Risk label processing method and device, risk control method and equipment, and storage medium

Abstract

This invention discloses a risk label processing method, apparatus, risk control method, device, and storage medium, comprising: responding to a risk scenario processing request and obtaining the feature matrix of the training samples corresponding to the risk model and their training sample labels; re-encoding the feature matrix of the training samples using labels to generate a labeled feature matrix; sorting the labeled feature matrix by label importance and extracting multiple important labels based on the sorting results; for each important label, obtaining real feature data of the important label over multiple historical periods, and using the risk model to obtain predicted feature data of the important label over multiple historical periods; and calculating the risk level of each important label based on the combined feature data of each important label using a maximum risk drawdown model. Based on the maximum risk drawdown algorithm, this invention can automatically select high-risk labeled customer groups and can periodically and automatically generate monitoring SQL for high-risk labeled customer groups for monitoring deployment.

Description

Technical Field

[0001] This invention belongs to the field of risk identification and monitoring control technology, specifically relating to a risk label processing method, device, risk control method, equipment, and storage medium. Background Technology

[0002] With the rapid development of information and internet technologies, online businesses have experienced rapid growth and widespread application, such as payment, registration, marketing, and credit lending. However, this rapid development has also brought with it frequent instances of fraudulent activities. Therefore, improving the security of online businesses and controlling their risks has received increasing attention and importance. Against this backdrop, risk models corresponding to online business (risk scenarios) have emerged: These models involve creating user profiles of the customer groups or users corresponding to the business scenarios being analyzed, establishing customer group tags across various dimensions in a database, extracting user characteristic data under these multi-dimensional tags, and using this data sample to train an AI model to form a corresponding risk assessment model.

[0003] In the field of online business risk control, after a risk model is deployed, it is often necessary to analyze, issue early warnings, and monitor the performance of high-risk customer groups based on new data to ensure the robustness and effectiveness of the model after deployment—that is, risk control. In practice, determining which tags are risky and which customer groups to analyze is often done manually by business personnel. This manual selection is time-consuming and inefficient. Furthermore, the actual meaning of manually selected tags may gradually become invalid over time, and it is difficult to quantify the risk level of the tags. Consequently, the resulting risk control strategies also suffer from poor accuracy. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a risk label processing method, device, risk control method, equipment and storage medium. The method is based on the maximum risk drawdown algorithm to quantify the high-risk labeled customer groups corresponding to the target risk scenario, thereby achieving accurate risk monitoring and control.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0006] A risk label processing method, the method comprising:

[0007] In response to a risk scenario processing request, the risk scenario processing request includes: a target risk scenario and its corresponding risk model; obtaining the feature matrix of the training samples corresponding to the risk model and the labels of the training samples;

[0008] The feature matrix of the training samples is re-encoded with labels to generate a labeled feature matrix; and the labeled feature matrix is ​​sorted by label importance based on the labels of the training samples.

[0009] Based on the sorting results, multiple important labels are extracted from the labeled feature matrix; for each important label, the real feature data of the important label under multiple historical periods are obtained, and the risk model is used to obtain the predicted feature data of the important label under multiple historical periods. The real feature data of the important label under multiple historical periods and the predicted feature data are combined to obtain the combined feature data corresponding to each important label.

[0010] Based on the combined feature data of each of the important labels, the risk level of each of the important labels is calculated using the maximum risk drawdown model.

[0011] According to a specific implementation, in the above-mentioned risk label processing method, the step of performing label-based recoding on the training data feature matrix based on the training data labels includes:

[0012] For each feature x in the feature matrix X of the training data i Determine x i Is it an enumeration value? If yes, then an enumeration value is a label; otherwise, it is x. i If it is a continuous value, then a density binning strategy is used to divide x... i Divide into several labels, and apply the resulting labels to x. i Perform recoding;

[0013] The labeled feature matrix is ​​generated by iterating through all features of the training data feature matrix X in sequence.

[0014] According to a specific implementation, in the above-mentioned risk label processing method, the step of ranking the labeled feature matrix by label importance based on the training sample labels includes:

[0015] Using the XGBOOST model, the importance of the labels in the labeled feature matrix is ​​ranked based on the labels of the training samples.

[0016] According to a specific implementation method, the target risk scenarios in the above-mentioned risk label processing method include: violation and prohibition risk scenarios, anti-fraud risk scenarios, investment and wealth management risk scenarios, fraud risk scenarios, lending risk scenarios, and insurance business scenarios.

[0017] Another aspect of the present invention provides a risk control method, comprising:

[0018] The risk level of the important labels corresponding to the target risk scenario is calculated using the risk label processing method described above.

[0019] Important tags are sorted according to their risk level, and multiple high-risk tags are extracted based on the sorting results;

[0020] For each of the high-risk tags, a corresponding monitoring dimension is set, and based on the monitoring dimension of each high-risk tag, an SQL monitoring description is generated for each high-risk tag to monitor the high-risk tags under the target risk scenario.

[0021] According to one specific implementation, in the above risk control method, generating the SQL monitoring description corresponding to each high-risk tag includes:

[0022] The SQL monitoring description corresponding to each high-risk label is generated using regular expressions or abstract syntax tree algorithms.

[0023] According to a specific implementation method, the monitoring dimensions set for each high-risk label in the above risk control method include: the ratio of the label customer group to all customer groups, the change rate of the label customer group in the current period compared to the previous period, and the instantaneous pulse volume of the label customer group.

[0024] In another aspect, the present invention provides a risk label processing apparatus, the apparatus comprising:

[0025] The response module is used to respond to risk scenario processing requests, which include: the target risk scenario and its corresponding risk model.

[0026] The acquisition module is used to obtain the feature matrix and training sample labels of the training samples corresponding to the risk model;

[0027] The encoding module is used to perform label re-encoding on the feature matrix of the training samples to generate a label-based feature matrix; and to sort the label-based feature matrix by label importance based on the labels of the training samples.

[0028] The first calculation module is used to extract multiple important labels from the labeled feature matrix according to the sorting result; for each important label, the module obtains the real feature data of the important label under multiple historical periods, and uses the risk control module to combine the real feature data of the important label under multiple historical periods with the predicted feature data to obtain the combined feature data corresponding to each important label.

[0029] The second calculation module is used to calculate the risk level of each important label based on the combined feature data of each important label using the maximum risk drawdown model.

[0030] In another aspect, the present invention provides an electronic device comprising:

[0031] One or more processors;

[0032] A memory is used to store one or more programs that, when executed by one or more processors, cause the one or more processors to implement the aforementioned risk label processing method or the aforementioned risk control method.

[0033] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the above-described risk label processing method, or implements the above-described risk control method.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] The method provided in this invention obtains multiple important labels of a risk model corresponding to a target risk scenario. For each important label, it acquires real feature data over multiple historical periods and uses the risk model to obtain predicted feature data for the important label over multiple historical periods. The real and predicted feature data of the important label over multiple historical periods are then combined to obtain combined feature data corresponding to each important label. Based on the combined feature data of each important label, the risk level of each important label is calculated using a maximum risk drawdown model. Therefore, this invention objectively quantifies the risk level of labels through a maximum risk drawdown algorithm, providing a basis for the generation of risk control strategies and improving the efficiency and accuracy of risk control. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the risk label processing method according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the risk control method according to an embodiment of the present invention.

[0038] Figure 3 This is a schematic diagram of the risk label processing device architecture according to an embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the risk control system architecture according to an embodiment of the present invention.

[0040] Figure 5 This is a schematic diagram of the electronic device architecture according to an embodiment of the present invention. Detailed Implementation

[0041] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0042] Example 1

[0043] Please see Figure 1 , Figure 1 The risk label processing method provided in this embodiment of the invention includes the following steps:

[0044] S1: Respond to a risk scenario processing request, the risk scenario processing request including: the target risk scenario and its corresponding risk model; obtain the feature matrix of the training samples corresponding to the risk model and the labels of the training samples;

[0045] S2: Perform label recoding on the feature matrix of the training samples to generate a labeled feature matrix; and, based on the labels of the training samples, sort the labeled feature matrix by label importance;

[0046] S3: Extract multiple important labels from the labeled feature matrix according to the sorting results; for each important label, obtain the real feature data of the important label under multiple historical periods, and use the risk model to obtain the predicted feature data of the important label under multiple historical periods, and combine the real feature data of the important label under multiple historical periods with the predicted feature data to obtain the combined feature data corresponding to each important label;

[0047] S4: Based on the combined feature data of each of the important labels, calculate the risk level of each of the important labels using the maximum risk drawdown model.

[0048] In this embodiment, multiple important labels of the risk model corresponding to the target risk scenario are obtained. For each important label, real feature data of the important label under multiple historical periods are obtained, and the risk model is used to combine the real feature data of the important label under multiple historical periods with the predicted feature data to obtain combined feature data corresponding to each important label. Based on the combined feature data of each important label, the risk degree of each important label is calculated using the maximum risk drawdown model. Accordingly, this embodiment objectively quantifies the risk degree of the label through the maximum risk drawdown algorithm, providing a basis for the generation of risk control strategies and improving the efficiency and accuracy of risk control.

[0049] Example 2

[0050] In one possible implementation, in S1 above, when a user wants to assess the risk model corresponding to their business, the user can log in to the risk label processing device (which can be understood as a related application, application client, or webpage on the terminal) through a terminal to input and submit a risk label processing request. The risk scenario processing request includes: the target risk scenario and its corresponding risk model. The risk label processing device (server) can receive risk label processing requests for different target risk scenarios. After receiving the request, it extracts information about the risk model of the target business scenario, extracting the training sample data set of the risk model (the feature matrix of the training samples and their training sample labels). Specifically, the target risk scenarios include: violation and prohibition risk scenarios, anti-fraud risk scenarios, investment and wealth management risk scenarios, fraud risk scenarios, credit lending risk scenarios, insurance business scenarios, etc.

[0051] Taking the user's risk scenario as an example of insurance business, the user sends a risk tag processing request to the risk tag processing device provided in this embodiment of the invention, attaching their own business scenario information and relevant information of the insurance business scenario risk model established by the user. By parsing the user's request, the risk tag device provided in this embodiment of the invention can extract the following information: 1. The risk tags set by the user can be divided into the following data dimensions: customer identity information, credit information, fund flow information, internet information, and historical policy information. 2. Each data dimension can include multiple risk tags, such as customer identity information including: gender tag, occupation tag, age tag, etc.; credit information including: loan tag, loan overdue tag, etc.; fund flow information including: monthly fund inflow tag, monthly fund outflow tag, etc.; internet information including: purchase behavior tag, etc.; historical policy information including: policy change tag, policy category tag, policy amount tag, etc. 3. The user obtains multiple raw feature data corresponding to pre-configured risk tags from the customer database. Based on these raw feature data, a training sample data set for the insurance business risk model can be established. The training sample data set includes the feature matrix of the training samples and their training sample tags. Accordingly, this device can extract the feature matrix and training sample tag data information of the training samples corresponding to the risk model.

[0052] In one possible implementation, S2 specifically includes: traversing all features of the feature matrix X, and when traversing to the i-th feature x... i When, for feature x i Perform tagging and encoding. Determine x. i Is it an enumeration value? If yes, then an enumeration value is a label; otherwise, it is x. iIf the value is continuous, it is divided into several labels using a density binning strategy, and then filled and replaced.

[0053] It is understood that the features in the feature matrix of the training samples described in this embodiment of the invention refer to a column of data in the feature matrix other than the label column. The features are coded with labels, and the corresponding labels are distinct from the training sample labels (data labels, i.e., sample labels) described in this embodiment of the invention. The labels in the coded labels can be each specific value of an enumerated feature, such as gender, which has two labels: male and female. They can also be each specific part of a continuous feature after binning. For example, a user's loan balance from last month is a feature. After binning the user's loan balance from last month with labels, four labels can be obtained: less than 5000, 5000-10000, 10000-20000, and greater than 20000; a new feature matrix X_new is generated based on this.

[0054] The corresponding padding and replacement process includes: assuming feature x i The data is divided into four labels: xi_1, xi_2, xi_3, and xi_4. Four new columns are added to the original feature matrix X. Each column is filled with zero values ​​except for the corresponding label positions. After adding these four columns, the original feature matrix X is deleted. i After traversing all features, a fully labeled feature matrix table X_new is generated. The new feature matrix X_new and the data labels of the training samples are then modeled using the XGBOOST model. The XGBOOST model is chosen because it is insensitive to outliers and missing data; after the labeling process described above, X_new is a very sparse matrix, making it suitable for the XGBOOST model. The importance ranking of the labels is then output from X_new.

[0055] In one possible implementation, S3 specifically includes: extracting N important labels from the labeled feature matrix based on the sorting result, where N can be set as needed; sequentially traversing each important label, and when label N is reached, assuming the current data period is T, selecting the feature matrix data for periods T-1, T-2, ..., TM of label N, where M is a hyperparameter that can be adjusted; simultaneously using a risk model to predict the customer group data for each period of label N, and combining the prediction results for period T with the prediction results for each historical period, denoted as T-1_N, T-2_N, ..., T-M_N. The data for M data periods of label N and the prediction results of the risk model for the data for M data periods of label N are combined to form the combined adjustment data for label N.

[0056] In one possible implementation, S4 above includes: calculating the maximum risk drawdown of label N over M historical periods based on the maximum risk drawdown model, denoted as score_n, and using this maximum drawdown value as a risk quantification indicator for label N. The maximum risk drawdown model calculates the maximum drawdown value through the following steps: A. Traversing the M data periods, calculating relevant statistical indicators for label i in the M data periods, including but not limited to: the proportion r of the label's customer group in the current period to the total customer group. B. Subtracting the largest r from the smallest r in the M periods, denoted as the current maximum statistical risk drawdown s1. C. Using the risk model to predict data with data labels in the M data periods, calculating the F1 score of the model in each period based on the actual data labels. D. Subtracting the largest F1 score from the smallest F1 score in step C, denoted as the current maximum model predictive ability risk drawdown s2. E. The final maximum risk drawdown score is denoted as (s1+s2) / s1*S2. This yields the risk level of label N. After traversing multiple important labels, sort the labels by risk level from highest to lowest according to score_N.

[0057] Another aspect of the present invention, such as Figure 2 As shown, a risk control method is also provided, including:

[0058] S201. The risk level of the important labels corresponding to the target risk scenario is calculated using the above-mentioned risk label processing method;

[0059] S202. Sort important labels according to their risk level, and extract multiple high-risk labels based on the sorting results; among them, a preset number of high-risk labels can be extracted according to the risk monitoring scenario.

[0060] S203. Set corresponding monitoring dimensions for each of the high-risk tags, and generate SQL monitoring descriptions for each of the high-risk tags based on the monitoring dimensions of each high-risk tag, so as to monitor the high-risk tags in the target risk scenario.

[0061] In one possible implementation, the monitoring dimensions set by the server in S203 above include, but are not limited to: the ratio of tagged customer groups to all customer groups, the rate of change of tagged customer groups in the current period compared to the previous period, and the instantaneous pulse volume of tagged customer groups (pulse volume = ratio of tagged customer groups to all customer groups * (current number of tagged customer groups / previous number of tagged customer groups)^2, used to measure the instantaneous surge of this customer group. Based on the monitoring dimensions, a monitoring SQL code is automatically generated using regular expressions or abstract syntax trees to continuously monitor the relevant statistical indicators of risky customer groups. Implementation examples of the two methods are as follows:

[0062]

[0063]

[0064] In this embodiment, based on the maximum risk drawdown algorithm, this application can automatically select high-risk tagged customer groups and periodically and automatically generate monitoring SQL for these high-risk tagged customer groups for monitoring deployment. This invention is designed based on the robustness requirements of relevant models in the risk control field. It eliminates the need for manual selection of risk tags; instead, it automatically outputs risk tags for risk analysis based on actual online data through a tag risk ranking algorithm. Even when no corresponding data tags are generated, it automatically provides risk warnings for online data. After outputting risk tags, it can also automatically generate SQL code for tag statistical monitoring, automatically completing the monitoring deployment of risk tags.

[0065] Example 2

[0066] In another aspect of the present invention, Figure 3 The present application illustrates a risk label processing apparatus provided in an embodiment, comprising:

[0067] The response module is used to respond to risk scenario processing requests, which include: the target risk scenario and its corresponding risk model.

[0068] The acquisition module is used to obtain the feature matrix and training sample labels of the training samples corresponding to the risk model;

[0069] The encoding module is used to perform label re-encoding on the feature matrix of the training samples to generate a label-based feature matrix; and to sort the label-based feature matrix by label importance based on the labels of the training samples.

[0070] The first calculation module is used to extract multiple important labels from the labeled feature matrix according to the sorting result; for each important label, the module obtains the real feature data of the important label under multiple historical periods, and uses the risk control module to combine the real feature data of the important label under multiple historical periods with the predicted feature data to obtain the combined feature data corresponding to each important label.

[0071] The second calculation module is used to calculate the risk level of each important label based on the combined feature data of each important label using the maximum risk drawdown model.

[0072] In one possible implementation, the above-mentioned encoding module is configured to perform label recoding on the training data feature matrix by means of: for each feature x in the training data feature matrix X i Determine xi Is it an enumeration value? If yes, then an enumeration value is a label; otherwise, it is x. i If it is a continuous value, then a density binning strategy is used to divide x... i Divide into several labels, and apply the resulting labels to x. i Re-encoding is performed; all features of the training data feature matrix X are sequentially traversed to generate the labeled feature matrix. Furthermore, using the XGBOOST model, the label importance of the labeled feature matrix is ​​ranked based on the labels of the training samples.

[0073] In another aspect of the present invention, Figure 4 This application illustrates a risk control system provided in an embodiment, comprising: the aforementioned risk label processing device, a sorting module, an SQL component generation module, and an execution module. The sorting module sorts important labels according to their risk level and extracts multiple high-risk labels based on the sorting results. The SQL component generation module sets corresponding monitoring dimensions for each high-risk label and generates an SQL monitoring description for each high-risk label based on its monitoring dimensions. The execution module monitors the high-risk labels under the target risk scenario based on the generated SQL monitoring descriptions.

[0074] In one possible implementation, the SQL component generation module is configured to generate an SQL monitoring description corresponding to each high-risk tag using a regular expression or abstract syntax tree algorithm.

[0075] In another aspect of the invention, an electronic device is also provided, such as... Figure 5 As shown, the electronic device includes a processor, a network interface, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the aforementioned risk label processing method or risk control method.

[0076] In embodiments of the present invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0077] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0078] In another aspect, the present invention provides a computer storage medium storing program instructions, which, when executed by at least one processor, are used to implement the above-described risk label processing method or risk control method.

[0079] In one possible implementation, the storage medium may be a memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0080] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0081] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0082] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0083] It should be understood that the system disclosed in this invention can be implemented in other ways. For example, the division of modules is merely a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the communication connection between modules can be through some interfaces, indirect coupling or communication connections between servers or units, and can be electrical or other forms.

[0084] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one processing unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0085] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

Claims

1. A method for processing risk labels, characterized in that, The method includes: In response to a risk scenario processing request, the risk scenario processing request includes: a target risk scenario and its corresponding risk model; obtaining the feature matrix of the training samples corresponding to the risk model and the labels of the training samples; The feature matrix of the training samples is re-encoded with labels to generate a labeled feature matrix; and the labeled feature matrix is ​​sorted by label importance based on the labels of the training samples. Based on the sorting results, multiple important labels are extracted from the labeled feature matrix; for each important label, the real feature data of the important label under multiple historical periods are obtained, and the risk model is used to obtain the predicted feature data of the important label under multiple historical periods. The real feature data of the important label under multiple historical periods and the predicted feature data are combined to obtain the combined feature data corresponding to each important label. Based on the combined feature data of each important label, the risk level of each important label is calculated using the maximum risk drawdown model. The maximum risk drawdown value of the label in multiple historical periods is calculated based on the maximum risk drawdown model, and this maximum drawdown value is used as the risk quantification indicator of the label. The step of ranking the labeled feature matrix by label importance based on the training sample labels includes: Using the XGBOOST model, the importance of the labels in the labeled feature matrix is ​​ranked based on the labels of the training samples; The target risk scenarios include: violation and prohibition risk scenarios, anti-fraud risk scenarios, investment and wealth management risk scenarios, fraud risk scenarios, lending risk scenarios, and insurance business scenarios.

2. The risk label processing method according to claim 1, characterized in that, The step of re-encoding the training data feature matrix based on the training data labels includes: For the training data feature matrix X Each feature in x i ,judge x i Whether it is an enumeration value; if yes, then an enumeration value is a label; otherwise, it is... x i If the value is continuous, then a density binning strategy is used to divide it into... x i Divide into several labels, and then apply the resulting labels to... x i Perform recoding; Iterate through the feature matrix of the training data in sequence X All features are used to generate the labeled feature matrix.

3. A risk control method, characterized in that, include: The risk level of important labels corresponding to the target risk scenario is calculated using the risk label processing method described in claim 1 or 2. Important tags are sorted according to their risk level, and multiple high-risk tags are extracted based on the sorting results; For each of the high-risk tags, a corresponding monitoring dimension is set, and based on the monitoring dimension of each high-risk tag, an SQL monitoring description is generated for each high-risk tag to monitor the high-risk tags under the target risk scenario.

4. The risk control method as described in claim 3, characterized in that, The generated SQL monitoring description for each high-risk label includes: The SQL monitoring description corresponding to each high-risk label is generated using regular expressions or abstract syntax tree algorithms.

5. The risk control method as described in claim 3, characterized in that, The monitoring dimensions set for each high-risk label include: the percentage of the labeled customer group to all customer groups, the change rate of the labeled customer group in the current period compared to the previous period, and the instantaneous pulse volume of the labeled customer group. Wherein, pulse quantity = the ratio of the tagged customer group to all customer groups * (the number of people in the current tagged customer group / the number of people in the previous tagged customer group)^2, and this pulse quantity is used to measure the instantaneous pulse quantity of the tagged customer group.

6. A risk label processing device, characterized in that, The device includes: The response module is used to respond to risk scenario processing requests, which include: the target risk scenario and its corresponding risk model. The acquisition module is used to obtain the feature matrix and training sample labels of the training samples corresponding to the risk model; The encoding module is used to perform label re-encoding on the feature matrix of the training samples to generate a label-based feature matrix; and to sort the label-based feature matrix by label importance based on the labels of the training samples. The first calculation module is used to extract multiple important labels from the labeled feature matrix according to the sorting result; for each important label, the module obtains the real feature data of the important label under multiple historical periods, and uses the risk control module to combine the real feature data of the important label under multiple historical periods with the predicted feature data to obtain the combined feature data corresponding to each important label. The second calculation module is used to calculate the risk level of each important label based on the combined feature data of each important label using the maximum risk drawdown model. The maximum risk drawdown value of the label in multiple historical periods is calculated based on the maximum risk drawdown model, and this maximum drawdown value is used as the risk quantification indicator of the label. The step of ranking the labeled feature matrix by label importance based on the training sample labels includes: Using the XGBOOST model, the importance of the labels in the labeled feature matrix is ​​ranked based on the labels of the training samples; The target risk scenarios include: violation and prohibition risk scenarios, anti-fraud risk scenarios, investment and wealth management risk scenarios, fraud risk scenarios, lending risk scenarios, and insurance business scenarios.

7. An electronic device, characterized in that, The electronic device includes: One or more processors; A memory for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1-4, or the method as described in any one of claims 3-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 1 or 2, or the method as described in any one of claims 3-5.

Images