer groups on the current date as the matching processing customer groups, when it is determined that the training processing risk of the credit model does not meet the requirements according to the change situation of the amplified data of different matching processing customer groups, proceed to the next step; When there is a matching processing customer group with the amount of amplified data greater than the preset data amount threshold, it is determined that the training processing risk of the credit model does not meet the requirements.

[0029] S3 Based on the change situation of the credit user data of different customer groups, the credit user data of different matching processing customer groups, and the change situation of the amplified data, determine the target impact customer groups in the customer groups and the impact correlation coefficients with different matching processing customer groups; Based on the change situation of the credit user data of different customer groups, determine the change quantity of the credit users in the customer group; Process the credit user data and the changes in the amplified data of different customer groups with different matching methods, determine the change quantity of credit users and the quantity of amplified users in different customer groups with different matching methods, and determine the influence correlation coefficients of different customer groups with different matching methods based on the average value of the ratios of the change quantity of credit users, the quantity of amplified users in different customer groups with different matching methods to the change quantity of credit users in the customer group. When the sum of the influence correlation coefficients of different customer groups with different matching methods is greater than the preset correlation coefficient threshold, it is determined that the customer group belongs to the target influence customer group.

[0030] Table 1 shows the change data of credit users in different customer groups on different dates

[0031] S4 determines the verification processing strategies for different target influence customer groups during the training processing of different matching processing customer groups according to the influence correlation coefficients between the target influence customer groups and different matching processing customer groups and the data amplification situation on the current date, determines the training processing order of different matching processing customer groups based on the verification processing strategies, and conducts credit risk control processing based on the updated credit model.

[0032] Verify the target influence customer groups in the matching processing customer groups where the influence correlation coefficient is greater than the preset influence correlation coefficient threshold or the quantity of amplified credit users is greater than the preset amplification quantity threshold.

[0033] Determine the training processing order of the matching processing customer groups from large to small according to the sum of the influence correlation coefficients with different target influence customer groups.

[0034] Further, the credit users are divided into different customer groups, specifically including: Determine the quantity of the same information items in the credit application information of different credit users according to the similarity of the credit application information of the credit users; Divide the credit users into different customer groups according to the quantity of the same information items.

[0035] Specifically, divide the credit users with the quantity of the same information items greater than the preset quantity threshold of the same information items into the same customer group.

[0036] It should be noted that the distribution of the customer group in the training data of the credit model includes the quantity of credit users in the customer group in the credit model and the proportion of the quantity of credit users in the customer group in all credit users in the credit model.

[0037] It can be understood that, as Figure 3 shown, the method for determining the adversarial processing customer group in the customer group is: Determine the proportion of the number of credit users of the customer group in the historical training data of the credit model based on the distribution of the customer group in the training data of the credit model, and use it as the proportion of the number of customer group users. Determine the dates when there are credit users in the customer group based on the distribution data of credit users on different dates, and use them as the matching credit dates. Determine the anti-processing demand coefficient of the customer group based on the ratio of the proportion of the number of matching credit dates to the proportion of the number of customer group users, and determine whether the customer group belongs to the anti-processing customer group based on the anti-processing demand coefficient.

[0038] Further, when the anti-processing demand coefficient is greater than the preset processing demand coefficient threshold, it is determined that the customer group belongs to the anti-processing customer group.

[0039] In another possible embodiment, the method for determining the anti-processing customer group in the customer group is as follows: Determine the number of credit users of the customer group in the historical training data of the credit model based on the distribution of the customer group in the training data of the credit model, and use it as the number of matching users. Determine the dates when there are credit users in the customer group based on the distribution data of credit users on different dates, and use them as the matching credit dates. Determine whether the customer group belongs to the anti-processing customer group based on the matching credit dates and the number of matching users.

[0040] Further, when the number of matching credit dates of the customer group is greater than the preset credit date number threshold and the number of matching users is less than the preset matching user number threshold, it is determined that the customer group belongs to the anti-processing customer group.

[0041] Further, the amplified data is determined based on the amplified credit users after the anti-processing of the credit data of the original credit users by the matching processing customer group.

[0042] Specifically, as Figure 4 shown, determining that the training processing risk of the credit model does not meet the requirements specifically includes: Determine the number of amplified credit users of different matching processing customer groups based on the change situation of the amplified data of different matching processing customer groups, and use it as the number of amplified users. Determine the matching processing customer groups with the number of amplified users greater than the preset amplified number threshold based on the number of amplified users of different matching processing customer groups, and use them as the amplified processing customer groups. Determine whether the training processing risk of the credit model meets the requirements based on the number of the amplified processing customer groups.

[0043] Further, when the number of customers in the amplification process is greater than the preset threshold of the number of customers in the amplification process, it is determined that the training process risk of the credit model does not meet the requirements.

[0044] It can be understood that when the training process risk of the credit model meets the requirements, after directly amplifying all the matched customers, the training process of the credit model is carried out, and there is no need to verify the credit model after the training process is completed.

[0045] In another possible embodiment, determining that the training process risk of the credit model does not meet the requirements specifically includes: Based on the changes in the amplification data of different matched customer groups, determine the number of credit users amplified for different matched customer groups, and use it as the amplified user number; According to the amplified user numbers of different matched customer groups, sum up the amplified user numbers of different matched customer groups, and use it as the total amplification number; Based on the total amplification number, determine whether the training process risk of the credit model meets the requirements.

[0046] Further, when the total amplification number is greater than the preset amplification value, it is determined that the training process risk of the credit model does not meet the requirements.

[0047] In another possible embodiment, determining that the training process risk of the credit model does not meet the requirements specifically includes: S41 Based on the changes in the amplification data of different matched customer groups, determine the number of credit users amplified for different matched customer groups, and use it as the amplified user number. According to the amplified user numbers of different matched customer groups, sum up the amplified user numbers of different matched customer groups, and use it as the total amplification number; Optionally, before entering step S42, it is also necessary to determine whether the total amplification number meets the requirements. When the total amplification number does not meet the requirements, it is determined that the training process risk of the credit model does not meet the requirements. Only when the total amplification number meets the requirements, then proceed to step S42.

[0048] In the case where the total amplification number does not meet the requirements, it indicates that the number of customers in the amplification process is greater than a certain threshold. Therefore, it will inevitably lead to deviations in the training data of the credit model. Therefore, it can be undoubtedly determined that the training process risk of the credit model does not meet the requirements at this time.

[0049] S42 Based on the number of credit users of different customer groups, determine the amplification influence value of different customer groups based on the ratio of the total amplification number to the number of credit users of the customer groups; Optionally, before entering step S43, it is also necessary to determine whether there is a customer group with a relatively large amplification influence value. When there is no customer group with an amplification influence value greater than the preset threshold, it can be directly determined that the training processing risk of the credit model meets the requirements.

[0050] Optionally, before entering step S43, it is also necessary to obtain the number of customer groups whose amplification influence value does not meet the requirements. When the number of customer groups whose amplification influence value does not meet the requirements is greater than the preset threshold for the number of customer groups with amplification influence, it is determined that the training processing risk of the credit model does not meet the requirements; Only when the number of customer groups whose amplification influence value does not meet the requirements is not greater than the preset threshold for the number of customer groups with amplification influence, then proceed to step S43.

[0051] S43 determines the matching processing customer groups with the number of amplified users greater than the preset amplification number threshold according to the number of amplified users in different matching processing customer groups, and uses them as the amplified processing customer groups. According to the ratio of the number of amplified users in different amplified processing customer groups to the number of credit users in different customer groups, determine different amplified processing customer groups, and combine the amplification influence values of different customer groups to determine the training processing risk coefficient of the credit model. Based on the training processing risk coefficient, determine whether the training processing risk of the credit model meets the requirements.

[0052] Optionally, the above step S43 includes the following content: Before evaluating the amplification influence coefficient, it is also necessary to determine whether the number of the amplified processing customer groups meets the requirements. If so, it is determined that the training processing risk of the credit model does not meet the requirements. If not, then evaluate the amplification influence coefficient.

[0053] Furthermore, when the training processing risk coefficient is greater than the preset training risk coefficient threshold, it is determined that the training processing risk of the credit model does not meet the requirements.

[0054] Specifically, as Figure 5 shown, the method for determining the target influence customer group in the customer group is: Determine the change quantity of the credit users in the customer group based on the change situation of the credit user data of different customer groups; Determine the change quantity of the credit users and the number of amplified users in different matching processing customer groups based on the change situation of the credit user data and the amplified data of different matching processing customer groups; Based on the average value of the ratio of the change quantity of the credit users in different matching processing customer groups, the number of amplified users to the change quantity of the credit users in the customer group, determine the influence correlation coefficient of different matching processing customer groups, and determine whether the customer group belongs to the target influence customer group according to the influence correlation coefficient of different matching processing customer groups.

[0055] Further, the change quantity of the credit users in the customer group is determined based on the number of credit users in the customer group during the initial training process of the credit model and the current number of credit users in the customer group.

[0056] It can be understood that determining whether the customer group belongs to the target influencing customer group according to the influence correlation coefficient of the customer group under different matching processes specifically includes: When there is a matching process customer group with an influence correlation coefficient greater than the preset influence correlation coefficient threshold, it is determined that the customer group is the target influencing customer group.

[0057] In another possible embodiment, the method for determining the target influencing customer group in the customer group is as follows: S31 Determine the change quantity of the credit users in the customer group based on the change situation of the credit user data of different customer groups, determine the change quantity of the credit users in different matching process customer groups and the amplified user quantity based on the change situation of the credit user data and the amplified data of different matching process customer groups, and determine the influence correlation coefficient of different matching process customer groups based on the ratio of the change quantity of the credit users, the amplified user quantity in different matching process customer groups to the change quantity of the credit users in the customer group; S32 Determine the change influence coefficient of different matching process customer groups based on the change quantity of the credit users and the amplified user quantity in different matching process customer groups, and in combination with the current number of credit users in the customer group; S33 Determine the comprehensive influence value of the customer group according to the influence correlation coefficient and the associated influence coefficient of different matching process customer groups, and determine whether the customer group belongs to the target influencing customer group based on the comprehensive influence value.

[0058] Further, when the comprehensive influence value of the customer group is greater than the preset comprehensive influence threshold, it is determined that the customer group is the target influencing customer group.

[0059] Optionally, the above step S31 includes the following content: S311 Determine the change quantity of the credit users in the customer group based on the change situation of the credit user data of different customer groups. When the change quantity of the credit users in the customer group is less than the preset change quantity threshold, go to step S312; when the change quantity of the credit users in the customer group is not less than the preset change quantity threshold, go to step S313; S312 When the current number of credit users in the customer group is less than the preset credit user number threshold, it is determined that the customer group belongs to the target influencing customer group; when the current number of credit users in the customer group is not less than the preset credit user number threshold, go to step S313; S313 processes the credit user data and the changes in the amplified data of different matching treatment customer groups, determines the change quantity of the credit users and the amplified user quantity of different matching treatment customer groups, and determines the impact correlation coefficients of different matching treatment customer groups based on the ratio of the change quantity of the credit users, the amplified user quantity of different matching treatment customer groups to the change quantity of the credit users of the customer group; S314 When there is a matching treatment customer group with an impact correlation coefficient greater than the preset impact correlation coefficient threshold, it is determined that the customer group belongs to the target impact customer group. When there is no matching treatment customer group with an impact correlation coefficient greater than the preset impact correlation coefficient threshold, it proceeds to step S315; S315 When the sum of the impact correlation coefficients of different matching treatment customer groups does not meet the requirements, it is determined that the customer group belongs to the target impact customer group. When the sum of the impact correlation coefficients of different matching treatment customer groups meets the requirements, it proceeds to step S32.

[0060] Optionally, the following content is included in the above step S32: S321 Based on the change quantity of the credit users and the amplified user quantity of different matching treatment customer groups, and in combination with the current quantity of the credit users of the customer group, determines the change impact coefficients of different matching treatment customer groups. When the change impact coefficients of different matching treatment customer groups are all less than the preset change impact coefficient threshold, it proceeds to step S322. When there is a matching treatment customer group with a change impact coefficient not less than the preset change impact coefficient threshold, it proceeds to step S323; S322 When the sum of the impact correlation coefficients of different matching treatment customer groups is less than the preset value of the impact correlation coefficient, it is determined that the customer group does not belong to the target impact customer group. When the sum of the impact correlation coefficients of different matching treatment customer groups is not less than the preset value of the impact correlation coefficient, it proceeds to step S323; S323 When the quantity of the matching treatment customer groups with a change impact coefficient not less than the preset change impact coefficient threshold does not meet the requirements, it is determined that the customer group belongs to the target impact customer group. When the quantity of the matching treatment customer groups with a change impact coefficient not less than the preset change impact coefficient threshold meets the requirements, it proceeds to step S33.

[0061] Further, the method for determining the verification processing strategy when the target impact customer group is trained and processed by different matching treatment customer groups is: Based on the data amplification situation of the matching treatment customer group on the current date, determines the amplified user quantity of the matching treatment customer group on the current date; Based on the proportion of the amplified user quantity of different matching treatment customer groups in the amplified user quantity on the current date, determines the customer group weight coefficients of different matching treatment customer groups; Determine the customer group impact value of the target impact customer group according to the sum of the products of the impact correlation coefficients and customer group weight coefficients of the target impact customer group and different matching processing customer groups, and determine the verification processing strategy of the target impact customer group during different matching processing customer group training processes based on the customer group impact value.

[0062] Specifically, determining the verification processing strategy of the target impact customer group during different matching processing customer group training processes based on the customer group impact value specifically includes: When the customer group impact value is greater than the preset customer group impact value threshold, verification processing of the target impact customer group is required after different matching processing customer group training processes and all customer group training processes are completed. When the customer group impact value is not greater than the preset customer group impact value threshold, and when the customer group impact value is less than the preset impact threshold, verification processing of the target impact customer group is not required during different matching processing customer group training processes. When the customer group impact value is not less than the preset impact threshold, determine the verification processing strategy of the target impact customer group during the training process of different matching processing customer groups according to the product of the impact correlation coefficient and customer group weight coefficient of different matching processing customer groups and the target impact customer group.

[0063] Further, when the product of the impact correlation coefficient and customer group weight coefficient of the matching processing customer group and the target impact customer group is greater than the preset weight coefficient threshold, it is determined that verification processing needs to be performed in the target impact customer group during the training process of the matching processing customer group.

[0064] It should be noted that the method for determining the training processing order of the matching processing customer group is as follows: Determine the target impact customer groups that need to be verified for different matching processing customer groups according to the verification processing strategies of different matching processing customer groups, and use them as verification impact customer groups. According to the verification processing strategies of different verification impact customer groups in different matching processing customer groups, determine the proportion of the number of matching processing customer groups with verification processing in the matching processing customer groups, and use it as the verification requirement coefficient. Based on the sum of the verification requirement coefficients of different verification impact customer groups of different matching processing customer groups, determine the verification requirement value of different matching processing customer groups, and determine the training processing order of different matching processing customer groups based on the verification requirement value.

[0065] Further, determining the training processing order of different matching processing customer groups based on the verification requirement value specifically includes: Determine the training processing order of different matching processing customer groups in descending order of the verification requirement values. After all the matching processing customer groups are trained in accordance with the training processing order, perform the training processing of the credit model based on all the customer groups except the matching processing customer groups.

[0066] It can be understood that the control and management of credit risk are performed based on the updated credit model, which specifically includes: Based on the updated credit model, and combining the credit application information of different credit users as the input quantity, determine the credit risk of the credit user based on the output quantity of the updated credit model.

[0067] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between each embodiment, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the embodiments of the device, equipment, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0068] The above specifically describes a particular embodiment of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] The above is only one or more embodiments of this specification and is not used to limit this specification. For those skilled in the art, one or more embodiments of this specification can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of one or more embodiments of this specification shall be included within the scope of the claims of this specification.

Claims

1. An automated model framework based on quasi-real-time effects, characterized in that, Specifically include: Variable pool module, quasi-real-time model module, data monitoring module; The variable pool module is responsible for extracting the credit characteristics of the credit application information of credit users to obtain a variable pool; The quasi-real-time model module is responsible for performing quasi-real-time update processing of samples based on the variable pool according to a preset cycle, and based on the basic model, using the processed samples after adversarial processing to perform incremental training of the basic model, and performing update processing of the credit model. After the update is completed, online processing is performed; The data monitoring module is responsible for adding timestamp features to the credit model and analyzing the change trends of the credit model and the customer group using the timestamp features.

2. The automated model framework based on the quasi-real-time effect according to claim 1, wherein The basic model is constructed by building an AutoML framework.

3. A risk control method, applied to an automated model framework based on quasi-real-time effects according to any one of claims 1-2, characterized in that, Specifically include: Based on the credit application information of credit users, divide the credit users into different customer groups, and determine the adversarial processing customer groups in the customer groups based on the distribution of the customer groups in the training data of the credit model and the distribution data of credit users on different dates; Take the adversarial processing customer group on the current date as the matching processing customer group, and when the change situation of the amplified data of different matching processing customer groups determines that the training processing risk of the credit model does not meet the requirements, proceed to the next step; Based on the change situation of the credit user data of different customer groups, the credit user data of different matching processing customer groups, and the change situation of the amplified data, determine the target impact customer groups in the customer groups and the impact correlation coefficients with different matching processing customer groups; According to the impact correlation coefficients between the target impact customer groups and different matching processing customer groups and the data amplification situation on the current date, determine the verification processing strategies for different target impact customer groups during the training processing of different matching processing customer groups, determine the training processing order of different matching processing customer groups based on the verification processing strategies, and perform credit risk control processing based on the updated credit model.

4. The risk control method according to claim 3, characterized in that Dividing the credit users into different customer groups specifically includes: Based on the similarity of the credit application information of the credit users, determine the number of identical information items in the credit application information of different credit users; Divide the credit users into different customer groups according to the number of identical information items.

5. The risk control method according to claim 4, wherein Credit users with the number of identical information items greater than the preset threshold of the number of identical information items are divided into the same customer group.

6. The risk control method according to claim 3, wherein The distribution of the customer groups in the training data of the credit model includes the number of credit users in the customer groups in the credit model and the proportion of the number of all credit users in the credit model.

7. The risk control method according to claim 3, characterized in that, The method for determining the adversarial processing customer groups in the customer groups is: Based on the distribution of the customer groups in the training data of the credit model, determine the proportion of the number of credit users in the customer groups in the historical training data of the credit model, and use it as the proportion of customer group users; According to the distribution data of credit users on different dates, determine the dates when there are credit users in the customer groups, and use them as the matching credit dates; Determine the confrontation processing demand coefficient of the customer group based on the ratio of the number ratio of the matched credit dates to the number ratio of the customer group users, and determine whether the customer group belongs to the confrontation processing customer group based on the confrontation processing demand coefficient.

8. The risk control method according to claim 7, wherein When the confrontation processing demand coefficient is greater than the preset processing demand coefficient threshold, it is determined that the customer group belongs to the confrontation processing customer group.

9. The risk control method according to claim 3, wherein The method for determining the training processing order of the matched processing customer group is as follows: With different verification processing strategies for the matched processing customer group, determine the target influencing customer group that needs to be verified for different matched processing customer groups, and use it as the verification influencing customer group; According to different verification influencing customer groups in different verification processing strategies of the matched processing customer group, determine the number ratio of the matched processing customer groups with verification processing in the matched processing customer group, and use it as the verification demand coefficient; Based on the sum of the verification demand coefficients of different verification influencing customer groups of different matched processing customer groups, determine the verification demand value of different matched processing customer groups, and determine the training processing order of different matched processing customer groups based on the verification demand value.

10. The risk control method according to claim 9, characterized in that Determining the training processing order of different matched processing customer groups based on the verification demand value specifically includes: Determine the training processing order of different matched processing customer groups in descending order of the verification demand value, and after all the matched processing customer groups are trained in accordance with the training processing order, perform the training processing of the credit model according to all the customer groups except the matched processing customer groups.

Images