Risk control strategy mining method and device based on large model, equipment and storage medium

Abstract

The invention provides a risk control strategy mining method and device based on a large model, equipment and a storage medium. The method comprises the following steps: acquiring sample data; extracting feature information of the sample data to form a first feature information set; screening the first feature information set to generate a second feature information set; obtaining a model training algorithm based on the second feature information set; generating a risk control strategy mining model based on sample data corresponding to the second feature information set and the model training algorithm; and generating a risk control strategy based on the risk control strategy mining model. A more accurate risk control strategy mining model can be generated for generating a risk control strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk management, and in particular to a risk control strategy mining method and device, equipment and storage medium based on a large model. Background Art

[0002] With the rapid development of financial technology, financial institutions are facing increasingly complex market environments and regulatory requirements, and risk control has become a key link in ensuring financial stability and sustainable development.

[0003] Traditional risk control strategies often rely on manual experience and simple statistical models, which are insufficient in today's explosive growth of data. Therefore, there is an urgent need for a more intelligent, efficient and flexible risk control strategy mining method to cope with the ever-changing market conditions and risk management needs. Summary of the invention

[0004] The present invention provides a risk control strategy mining method and device, equipment and storage medium based on a large model, which can generate accurate risk control strategy mining models for various business scenarios, thereby generating more scientific and reasonable risk control strategies.

[0005] On the one hand, a risk control strategy mining method based on a large model is provided, including: Get sample data; Extracting feature information of the sample data to form a first feature information set; Screening the first feature information set to generate a second feature information set; Acquire a model training algorithm based on the second feature information set; Generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm; Generate a risk control strategy based on the risk control strategy mining model.

[0006] Optionally, screening the first feature information set to generate a second feature information set includes: Scoring the importance of feature information in the first feature information set; Performing statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information; Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

[0007] Optionally, acquiring a model training algorithm based on the second feature information set includes: Determine model performance indicators based on business requirements and sample data; A model training algorithm is obtained from an algorithm library based on the model performance indicator, and the algorithm library stores a plurality of pre-set model training algorithms.

[0008] Optionally, generating a risk control strategy mining model based on sample data corresponding to the second feature information set and the model training algorithm includes: Training an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set; The initial risk control strategy mining model is evaluated to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, the initial risk control strategy mining model is output as the risk control strategy mining model.

[0009] On the other hand, a risk control strategy mining device based on a large model is provided, comprising: An acquisition module, used to acquire sample data; and acquire a model training algorithm based on the second feature information set; An extraction module, used for extracting feature information of the sample data to form a first feature information set; A screening module, used for screening the first feature information set to generate a second feature information set; A generation module is used to generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm; and generate a risk control strategy based on the risk control strategy mining model.

[0010] Optionally, the screening module is used to: Scoring the importance of feature information in the first feature information set; Performing statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information; Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

[0011] Optionally, the acquisition module is used to: Determine model performance indicators based on business requirements and sample data; A model training algorithm is obtained from an algorithm library based on the model performance indicator, and the algorithm library stores a plurality of pre-set model training algorithms.

[0012] Optionally, the generating module is used to: Training an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set; The initial risk control strategy mining model is evaluated to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, the initial risk control strategy mining model is output as the risk control strategy mining model.

[0013] On the other hand, an electronic device is provided, comprising a risk control strategy mining device based on a large model as described in any of the above items.

[0014] On the other hand, a computer-readable storage medium is provided, wherein at least one program code is stored in the computer-readable storage medium, and the program code is executed by a processor to implement the risk control strategy mining method based on a large model as described in any of the above items.

[0015] The beneficial effects of the technical solution provided by the present disclosure are: In an embodiment of the present disclosure, a risk control strategy mining method is provided. The method obtains characteristic information of the sample data after analyzing the sample data, and filters the characteristic information of the sample data, and obtains a model training algorithm according to the filtered characteristic information, thereby generating a risk control strategy mining model, thereby improving the matching degree between the risk control strategy mining model and the business scenario, and improving the accuracy of the risk control strategy mining model, thereby generating a risk control strategy that is more in line with the business scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A method flow chart of a risk control strategy mining method provided by an embodiment of the present disclosure; Figure 2 A structural block diagram of a risk control strategy mining device provided by an embodiment of the present disclosure; Figure 3 A structural block diagram of an electronic device provided in an embodiment of the present disclosure.

[0018] 21: acquisition module; 22: extraction module; 23: screening module; 24: generation module; 31: processor; 32: memory. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] In recent years, artificial intelligence and machine learning technologies have been widely used in various fields, especially in financial risk management. Large models, especially deep learning models, are favored for their powerful data processing and pattern recognition capabilities. These models can automatically learn and extract complex features from data, which can then be used for prediction and decision support.

[0021] The purpose of this invention is to combine deep learning models (big models) with risk control strategy mining, use the powerful data processing capabilities of the deep learning model itself and the functions of the deep learning model itself to analyze the data, and select appropriate algorithms to generate risk control strategy mining models, thereby generating risk control strategies that are consistent with actual business scenarios.

[0022] Figure 1 A method flow chart of a risk control strategy mining method provided by an embodiment of the present disclosure. Figure 1 The method steps include: S11. Obtain sample data.

[0023] In one example, step S11 includes: Step 1: Get the initial sample data.

[0024] In one example, step 1 includes: The first step is to build a prompt word framework for data and business identification through the prompt word engineering of the large model.

[0025] The prompt word framework includes database type, database table structure, business information of the database table, detailed field information of the database table, detailed task information, task constraint information and output information data format.

[0026] In the disclosed embodiment, the big model may be an open source big model such as Tongyi Qianwen or Wenxin Yiyan.

[0027] In the second step, according to the specific database information configured, the knowledge base capability of the large model is used to query the information of the specified database in the knowledge base and reassemble it into the format determined in the prompt word framework and embed it in the prompt word.

[0028] The third step is to call the big model capability to generate accurate data source query statements based on the complete prompt word project completed above, and output them in the specified format.

[0029] The fourth step is to read the specific database type configured by the user and create a corresponding data connection based on the configured database type and database connection information (currently most mainstream databases are supported).

[0030] The fifth step is to query the database connection through the query statement information and process the results into the specified data format.

[0031] In step 1, information from different databases will be called for different business scenarios. Before calling the external database, the big model needs to build a database call path. Through the database call path, the big model can accurately call the external database.

[0032] Step 2: preprocess the initial sample data to form sample data.

[0033] In one example, step 2 includes: The first step is to test the initial sample data.

[0034] Load the data set (initial sample data) into memory, check the basic information of the data such as the number of rows, columns, data type, storage format, etc., and perform a quick statistical summary to view data distribution, missing values, outliers, etc.

[0035] The second step is to preprocess the initial sample data.

[0036] Perform data cleaning, remove duplicate records, handle inconsistent entries, delete or correct erroneous entries; handle missing values, determine whether they are random missing or non-random missing and fill in missing values. The large model selects a specific filling method from the mean / median / mode, forward / backward, prediction and other filling methods to fill in.

[0037] The third step is to preprocess the initial sample data for outlier detection.

[0038] The big model selects statistical methods (such as standard deviation, interquartile range) or machine learning methods (such as isolation forest) to identify outliers and deal with them.

[0039] In step 2, the initial sample data is preprocessed to form sample data, and the sample data is used for the subsequent training of the risk control strategy mining model, which can ensure that the training data used by the risk control strategy mining model is relatively accurate.

[0040] The fourth step is to normalize the initial sample data to form sample data.

[0041] The data type of the initial sample data is set to a numerical type, such as the user's age, etc., and the numerical type of the initial sample data is set to a categorical type, such as the user's credit information, etc. The initial sample data is normalized to form sample data.

[0042] S12: Extract feature information of the sample data to form a first feature information set.

[0043] In one example, step S12 includes: The large model performs feature engineering on the sample data, evaluates and establishes new features or converts existing features to form a first feature information set, and outputs the first feature information set in a certain format. The specific format of the first feature information is not limited here.

[0044] S13: Screen the first feature information set to generate a second feature information set.

[0045] In one example, step S13 includes: Step 1: Score the importance of feature information in the first feature information set.

[0046] In one example, step 1 includes: The large model selection scoring algorithm calculates the importance of each feature in the first feature information set, wherein the scoring algorithm may be a random forest, a gradient boosting tree, a least absolute shrinkage and selection algorithm (Least Absolute Shrinkage and Selection Operator, LASSO), a ridge regression, or the like.

[0047] In different business scenarios, the focus of the risk control strategy mining model may be different, and some of the feature information extracted from the first feature information set may have little relevance to the actual business scenario. Therefore, in step 1, the feature information in the first feature information set is scored to facilitate the subsequent removal of feature information with a low degree of relevance to the actual business scenario, that is, to remove unimportant feature information.

[0048] Step 2: Perform statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information.

[0049] In one example, step 2 includes: The big model performs statistical summary analysis on the first feature information set, calculates the statistics of each feature, such as the mean, median, standard deviation, minimum, maximum, quartile, etc., analyzes the correlation between features, and identifies multicollinearity problems. Among them, the correlation between score data is a function that the big model itself has.

[0050] In step 2, the correlation between the feature information in the first feature information set is further analyzed. It is only necessary to explain the definition of the correlation in the large model. The large model can automatically analyze the correlation between the feature information. The higher the correlation between the feature information, the more conducive it is to form a risk control strategy mining model with higher accuracy.

[0051] Step 3: Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

[0052] In one example, step 3 includes: The first step is to perform feature screening on the feature information in the first feature information set.

[0053] According to the result of the scoring algorithm, a threshold is set to remove features below this threshold. Recursive feature elimination or feature selection can be used. If the first feature information set does not meet expectations after evaluation, the large model can automatically reselect features of the sample data. Among them, the large model can reselect features of the sample data multiple times, with a maximum of 5 loop iterations. Of course, the number of loop iterations given above is only an example, and the present disclosure does not limit the maximum number of loop iterations.

[0054] In the second step, the importance and relevance of the feature information in the first feature information set are displayed through a visual chart.

[0055] According to the evaluation results, each feature is sorted by score and a corresponding visualization chart is generated. For example, a histogram is generated to show the distribution of features, a scatter plot is generated to show the linear and nonlinear relationship between features, and a box plot is generated to show the degree of affinity between potential outliers and data distribution.

[0056] Visual charts summarize feature importance, relevance, and distribution, and highlight removed or added features and their impact on the feature set.

[0057] The third step is to display the filtered first feature information set and the visualization chart to the user. The user can manually filter the feature information in the first feature information set according to the information displayed in the visualization chart. The user can manually intervene in the feature selection process according to the evaluation results and the chart, and feed back the final results to the big model.

[0058] In step 3, the user may filter the feature information in the first feature information set by manual selection, such as filtering out feature information with low relevance. Of course, the large model may also automatically filter according to the relevance between feature information.

[0059] In step S13, the feature information in the first feature information set is screened by the importance and relevance of the feature information, thereby generating a second feature information set, which is conducive to the subsequent generation of a more accurate risk control strategy mining model.

[0060] S14. Acquire a model training algorithm based on the second feature information set.

[0061] In one example, step S14 includes: Step 1: Determine model performance indicators based on business requirements and sample data.

[0062] Among them, in actual business scenarios, the risk control strategy mining model has multiple performance indicators, such as accuracy, recall, F1 score, etc. The focus of users' business needs in business scenarios may be different. For example, some users will focus on accuracy, so the risk control strategy mining model needs to focus on accuracy when establishing it.

[0063] Step 2: Obtain a model training algorithm from an algorithm library based on the model performance indicator, wherein the algorithm library stores a plurality of pre-set model training algorithms.

[0064] The big model obtains the contents of the algorithm library, evaluates and recommends the algorithm based on data characteristics, business requirements, algorithm content in the algorithm library, algorithm complexity, training speed, prediction accuracy, and model interpretability.

[0065] The algorithm library is pre-established and contains a variety of different model training algorithms, such as machine learning model training algorithms. By screening out multiple model training algorithms based on the feature information of the second feature information set, sample data corresponding to the second feature information set, business requirements, and features of the algorithms in the algorithm library, it is possible to further ensure that the trained model matches the actual business requirements and improve the accuracy of the model.

[0066] In step 2, after the big model screens out multiple suitable model training algorithms from the algorithm library, it feeds back the model training algorithms to the users, allowing the users to select or exclude specific algorithms based on professional knowledge and additional information, collects user feedback, and determines the final algorithm based on their feedback.

[0067] Users can also manually filter out the algorithms selected from the large model, or they can manually select the model training algorithm from the algorithm library.

[0068] S15. Generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm.

[0069] In one example, step S15 includes: Step 1: Train an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set.

[0070] In step S14, multiple model training algorithms will be screened out. In step 1, multiple initial risk control strategy mining models will be generated by utilizing multiple model training algorithms. The purpose of generating multiple risk control strategy mining models is to facilitate the subsequent screening of the risk control strategy mining model that best matches the user's business needs.

[0071] Step 2: Evaluate the initial risk control strategy mining model to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, output the initial risk control strategy mining model as the risk control strategy mining model.

[0072] In step 2, the performance of the model is evaluated on the validation set, using a variety of evaluation indicators such as accuracy, precision, recall, F1 score, etc. to detect whether the model has overfitting and underfitting phenomena. If all the above conditions are met, the initial risk control strategy mining model meets the requirements.

[0073] After the initial risk control strategy mining model meets the requirements, the big model will also score the initial risk control strategy mining model that meets the requirements based on the performance of the model, and display the model and the model score to the user. The user can manually filter out the required model as the risk control strategy mining model.

[0074] S16. Generate a risk control strategy based on the risk control strategy mining model.

[0075] In one example, step S16 includes: Step 1: Generate risk control strategy.

[0076] The generalization ability of the risk control strategy mining model is analyzed through learning curve analysis, and the rules of the model are mined and extracted. The quality of the rules is evaluated using indicators such as support, confidence, and improvement, and high-value rules are selected as risk control strategies.

[0077] At the same time, the training results and filtered rules are output to the user for viewing. If the user is not satisfied with the results, he or she can manually adjust the model parameters (such as weights and biases under linear regression) to retrain the risk control strategy mining model. The parameters can be adjusted and trained up to 5 times. Of course, the above maximum number of parameter adjustments is only used as an example.

[0078] Step 2: Output risk control strategy.

[0079] The large model outputs a report in the specified format, including a description of the model logic and assumptions, a data overview and preprocessing summary, the model training and verification process, model performance evaluation, model interpretation and rule extraction, and strategy recommendations and application guidance.

[0080] It is worth noting that, in the embodiments of the present disclosure, using a large model to generate a risk control strategy mining model has the following advantages: First, the big model can automatically parse the structure of the data source, understand its business model, and design corresponding query statements to extract the required data from the database. This process greatly simplifies the complexity of data acquisition and facilitates subsequent data analysis.

[0081] Second, by calling data preprocessing tools, the large model can automatically clean the data, remove noise and outliers, and perform feature selection, feature transformation, and feature construction to generate higher quality data sets for modeling.

[0082] Third, the large model can automatically select the appropriate machine learning algorithm and optimize the model parameters in an iterative manner to improve prediction accuracy and decision-making results. At the same time, the model can generate interpretable rules that are easy for users to understand and apply.

[0083] Fourth, considering that the specific needs and preferences of different institutions may be different, the big model provides a user-friendly interface that allows users to participate in the process of feature selection, algorithm customization, and parameter tuning to ensure that the final strategy is both efficient and in line with actual application scenarios.

[0084] Fifth, the large model can comprehensively consider various indicators (such as support, confidence, etc.) to screen and evaluate the effectiveness of strategies, and generate detailed analysis reports, including model explanations, visualization results and strategy recommendations, to help decision makers better understand and implement strategies.

[0085] In summary, the risk control strategy mining method based on large models integrates data science, machine learning and human-computer interaction technology, providing a comprehensive and intelligent solution for modern financial risk management.

[0086] Figure 2 This is a structural block diagram of a risk control strategy mining device provided by an embodiment of the present disclosure. Figure 2 , the device comprises: An acquisition module 21 is used to acquire sample data; and acquire a model training algorithm based on the second feature information set; An extraction module 22, configured to extract feature information of the sample data to form a first feature information set; A screening module 23, configured to screen the first feature information set to generate a second feature information set; The generation module 24 is used to generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm; and generate a risk control strategy based on the risk control strategy mining model.

[0087] Optionally, the screening module 23 is used to: Scoring the importance of feature information in the first feature information set; Performing statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information; Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

[0088] Optionally, the acquisition module 21 is used to: Determine model performance indicators based on business requirements and sample data; A model training algorithm is obtained from an algorithm library based on the model performance indicator, and the algorithm library stores a plurality of pre-set model training algorithms.

[0089] Optionally, the generating module 24 is used to: Training an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set; The initial risk control strategy mining model is evaluated to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, the initial risk control strategy mining model is output as the risk control strategy mining model.

[0090] Figure 3 This is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 3 The electronic device may include the above-mentioned risk control strategy mining device based on a large model. Generally, the configuration device includes: a processor 31 and a memory 32.

[0091] The processor 31 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 31 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 31 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0092] The memory 32 may include one or more computer-readable storage media, which may be non-transitory. The memory 32 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 32 is used to store at least one instruction, which is used to be executed by the processor 31 to implement the risk control strategy mining method based on a large model performed by an electronic device provided in the method embodiment of the present application.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A risk control strategy mining method based on a large model, characterized in that: include: Get sample data; Extracting feature information of the sample data to form a first feature information set; Screening the first feature information set to generate a second feature information set; Acquire a model training algorithm based on the second feature information set; Generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm; Generate a risk control strategy based on the risk control strategy mining model.

2. The risk control strategy mining method based on a large model according to claim 1 is characterized in that: The first feature information set is screened to generate a second feature information set, including: Scoring the importance of feature information in the first feature information set; Performing statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information; Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

3. The risk control strategy mining method based on a large model according to claim 1 is characterized in that: Acquiring a model training algorithm based on the second feature information set includes: Determine model performance indicators based on business requirements and sample data; A model training algorithm is obtained from an algorithm library based on the model performance indicator, and the algorithm library stores a plurality of pre-set model training algorithms.

4. The risk control strategy mining method based on a large model according to claim 1 is characterized in that: Generating a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm includes: Training an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set; The initial risk control strategy mining model is evaluated to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, the initial risk control strategy mining model is output as the risk control strategy mining model.

5. A risk control strategy mining device based on a large model, characterized in that: include: An acquisition module is used to acquire sample data; Acquire a model training algorithm based on the second feature information set; An extraction module, used for extracting feature information of the sample data to form a first feature information set; A screening module, used for screening the first feature information set to generate a second feature information set; A generation module is used to generate a risk control strategy mining model based on the sample data corresponding to the second feature information set and the model training algorithm; and generate a risk control strategy based on the risk control strategy mining model.

6. The risk control strategy mining device based on a large model according to claim 5 is characterized in that: The screening module is used to: Scoring the importance of feature information in the first feature information set; Performing statistical summary analysis on the feature information in the first feature information set to determine the correlation between each piece of feature information; Based on the scores and relevance of the feature information in the first feature information set, the feature information in the first feature information set is screened to form a second feature information set.

7. The risk control strategy mining device based on a large model according to claim 5 is characterized in that: The acquisition module is used for: Determine model performance indicators based on business requirements and sample data; A model training algorithm is obtained from an algorithm library based on the model performance indicator, and the algorithm library stores a plurality of pre-set model training algorithms.

8. The risk control strategy mining device based on a large model according to claim 5, characterized in that: The generation module is used for: Training an initial risk control strategy mining model based on the model training algorithm and sample data corresponding to the second feature information set; The initial risk control strategy mining model is evaluated to determine whether the performance indicators of the initial risk control strategy mining model meet the requirements. When the performance indicators of the initial risk control strategy mining model meet the requirements, the initial risk control strategy mining model is output as the risk control strategy mining model.

9. An electronic device, characterized in that: It includes a risk control strategy mining device based on a large model as described in any one of claims 5 to 8.

10. A computer-readable storage medium, characterized in that: At least one program code is stored in the computer-readable storage medium, and the program code is executed by a processor to implement the risk control strategy mining method based on a large model as described in any one of claims 1 to 4.

Images