Training data screening method, risk assessment method, device, equipment and medium

Abstract

The application relates to the fields of artificial intelligence and big data, in particular to a training data screening method and risk assessment method, a training data screening device, a computer device, a storage medium and a computer program product. The screening method comprises the following steps: obtaining a test sample set and a training sample set; the test sample set comprises test data of multiple dimensions, and the training sample set comprises training data of the same dimensions as the test data; determining data stability values of the test sample set and the training sample set; the data stability value is used for representing the change degree of the values of the data in the corresponding data set; and based on the difference distance between the data stability values of the test sample set and the training sample set, at least part of the training data is screened for training a business data model. The method can improve the balance between the sample data and improve the stability of the data model trained by the sample data.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for screening training data, a risk assessment method, a device for screening training data, a computer device, a storage medium, and a computer program product. Background Technology

[0002] With the continuous development of banking technology and the deepening of the digital transformation of the banking industry, the application of artificial intelligence technology, represented by machine learning, is becoming increasingly rich. For example, machine learning models trained with a large number of training samples can change the traditional marketing and risk control work in the banking industry based on expert rules and business experience, greatly improving the accuracy of application.

[0003] Currently, building machine learning models in the banking industry mainly involves acquiring a large number of customer samples, dividing a portion of these samples into training samples for training the model and a portion into test samples for testing the model, so as to build the final machine learning model for use using the training and test samples.

[0004] However, since there are often extreme samples (i.e. extreme data) in the training samples, the distribution of data between the training samples and the test samples obtained by the current partitioning method is quite different, which makes the partitioning of the two types of samples unbalanced. Ultimately, this leads to the poor stability of the trained machine learning model and the poor computational processing effect of the machine learning model. Summary of the Invention

[0005] Therefore, it is necessary to provide a method for screening training data, a risk assessment method, a device for screening training data, a computer device, a storage medium, and a computer program product that can ensure a balanced data distribution between training samples and test samples, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for filtering training data, the method comprising:

[0007] Obtain a test sample set and a training sample set; the test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data; the training data and the test data are business data recorded by the banking customer system.

[0008] Determine the data stability value of the test sample set and the data stability value of the training sample set; the data stability value is used to characterize the degree of change of the values ​​of each data in the corresponding data set.

[0009] Based on the difference distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set, at least a portion of the training data is selected for training the business data model.

[0010] In one embodiment, the training data includes positive and negative sample data with the same multiple dimensions as the test data;

[0011] The process of obtaining the test sample set and the training sample set includes:

[0012] Multiple positive sample data and multiple negative sample data are obtained from a business data repository corresponding to the banking customer system, and the training sample set is obtained by combining them.

[0013] Multiple test data sets are obtained from the business data repository to form the test sample set.

[0014] In one embodiment, the number of negative sample data stored in the business data repository is much greater than the number of positive sample data;

[0015] Obtain multiple positive sample data and negative sample data from the business data repository, including:

[0016] Extract all positive sample data stored in the business data repository to obtain multiple positive sample data sets; and

[0017] A portion of negative sample data is extracted from the business data repository to obtain multiple negative sample data; wherein, the magnitude of the difference distance between the stable values ​​of the positive sample data and the negative sample data is negatively correlated with the stability of the business data model.

[0018] In one embodiment, determining the data stability values ​​of the test sample set and the training sample set includes:

[0019] Based on the first type of statistical data corresponding to the test data and training data of various dimensions, the single-type stable value of the test data and the single-type stable value of the training data of each dimension are determined; the single-type stable value is used to characterize the degree of change of the values ​​of each data in a dimension; the test data and training data of various dimensions include at least the dimensions of total assets, monthly deposits, monthly outflows, business revenue, and monthly transaction frequency.

[0020] Determine the second type of statistical data corresponding to multiple single-type stable values ​​of the test data, and the third type of statistical data corresponding to multiple single-type stable values ​​of the training data;

[0021] Based on the second type of statistical data and the third type of statistical data, the data stability value of the test sample set and the data stability value of the training sample set are determined respectively.

[0022] In one embodiment, the first type of statistical data includes the standard deviation and the median;

[0023] The determination of the single-type stable value of the test data and the single-type stable value of the training data for each dimension, based on the first type of statistical data corresponding to the test data and the training data for each dimension, includes:

[0024] Based on the standard deviation and median of the test data for each dimension, calculate the unitype stable value of the test data for each dimension; and

[0025] Based on the standard deviation and median of the training data for each dimension, the unitype stable value of the training data for each dimension is calculated.

[0026] In one embodiment, both the second type of statistical data and the third type of statistical data include standard deviation and mean;

[0027] The step of determining the data stability value of the test sample set and the data stability value of the training sample set based on the second type of statistical data and the third type of statistical data, respectively, includes:

[0028] Based on the standard deviation and mean of various single-type stable values ​​in the test data, the data stability value of the test sample set is calculated; and

[0029] Based on the standard deviation and mean of various single-type stable values ​​in the training data, the data stability value of the training sample set is calculated.

[0030] In one embodiment, the step of selecting at least a portion of the training data for training the business data model based on the difference distance between the stable data values ​​of the test sample set and the stable data values ​​of the training sample set includes:

[0031] The difference distance is obtained by using the quotient of the stable data values ​​of the test sample set and the stable data values ​​of the training sample set; the difference distance is used to characterize the degree of dispersion between the two stable data values; the magnitude of the difference distance is negatively correlated with the stability of the business data model.

[0032] Based on the relationship between the difference distance and the preset distance limit, at least a portion of the training data is selected for training the business data model.

[0033] In one embodiment, obtaining the difference distance by using the quotient of the data stability values ​​of the test sample set and the data stability values ​​of the training sample set includes:

[0034] Calculate the absolute value of the difference between the quotient and the preset constant to determine the difference distance;

[0035] The step of selecting at least a portion of the training data for training the business data model based on the relationship between the difference distance and the preset distance limit includes:

[0036] If the difference distance is less than or equal to the distance limit, all training data in the training sample set are selected for training the business data model.

[0037] In one embodiment, the larger the difference distance, the lower the stability of the business data model; the smaller the difference distance, the higher the stability of the business data model.

[0038] The step of selecting at least a portion of the training data for training the business data model based on the relationship between the difference distance and the preset distance limit includes:

[0039] If the difference distance is greater than the distance limit, multiple new negative sample data are retrieved from the business data repository corresponding to the banking customer system;

[0040] By combining the new negative sample data with the positive sample data in the training sample set, a new training sample set is obtained.

[0041] Based on the new training sample set and the test sample set, the steps of determining the data stability value of the test sample set and the data stability value of the training sample set are repeated until the difference distance is less than or equal to the distance limit.

[0042] Secondly, this application provides a risk assessment method, which is applied to a banking customer system, and the method includes:

[0043] Obtain customer investment data;

[0044] The investment data is input into a trained risk assessment model to calculate the risk, thereby obtaining the risk score of the customer's investment based on the investment data; the risk assessment model is constructed using test data in the test sample set and training data in the training sample set, and the training data in the training sample set is selected based on the above-mentioned training data selection method;

[0045] Based on the risk score, the risk assessment result for the customer is determined.

[0046] Thirdly, this application also provides a training data filtering device, the device comprising:

[0047] The sample acquisition unit is configured to acquire a test sample set and a training sample set; the test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data; the training data and the test data are business data recorded by the banking customer system.

[0048] The data processing unit is configured to determine the data stability value of the test sample set and the data stability value of the training sample set; the data stability value is used to characterize the degree of change of the values ​​of each data in the corresponding data set.

[0049] The data filtering unit is configured to perform a distance-based filtering of at least a portion of the training data for training the business data model, based on the difference between the stable values ​​of the data in the test sample set and the stable values ​​of the data in the training sample set.

[0050] Fourthly, this application also provides an electronic device, the electronic device comprising:

[0051] processor;

[0052] Memory for storing the executable instructions of the processor;

[0053] The processor is configured to execute the executable instructions to implement the training data screening method and / or risk assessment method as described above.

[0054] Fifthly, this application also provides a computer-readable storage medium. The computer-readable storage medium includes program data that, when executed by a processor of an electronic device, enables the electronic device to perform the training data screening method and / or risk assessment method as described above.

[0055] Sixthly, this application also provides a computer program product. The computer program product includes program instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the training data screening method and / or risk assessment method as described above.

[0056] The aforementioned training data screening method, risk assessment method, training data screening device, computer equipment, storage medium, and computer program product firstly involve acquiring a test sample set and a training sample set. The test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data. Both the training and test data are business data recorded by a banking customer system. The process involves determining the stable values ​​of the test and training sample sets. These stable values ​​characterize the degree of variation in the values ​​of the corresponding data sets. Based on the distance between the stable values ​​of the test and training sample sets, at least a portion of the training data is selected for training the business data model. This approach achieves two advantages: firstly, by using the stable values ​​of the test and training sample sets to determine the degree of variation in the values ​​of the data, the distribution of data in each sample set can be accurately determined, thereby improving the balance among the data in the sample sets; secondly, by using the selected training data to train the data model in the banking customer system, the model can be trained using training data with improved sample balance, thus enhancing the stability of the trained data model. Attached Figure Description

[0057] Figure 1 This is an application environment diagram illustrating a training data filtering method according to an exemplary embodiment;

[0058] Figure 2 This is a flowchart illustrating a method for filtering training data according to an exemplary embodiment;

[0059] Figure 3 This is a schematic diagram illustrating a process for obtaining a sample set according to an exemplary embodiment;

[0060] Figure 4 This is a schematic diagram illustrating a process for determining the data stability value of a sample set according to an exemplary embodiment;

[0061] Figure 5 This is a schematic diagram illustrating a process for filtering at least a portion of training data according to an exemplary embodiment;

[0062] Figure 6 This is a schematic diagram illustrating another process for filtering at least a portion of the training data, according to an exemplary embodiment.

[0063] Figure 7 This is a flowchart illustrating a risk assessment method according to an exemplary embodiment;

[0064] Figure 8 This is a schematic diagram illustrating another process for filtering training data according to an exemplary embodiment;

[0065] Figure 9 This is a block diagram illustrating a training data filtering device according to an exemplary embodiment;

[0066] Figure 10 This is a block diagram of an electronic device for filtering training data according to an exemplary embodiment;

[0067] Figure 11 This is a block diagram of a computer-readable storage medium illustrating a screening device for training data according to an exemplary embodiment;

[0068] Figure 12 This is a block diagram of a computer program product for a training data filtering device according to an exemplary embodiment. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0070] The terms "first," "second," etc., used in this application are used to distinguish different objects, not to describe a specific order. Furthermore, although the terms "first," "second," etc., are used repeatedly to describe various operations (or various thresholds, or various applications, or various instructions, or various components), these operations (or thresholds, applications, instructions, or components) should not be limited by these terms. These terms are only used to distinguish one operation (or threshold, application, instruction, or component) from another operation (or threshold, application, instruction, or component). For example, a first type of statistical data can be called a second type of statistical data, and a second type of statistical data can be called a first type of statistical data, without departing from the scope of this application. Both the first type of statistical data and the second type of statistical data are data obtained after statistical analysis of corresponding data sets; they are simply data obtained from different statistical analyses.

[0071] The training data filtering method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with data storage system 103 via a network, and data storage system 103 is communicatively connected to server 104. Data storage system 103 can store data that server 104 needs to process. Data storage system 103 can be integrated onto server 104, or it can be placed on the cloud or other network servers.

[0072] In some embodiments, the terminal 102 is equipped with a banking customer system. Customers generate a large amount of business data through various financial activities performed on the terminal 102, and then the data storage system 103 stores this business data.

[0073] In some embodiments, terminal 102 (such as a mobile terminal or a fixed terminal) can be implemented in various forms. Terminal 102 can be a mobile terminal, including mobile phones, smartphones, laptops, portable handheld devices, personal digital assistants (PDAs), tablet computers (PADs), etc., equipped with a banking customer system and capable of generating large amounts of business data through various customer financial activities. Terminal 102 can also be a fixed terminal, including automated teller machines (ATMs), access control machines, digital TVs, desktop computers, fixed-line computers, etc., equipped with a banking customer system and capable of generating large amounts of business data through various customer financial activities. Hereinafter, it is assumed that terminal 102 is a fixed terminal. However, those skilled in the art will understand that, if there are operations or elements specifically designed for mobile purposes, the construction according to the embodiments disclosed in this application can also be applied to mobile-type terminal 102.

[0074] In some embodiments, reference Figure 1 Server 104 acquires a test sample set and a training sample set. The test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data. Both the training data and the test data are business data recorded by the banking customer system. Then, server 104 determines the stable data values ​​of the test sample set and the training sample set. The stable data values ​​are used to characterize the degree of change in the values ​​of each data point in the corresponding data set. Finally, based on the difference distance between the stable data values ​​of the test sample set and the stable data values ​​of the training sample set, server 104 selects at least a portion of the training data to train the business data model.

[0075] In some embodiments, the data processing component running on server 104 may load any of the various additional server applications and / or middleware applications being executed, such as HTTP (Hypertext Transfer Protocol), FTP (File Transfer Protocol), CGI (Common Gateway Interface), RDBMS (Relational Database Management System), etc.

[0076] In some embodiments, server 104 may be implemented using a standalone server or a server cluster consisting of multiple servers. Server 104 may be adapted to run one or more application services or software components that provide the terminal 102 described in the foregoing disclosure.

[0077] In one embodiment, such as Figure 2 As shown, a method for filtering training data is provided, which can be applied to... Figure 1 Taking server 104 as an example, the method includes the following steps:

[0078] Step S11: Obtain the test sample set and the training sample set.

[0079] In one embodiment, the test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data. The training data and test data are business data recorded by the banking customer system.

[0080] In some embodiments, a banking customer system may include a data repository for storing large amounts of business data. In response to initiating a screening process for training data, the server retrieves test data and training data from the data repository corresponding to the banking customer system to obtain a test sample set and a training sample set.

[0081] In some embodiments, at least a portion of the training data in the training sample set is selected for training a business data model in the banking industry, and test data is used to test the stability of the business data model.

[0082] In some embodiments, the dimensions included in the training data and test data can be various business data such as "total assets", "monthly deposits", "monthly outflows", "fee income", "monthly transaction count", etc., recorded by the banking customer system.

[0083] Step S12: Determine the stable values ​​of the test sample set and the stable values ​​of the training sample set.

[0084] In some embodiments, the server determines the data stability value of the test sample set based on preset calculation rules and the values ​​of test data in various dimensions in the test sample set; and the server determines the data stability value of the training sample set based on preset calculation rules and the values ​​of training data in various dimensions in the training sample set.

[0085] In one embodiment, the data stability value is used to characterize the degree of change in the values ​​of the data in the corresponding data set.

[0086] For example, the greater the variation in the values ​​of the data in the test sample set and / or training sample set, the smaller the stable value of the corresponding data in the test sample set and / or training sample set; conversely, the smaller the variation in the values ​​of the data in the test sample set and / or training sample set, the larger the stable value of the corresponding data in the test sample set and / or training sample set.

[0087] Step S13: Based on the difference distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set, at least a portion of the training data is selected for training the business data model.

[0088] In one embodiment, the server can determine the degree of balance of the numerical distribution of the data included in the test sample set and the training sample set based on the difference distance between the stable values ​​of the data in the test sample set and the stable values ​​of the data in the training sample set. The server then selects at least a portion of the training data from the training sample set for training the business data model based on the degree of balance of the numerical distribution of the data in the two sample sets.

[0089] In the aforementioned method for selecting training data, firstly, a test sample set and a training sample set are obtained. The test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data. Both the training and test data are business data recorded by the banking customer system. The stable values ​​of the test sample set and the training sample set are determined. These stable values ​​characterize the degree of variation among the values ​​of each data point in the corresponding data set. Based on the distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set, at least a portion of the training data is selected for training the business data model. In this way, on the one hand, by using the stable values ​​of the test and training sample sets to determine the degree of variation among the data points, the distribution of data in each sample set can be accurately determined, thereby improving the balance among the data points in the sample sets. On the other hand, using the selected training data to train the data model in the banking customer system allows for the use of training data with improved sample balance to train the model, thus enhancing the stability of the trained data model.

[0090] Those skilled in the art will understand that the methods disclosed in the above-described specific embodiments can be implemented in more concrete ways. For example, the implementation of the training data screening process described above is merely an illustrative description.

[0091] For example, the process by which the server determines the stable values ​​of the test sample set and the stable values ​​of the training sample set; or the process by which the server filters at least a portion of the training data based on the difference distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set, etc., is merely one way of aggregating data. In actual implementation, there may be other ways of partitioning the data. For example, the stable values ​​of the test sample set and the stable values ​​of the training sample set determined by the server may be combined or aggregated into another system, or some features may be ignored or not executed.

[0092] In one embodiment, the training data of the training sample set includes positive and negative sample data with the same multiple dimensions as the test data.

[0093] Both positive and negative sample data are training data, but they differ in the magnitude of the sample data. The values ​​of positive sample data tend to be more general, while the values ​​of negative sample data tend to be more extreme. Both types of sample data have the same multiple dimensions as the test data.

[0094] In one embodiment, the difference in data distribution between positive and negative sample data is negatively correlated with the stability of the business data model obtained through subsequent training; that is, the magnitude of the difference in the stable values ​​between positive and negative sample data is negatively correlated with the stability of the business data model.

[0095] In some embodiments, the greater the difference in the distribution of positive and negative sample data, the lower the stability of the business data model obtained through subsequent training; conversely, the smaller the difference in the distribution of positive and negative sample data, the higher the stability of the business data model obtained through subsequent training.

[0096] In one embodiment, the test data and training data of various dimensions include at least the dimensions of total assets, monthly deposits, monthly outflows, business revenue, and monthly transaction counts.

[0097] As an example, the dimensions of the positive and negative sample data in the training sample set include "Total Assets" and "Monthly Deposits". Specifically, the values ​​for "Total Assets" in the positive sample data are 2000, 3200, 3600, and 3800, respectively, while the values ​​for "Total Assets" in the negative sample data are 10, 30, 40000, and 68000, respectively. Similarly, the values ​​for "Monthly Deposits" in the positive sample data are 200, 320, 360, and 380, respectively, while the values ​​for "Monthly Deposits" in the negative sample data are 5, 20, 40000, and 58000, respectively.

[0098] In one exemplary embodiment, see Figure 3 , Figure 3 This is a schematic flowchart illustrating an embodiment of obtaining a sample set in this application. In step S11, the process of the server obtaining the test sample set and the training sample set can be implemented in the following way:

[0099] Step S111: Obtain multiple positive sample data and multiple negative sample data from a business data repository corresponding to the banking customer system, and combine them to obtain a training sample set.

[0100] In one embodiment, the business data repository of the banking customer system stores a large amount of negative sample data and a small amount of positive sample data, that is, the number of negative sample data stored in the business data repository is much greater than the number of positive sample data.

[0101] In one embodiment, the server extracts all stored positive sample data from the business data repository to obtain multiple positive sample data.

[0102] In one embodiment, the server extracts a portion of the stored negative sample data from the business data repository using a preset sampling method to obtain multiple negative sample data.

[0103] In one embodiment, the server combines the multiple positive sample data and the multiple negative sample data together to obtain a training sample set.

[0104] The preset sampling method can be a preset random sampling method, stratified sampling method, or probability sampling method, etc., and no specific limitation is made here.

[0105] In some embodiments, the number of positive sample data and the number of negative sample data acquired by the server may be the same or different. For example, the number of positive sample data acquired by the server may be 100, and the number of negative sample data acquired by the server may be 100, 120, 150, etc.

[0106] Step S112: Obtain multiple test data from the business data repository to form a test sample set.

[0107] In one embodiment, the server extracts a portion of the stored business data from the business data repository using a preset sampling method, and uses this as multiple test data. Then, the server combines these multiple test data to obtain a test sample set.

[0108] In one embodiment, the number of test data in the test sample set acquired by the server and the number of each training data in the training sample set may be the same or different. For example, the number of test data acquired by the server may be 200, while the number of training data acquired by the server may be 100, 150, 200, 300, etc.

[0109] In one exemplary embodiment, see Figure 4 , Figure 4 This is a flowchart illustrating an embodiment of determining the data stability value of a sample set in this application. In step S12, the process by which the server determines the data stability value of the test sample set and the data stability value of the training sample set can be implemented in the following way:

[0110] Step S121: Based on the first type of statistical data corresponding to the test data and training data of various dimensions, determine the single-type stable value of the test data of each dimension and the single-type stable value of the training data of each dimension.

[0111] In one embodiment, the first type of statistical data includes the standard deviation and median of test data for various dimensions, and the standard deviation and median of training data for various dimensions.

[0112] In one embodiment, a single-type stable value is used to characterize the degree of variation in the values ​​of individual data in a single dimension of data.

[0113] As an example, the greater the variation in the values ​​of the "monthly transaction count" type test data in the test sample set, the smaller the single-type stable value of the "monthly transaction count" type in the corresponding test sample set; conversely, the smaller the variation in the values ​​of the "monthly transaction count" type test data in the test sample set, the larger the single-type stable value of the "monthly transaction count" type in the corresponding test sample set.

[0114] In one embodiment, the server calculates the categorical stability value of the test data for each dimension based on the standard deviation and median of the test data for each dimension; and the server calculates the categorical stability value of the training data for each dimension based on the standard deviation and median of the training data for each dimension.

[0115] In some embodiments, the server can obtain the single-type stable value of the test data for each dimension based on the quotient of the standard deviation and the median of the test data for each dimension; and the server can obtain the single-type stable value of the training data for each dimension based on the quotient of the standard deviation and the median of the training data for each dimension.

[0116] Step S122: Determine the second type of statistical data corresponding to multiple single-type stable values ​​of the test data, and the third type of statistical data corresponding to multiple single-type stable values ​​of the training data.

[0117] In one embodiment, the second type of statistical data includes the standard deviation and mean of various single-type stable values ​​in the test data, and the third type of statistical data includes the standard deviation and mean of various single-type stable values ​​in the training data.

[0118] Step S123: Based on the second type of statistical data and the third type of statistical data, determine the stable data values ​​of the test sample set and the stable data values ​​of the training sample set, respectively.

[0119] In one embodiment, the server calculates the data stability value of the test sample set based on the standard deviation and mean of various single-type stable values ​​in the test data; and the server calculates the data stability value of the training sample set based on the standard deviation and mean of various single-type stable values ​​in the training data.

[0120] In some embodiments, the server may determine the data stability value of the test sample set based on the quotient of the standard deviation and the mean of various single-type stable values ​​in the training data; and the server may determine the data stability value of the training sample set based on the quotient of the standard deviation and the mean of various single-type stable values ​​in the training data.

[0121] In one exemplary embodiment, see Figure 5 , Figure 5 This is a schematic flowchart illustrating an embodiment of screening at least a portion of training data in this application. In step S13, the server screens at least a portion of the training data for training the business data model based on the difference distance between the data stability values ​​of the test sample set and the data stability values ​​of the training sample set. This process can be implemented in the following ways:

[0122] Step S131: Use the quotient of the data stability values ​​of the test sample set and the data stability values ​​of the training sample set to obtain the difference distance.

[0123] In one embodiment, the server first calculates the quotient of the data stability value of the test sample set and the data stability value of the training sample set. Then, the server calculates the absolute value of the difference between the quotient and a preset constant to determine the difference distance.

[0124] The server preset constant can be any constant, such as 0.5, 1, 1.5, etc., without any specific limitation.

[0125] In one embodiment, the difference distance is used to characterize the degree of dispersion between two stable data values.

[0126] As an example, the greater the difference between the stable values ​​of the test sample set and the stable values ​​of the training sample set, the greater the dispersion between them, and consequently, the weaker the balance of data distribution between the test data in the test sample set and the training data in the training sample set. Conversely, the smaller the difference between the stable values ​​of the test sample set and the stable values ​​of the training sample set, the smaller the dispersion between them, and consequently, the stronger the balance of data distribution between them.

[0127] Step S132: Based on the relationship between the difference distance and the preset distance limit, at least a portion of the training data is selected for training the business data model.

[0128] In one embodiment, if the difference distance is less than or equal to a preset distance limit, the server selects all training data in the training sample set for training the business data model.

[0129] In some embodiments, if the difference distance is less than or equal to a preset distance limit, it indicates that the data distribution balance between the test data in the test sample set and the training data in the training sample set is acceptable to the server. Therefore, the server filters out all positive and negative sample data from the training sample set for use in training the business data model.

[0130] In one embodiment, the difference in data distribution between the test data and the training data used to train and test the business data model is negatively correlated with the stability of the business data model; that is, the magnitude of the difference in the stable values ​​of the test data and the training data is negatively correlated with the stability of the business data model.

[0131] In some embodiments, the greater the difference in data distribution between the test data and the training data, the lower the stability of the trained business data model; conversely, the smaller the difference in data distribution between the test data and the training data, the higher the stability of the trained business data model.

[0132] In one exemplary embodiment, see Figure 6 , Figure 6 This is a flowchart illustrating another embodiment of the process for selecting at least a portion of the training data in this application. In step S132, the server selects at least a portion of the training data for training the business data model based on the relationship between the difference distance and a preset distance limit. This process can be implemented in the following ways:

[0133] Step a1: If the difference distance is greater than the distance limit, retrieve multiple new negative sample data from the business data repository corresponding to the banking customer system.

[0134] In one embodiment, if the difference distance is greater than a preset distance limit, the server will filter out the positive sample data in the training sample set, and then the server will retrieve multiple new negative sample data from the business data repository.

[0135] In some embodiments, if the difference distance is greater than a preset distance limit, it indicates that the data distribution balance between the test data in the test sample set and the training data in the training sample set is unacceptable to the server. Therefore, the server separately filters out all positive sample data in the training sample set to remove unusable negative sample data.

[0136] Step a2: Combine the new negative sample data with the positive sample data from the training sample set to obtain a new training sample set.

[0137] In one embodiment, the server combines multiple new negative sample data and separately selected positive sample data to obtain a new training sample set.

[0138] Step a3: Based on the new training sample set and test sample set, repeat the steps to determine the data stability value of the test sample set and the data stability value of the training sample set until the difference distance is less than or equal to the distance limit.

[0139] In one embodiment, the server re-executes the steps of setting the data stability value of the test sample set and the data stability value of the training sample set based on the newly acquired training sample set and the acquired test sample set, until the difference between the two is less than or equal to the distance limit.

[0140] In one embodiment, such as Figure 7 As shown, a risk assessment method is provided. Taking the application of this method to a banking customer system as described in the above embodiment as an example, the method includes the following steps:

[0141] Step S21: Obtain the client's investment data.

[0142] In one embodiment, the server retrieves multiple investment data of a customer from a business data repository in a banking customer system.

[0143] In some embodiments, multiple investment data for a customer can be business data such as the customer's "total assets", "monthly deposits", "monthly outflows", "fee income", "monthly transaction count", etc., without specific limitations.

[0144] Step S22: Input the investment data into a trained risk assessment model to calculate the risk and obtain the risk score for the customer's investment based on the investment data.

[0145] In one embodiment, the risk assessment model is constructed using test data from the test sample set and training data from the training sample set.

[0146] In some embodiments, the risk assessment model can be a neural network model (such as CNN, VGG, ResNet, etc.) or a semantic segmentation model (such as Transformer, Attention-based RNN, LSTM, etc.).

[0147] In one embodiment, the training data in the training sample set is obtained by filtering the training data using the method described in the above embodiments.

[0148] In one embodiment, the server inputs the customer's investment data into a risk assessment model to calculate the risk. After the risk assessment model completes the calculation, it outputs the risk score for the customer's investment based on the investment data.

[0149] In some embodiments, the risk assessment model can be a machine learning model. The machine learning model learns from the cumulative feedback values ​​obtained by taking different input parameters (i.e., the numerical values ​​of the training data) from a large amount of training data, in order to obtain the optimal range of feedback values ​​(i.e., the optimal range of feedback data for the model's learned parameters) for different action strategies (i.e., the numerical values ​​of the test data) under each initial input parameter (i.e., the numerical values ​​of the input training data).

[0150] As an example, the machine learning model sets the initial input parameter values ​​for each input parameter data as sp = (x1n, x2n, ..., xnn), where sp is the input parameter dimension, xnn is the initial input parameter data for one input parameter data, xi ∈ [li, hi] is the initial input parameter data for the i-th input parameter data, and a total of n ∈ N initial input parameter data need to be learned. The machine learning model uses the cumulative feedback value of different input parameter data as rt = score(st+1) - score(st), where the cumulative feedback value is set to a step size of 1 for each parameter tuning, i.e., a = (±1, ±1, ..., ±1), and score is the range of comprehensive optimal feedback values ​​determined by the engineer under the current input parameter data settings.

[0151] Step S23: Determine the customer's risk assessment result based on the risk score.

[0152] In one embodiment, the server compares the risk score output by the risk assessment model with a pre-set risk comparison table to obtain the customer's risk assessment result.

[0153] The aforementioned risk assessment method involves several steps. First, the server acquires the customer's investment data. Then, the server inputs this data into a trained risk assessment model to calculate the risk score for the customer's investments based on the data. Finally, the server determines the customer's risk assessment result based on the risk score. This approach utilizes at least a portion of the selected training data to train the risk assessment model within the banking customer system. This allows the model to be trained using training data with improved sample balance, thereby enhancing the stability of the trained risk assessment model.

[0154] To more clearly illustrate the training data filtering method provided in the embodiments of this disclosure, the following specific embodiment will be used to describe the training data filtering method in detail. In an exemplary embodiment, as follows... Figure 8 As shown, Figure 8 This is a flowchart illustrating a training data filtering method according to another exemplary embodiment. The training data filtering method is used in server 104 and specifically includes the following:

[0155] Step S31: Determine the number of test samples and training samples.

[0156] In the financial industry, the number of real samples available for machine learning models is typically massive. However, the number of positive samples used in the training samples is usually limited, while negative samples are abundant. Therefore, it is necessary to first determine the number of positive samples in the training samples, then the number of negative samples, and finally the number of test samples. The ratio of positive to negative samples can be specific, and the number of test samples should be the same as the total number of training samples.

[0157] In one embodiment, let the number of real test samples be M, the number of positive samples in the training samples be K, and the number of negative samples in the training samples be L, where M = 2K = 2L.

[0158] Step S32: Sample negative samples from the training samples.

[0159] In the financial industry modeling process, the number of negative samples L in the sample library is usually much greater than the number of negative samples K. In order to ensure the training speed of the machine learning model and the balance between positive and negative samples, it is necessary to sample the negative samples L, while the test samples M and positive samples K do not need to be sampled.

[0160] Step S33: Calculate the stability of each type of single feature in the test samples and training samples respectively.

[0161] Both the test samples and training samples consist of the same multiple data types, with each data type including one feature. For example, both the test samples and training samples consist of five data types: "total financial assets," "average monthly deposit balance," "monthly outflow of assets," "fee income," and "monthly transaction count," with each data type representing a data feature.

[0162] In this context, for the same feature in a sample, if the degree of variation between the training data and the actual data is different, it will significantly affect the model's performance. Single-feature stability mainly considers the variation of a single feature in the sample. Single-feature stability is defined as the ratio of the variation of the feature sequence values ​​to the average level. The variation is measured using the standard deviation. Considering the significant differences in data among different customers in the financial industry, the median is used to measure the average level of the data to exclude the influence of extreme values.

[0163] The stability of various types of single features in test samples and training samples can be expressed by the following formula:

[0164]

[0165] Where Wn is the stability of the nth single feature, σn is the standard deviation of the value sequence of the nth single feature, and M 0.5 This is the median of the single feature value sequence. The magnitude of Wn reflects the stability of the single feature; a larger Wn indicates a greater degree of variation in the single feature, and a smaller Wn indicates a smaller degree of variation in the single feature.

[0166] Step S34: Calculate the overall stability of single feature stability of various types in the test samples and the overall stability of single feature stability of various types in the training samples.

[0167] The overall stability, which can be calculated for various types of single-feature stability, can be expressed by the following formula:

[0168]

[0169] Where σ is the standard deviation of single-feature stability for each type, and mean is the mean of single-feature stability for each type.

[0170] The overall stability of the test sample M can be determined by W. M This indicates that the overall stability of K+L for both positive and negative samples can be determined by W. K+L express.

[0171] Step S35: Compare the difference between the overall stability of the test sample M and the overall stability of the positive and negative samples to select suitable training set samples for training the model.

[0172] The difference between the overall stability of the test sample M and the overall stability of the positive and negative samples K+L in the training samples can be expressed by the following formula:

[0173]

[0174] Among them, W Δ The smaller the value of W, the smaller the difference between the training samples and the samples to be predicted, and the better the model performs in practical applications. For example, in practical applications, W... Δ The judgment threshold can be 0.2, if W Δ If the value is greater than 0.2, then negative samples L can be repeatedly sampled from the sample library, and the difference W between the overall stability of the test sample M and the overall stability of the positive and negative samples K+L in the training samples can be compared. Δ Until W Δ Once the value falls below 0.2, using the positive and negative samples K+L from the training samples at this point for modeling training can ensure good practical application results.

[0175] As an example, if a sample set K+l consists of a negative sample l and a positive sample K, and its feature "total financial assets" has a value sequence of 100, 200, 300, 1000, 10000, then the standard deviation of the feature "total financial assets" is 4307, the median is 300, and the stability coefficient of the feature "total financial assets" is 14.35. If sample l has five features, in addition to "total financial assets," namely "average monthly deposit balance," "monthly outflow of assets," "intermediate business income," and "monthly transaction frequency," and the stability coefficients of these features are calculated respectively, and their stability coefficients, arranged from smallest to largest, are 14.35, 15, 25, 35, and 50, then the stability coefficient of sample K+l is 0.5371. If the stability coefficient of the sample M to be predicted is 0.55, the difference W between the stability coefficients of sample K+l and sample M is... Δ The value is 0.023, if W is used. Δ If the threshold for judgment is 0.2, then the negative sample K+l is retained and sampled again, and used as the training sample for the model. Otherwise, a new negative sample K+l is resampled, and W is calculated. Δ This continues until the set threshold is met. At this point, using the obtained samples for model training yields relatively good results.

[0176] It should be understood that, although Figures 2-8 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 2-8At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0177] It is understood that the same / similar parts between the various embodiments of the methods described above in this specification can be referred to each other. Each embodiment focuses on the differences from other embodiments, and relevant parts can be referred to the description of other method embodiments.

[0178] Figure 9 This is a block diagram illustrating a training data filtering device according to an exemplary embodiment. (Refer to...) Figure 9 The screening device 10 includes a sample acquisition unit 11, a data processing unit 12, and a data screening unit 13.

[0179] The sample acquisition unit 11 is configured to acquire a test sample set and a training sample set; the test sample set includes test data of multiple dimensions, and the training sample set includes training data of the same dimensions as the test data; the training data and the test data are business data recorded by the banking customer system.

[0180] The data processing unit 12 is configured to determine the data stability value of the test sample set and the data stability value of the training sample set; the data stability value is used to characterize the degree of change of the values ​​of each data in the corresponding data set.

[0181] The data filtering unit 13 is configured to perform a distance-based filtering of at least a portion of the training data for training the business data model, based on the difference between the stable values ​​of the data in the test sample set and the stable values ​​of the data in the training sample set.

[0182] In an exemplary embodiment, the sample acquisition unit 11 is further configured to acquire a plurality of positive sample data and a plurality of negative sample data from a business data repository corresponding to the banking customer system, so as to obtain the training sample set.

[0183] Multiple test data sets are obtained from the business data repository to form the test sample set, wherein the training data includes positive and negative sample data of the same dimension as the test data.

[0184] In one exemplary embodiment, the sample acquisition unit 11 is further configured to extract all positive sample data stored in the business data repository to obtain a plurality of positive sample data; and

[0185] A portion of negative sample data is extracted from the business data repository to obtain multiple negative sample data, wherein the number of negative sample data stored in the business data repository is much greater than the number of positive sample data.

[0186] In an exemplary embodiment, the data processing unit 12 is further configured to perform first type statistical data based on various dimensions of the test data and the training data to determine the single-type stable value of the test data for each dimension and the single-type stable value of the training data for each dimension; the single-type stable value is used to characterize the degree of change of values ​​among the data in a certain dimension.

[0187] Determine the second type of statistical data corresponding to multiple single-type stable values ​​of the test data, and the third type of statistical data corresponding to multiple single-type stable values ​​of the training data;

[0188] Based on the second type of statistical data and the third type of statistical data, the data stability value of the test sample set and the data stability value of the training sample set are determined respectively.

[0189] In one exemplary embodiment, the data processing unit 12 is further configured to perform calculations based on the standard deviation and median of the test data for each dimension, to obtain single-type stable values ​​for the test data corresponding to each dimension; and

[0190] Based on the standard deviation and median of the training data for each dimension, a single-type stable value for the training data corresponding to each dimension is calculated, wherein the first type of statistical data includes the standard deviation and median.

[0191] In one exemplary embodiment, the data processing unit 12 is further configured to calculate the data stability value of the test sample set based on the standard deviation and mean of various single-type stable values ​​in the test data; and

[0192] Based on the standard deviation and mean of various single-type stable values ​​in the training data, the data stability value of the training sample set is calculated, wherein the second type of statistical data and the third type of statistical data both include the standard deviation and the mean.

[0193] In one exemplary embodiment, the data filtering unit 13 is further configured to perform a quotient of the data stability values ​​of the test sample set and the data stability values ​​of the training sample set to obtain the difference distance; the difference distance is used to characterize the degree of dispersion between the two data stability values.

[0194] Based on the relationship between the difference distance and the preset distance limit, at least a portion of the training data is selected for training the business data model.

[0195] In an exemplary embodiment, the data filtering unit 13 is further configured to perform calculation of the absolute value of the difference between the quotient and a preset constant, and determine the difference distance;

[0196] The step of selecting at least a portion of the training data for training the business data model based on the relationship between the difference distance and the preset distance limit includes:

[0197] If the difference distance is less than or equal to the distance limit, all training data in the training sample set are selected for training the business data model.

[0198] In an exemplary embodiment, the data filtering unit 13 is further configured to perform the following: if the difference distance is greater than the distance limit, re-acquire multiple new negative sample data from the business data repository corresponding to the banking customer system;

[0199] By combining the new negative sample data with the positive sample data in the training sample set, a new training sample set is obtained.

[0200] Based on the new training sample set and the test sample set, the steps of determining the data stability value of the test sample set and the data stability value of the training sample set are repeated until the difference distance is less than or equal to the distance limit.

[0201] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0202] Figure 10 This is a block diagram illustrating an electronic device 20 for filtering training data according to an exemplary embodiment. For example, electronic device 20 can be a server. (Refer to...) Figure 10 The electronic device 20 includes a processing component 21, which further includes one or more processors, and memory resources represented by memory 22 for storing executable instructions, such as application programs, that can be executed by the processing component 21. The application programs stored in memory 22 may include one or more modules, each corresponding to a set of executable instructions. Furthermore, the processing component 21 is configured to execute the executable instructions to perform the methods described above.

[0203] In one embodiment, the electronic device 20 is a server, and the computing system within the server can run one or more operating systems, including any of the operating systems discussed above and any commercially available server operating system. The server can also run any of a variety of additional server applications and / or middleware applications, including HTTP (Hypertext Transfer Protocol) servers, FTP (File Transfer Protocol) servers, CGI (Common Gateway Interface) servers, database servers, etc. Exemplary database servers include, but are not limited to, commercially available database servers from companies such as IBM.

[0204] In one embodiment, the processing component 21 typically controls the overall operation of the electronic device 20, such as operations associated with display, data processing, data communication, and recording operations. The processing component 21 may include one or more processors to execute instructions to perform all or part of the steps of the methods described above. Furthermore, the processing component 21 may include one or more modules to facilitate interaction between the processing component 21 and other components. For example, the processing component 21 may include a multimedia module to facilitate control of the interaction between the user terminal and the processing component 21 using multimedia components.

[0205] In one embodiment, the processor in processing component 21 may also be referred to as a CPU (Central Processing Unit). The processor may be an electronic chip with signal processing capabilities. The processor may also 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. A general-purpose processor may be a microprocessor or any conventional processor. Furthermore, the processor may be implemented using integrated circuit chips.

[0206] In one embodiment, memory 22 is configured to store various types of data to support the operation of electronic device 20. Examples of such data include instructions for any application or method operating on electronic device 20, acquired data, messages, images, videos, etc. Memory 22 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, optical disk, or graphene memory.

[0207] In one embodiment, the memory 22 can be a memory module, TF card, etc., and can store all the information in the electronic device 20, including the input raw data, computer programs, intermediate running results, and final running results. It stores and retrieves information according to the location specified by the processor. With the memory 22 in this embodiment, the electronic device 20 has a memory function and can ensure normal operation. In one embodiment of the electronic device 20, the memory 22 can be classified according to its purpose as main memory (RAM) and auxiliary memory (external memory), or it can be classified as external memory and internal memory. External memory is usually magnetic media or optical discs, which can store information for a long time. RAM refers to the storage components on the motherboard, used to store currently executing data and programs, but it is only used for temporary storage of programs and data; the data will be lost when the power is turned off or disconnected.

[0208] Electronic device 20 may further include: a power supply component 23 configured to perform power management of electronic device 20, a wired or wireless network interface 24 configured to connect electronic device 20 to a network, and an input / output (I / O) interface 25. Electronic device 20 may operate on an operating system stored in memory 22, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or similar.

[0209] In one embodiment, power supply component 23 provides power to various components of electronic device 20. Power supply component 23 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 20.

[0210] In one embodiment, the wired or wireless network interface 24 is configured to facilitate wired or wireless communication between the electronic device 20 and other devices. The electronic device 20 can access wireless networks based on communication standards, such as WiFi, carrier networks (such as 2G, 3G, 4G, or 5G), or combinations thereof.

[0211] In one exemplary embodiment, the wired or wireless network interface 24 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In another exemplary embodiment, the wired or wireless network interface 24 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0212] In one embodiment, the input / output (I / O) interface 25 provides an interface between the processing component 21 and peripheral interface modules, such as a keyboard, click wheel, buttons, etc. These buttons may include, but are not limited to, a home button, volume buttons, a power button, and a lock button.

[0213] Figure 11 This is a block diagram illustrating a computer-readable storage medium 30 for filtering training data according to an exemplary embodiment. The computer-readable storage medium 30 stores program data 31 capable of implementing the methods described above.

[0214] If the integrated units of the various functional units in the various embodiments of this application are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium 30. Based on this understanding, the technical solution of this application, 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. The computer-readable storage medium 30 includes a program data 31, which includes several instructions to cause a computer device (which may be a personal computer, system server, or network device, etc.), an electronic device (e.g., MP3, MP4, etc., or a mobile phone, tablet computer, wearable device, etc., or a desktop computer, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of this application.

[0215] Figure 12 This is a block diagram illustrating a computer program product 40 for filtering training data according to an exemplary embodiment. The computer program product 40 includes program instructions 41, and the program data can be executed by the processor of the electronic device 20 to perform the above-described method.

[0216] Those skilled in the art will understand that embodiments of this application can be provided as a method for biometric identification, a training data screening device 10, an electronic device 20, a computer-readable storage medium 30, or a computer program product 40. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product 40 embodied on one or more computer program instructions 41 (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0217] This application is described with reference to flowchart illustrations and / or block diagrams of training data filtering methods, training data filtering devices 10, electronic devices 20, computer-readable storage media 30, or computer program products 40 according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by the computer program product 40. These computer program products 40 can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that program instructions 41, executable by the processor of the computer or other programmable data processing device, generate instructions for implementing the training data filtering method in the flowchart illustration. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0218] These computer program products 40 may also be stored in a computer-readable storage medium capable of directing a computer or other programmable data processing device to function in a particular manner, such that program instructions 41 stored in the computer program product 40 produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0219] These program instructions 41 may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing the program instructions 41 that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0220] It should be noted that the various methods, apparatuses, electronic devices, computer-readable storage media, computer program products, etc. described above may also include other implementation methods according to the description of the method embodiments. For specific implementation methods, please refer to the description of the relevant method embodiments, which will not be elaborated here.

[0221] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0222] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for selecting training data, characterized in that, The method includes: Obtain a test sample set and a training sample set; the test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data; the training data and the test data are business data recorded by the banking customer system. Determine the data stability value of the test sample set and the data stability value of the training sample set; the data stability value is used to characterize the degree of change of the values ​​of each data in the corresponding data set. Based on the difference distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set, at least a portion of the training data is selected for training the business data model. The step of selecting at least a portion of the training data for training the business data model based on the difference distance between the stable values ​​of the test sample set and the stable values ​​of the training sample set includes: The difference distance is obtained by using the quotient of the stable data values ​​of the test sample set and the stable data values ​​of the training sample set; the difference distance is used to characterize the degree of dispersion between the two stable data values; the magnitude of the difference distance is negatively correlated with the stability of the business data model. Based on the relationship between the difference distance and the preset distance limit, at least a portion of the training data is selected for training the business data model; The step of obtaining the difference distance by using the quotient of the data stability values ​​of the test sample set and the data stability values ​​of the training sample set includes: Calculate the absolute value of the difference between the quotient and the preset constant to determine the difference distance; The step of selecting at least a portion of the training data for training the business data model based on the relationship between the difference distance and the preset distance limit includes: If the difference distance is less than or equal to the distance limit, all training data in the training sample set are selected for training the business data model.

2. The method according to claim 1, characterized in that, The training data includes positive and negative sample data with the same multiple dimensions as the test data. The process of obtaining the test sample set and the training sample set includes: Multiple positive sample data and multiple negative sample data are obtained from a business data repository corresponding to the banking customer system, and the training sample set is obtained by combining them. Multiple test data sets are obtained from the business data repository to form the test sample set.

3. The method according to claim 2, characterized in that, The number of negative sample data stored in the business data repository is greater than the number of positive sample data; Obtain multiple positive sample data and negative sample data from the business data repository, including: Extract all positive sample data stored in the business data repository to obtain multiple positive sample data sets; and A portion of negative sample data is extracted from the business data repository to obtain multiple negative sample data; wherein, the magnitude of the difference distance between the stable values ​​of the positive sample data and the negative sample data is negatively correlated with the stability of the business data model.

4. The method according to claim 1, characterized in that, Determining the data stability values ​​of the test sample set and the training sample set includes: Based on the first type of statistical data corresponding to the test data and training data of various dimensions, the single-type stable value of the test data and the single-type stable value of the training data of each dimension are determined; the single-type stable value is used to characterize the degree of change of the values ​​of each data in a dimension; the test data and training data of various same dimensions include at least the dimensions of total assets, monthly deposits, monthly outflows, business revenue, and monthly transaction frequency. Determine the second type of statistical data corresponding to multiple single-type stable values ​​of the test data, and the third type of statistical data corresponding to multiple single-type stable values ​​of the training data; Based on the second type of statistical data and the third type of statistical data, the data stability value of the test sample set and the data stability value of the training sample set are determined respectively.

5. The method according to claim 4, characterized in that, The first type of statistical data includes the standard deviation and the median; The determination of the single-type stable value of the test data and the single-type stable value of the training data for each dimension, based on the first type of statistical data corresponding to the test data and the training data for each dimension, includes: Based on the standard deviation and median of the test data for each dimension, calculate the unitype stable value of the test data for each dimension; and Based on the standard deviation and median of the training data for each dimension, the unitype stable value of the training data for each dimension is calculated.

6. The method according to claim 5, characterized in that, Both the second type of statistical data and the third type of statistical data include standard deviation and mean; The step of determining the data stability value of the test sample set and the data stability value of the training sample set based on the second type of statistical data and the third type of statistical data, respectively, includes: Based on the standard deviation and mean of various single-type stable values ​​in the test data, the data stability value of the test sample set is calculated; and Based on the standard deviation and mean of various single-type stable values ​​in the training data, the data stability value of the training sample set is calculated.

7. The method according to claim 1, characterized in that, The larger the difference distance, the lower the stability of the business data model; the smaller the difference distance, the higher the stability of the business data model. If the difference distance is greater than the distance limit, multiple new negative sample data are retrieved from the business data repository corresponding to the banking customer system; By combining the new negative sample data with the positive sample data in the training sample set, a new training sample set is obtained. Based on the new training sample set and the test sample set, the steps of determining the data stability value of the test sample set and the data stability value of the training sample set are repeated until the difference distance is less than or equal to the distance limit.

8. A risk assessment method, said method being applied to a banking customer system, characterized in that, The method includes: Obtain customer investment data; The investment data is input into a trained risk assessment model to calculate the risk, thereby obtaining the risk score of the customer's investment based on the investment data; the risk assessment model is constructed using test data in the test sample set and training data in the training sample set, wherein the training data in the training sample set is obtained by filtering the training data according to the filtering method of any one of claims 1-7. Based on the risk score, the risk assessment result for the customer is determined.

9. A training data filtering device, characterized in that, include: The sample acquisition unit is configured to acquire a test sample set and a training sample set; the test sample set includes test data with multiple dimensions, and the training sample set includes training data with the same multiple dimensions as the test data; the training data and the test data are business data recorded by the banking customer system. The data processing unit is configured to determine the data stability value of the test sample set and the data stability value of the training sample set; the data stability value is used to characterize the degree of change of the values ​​of each data in the corresponding data set. A data filtering unit is configured to perform a difference distance between the data stability values ​​of the test sample set and the data stability values ​​of the training sample set to filter at least a portion of the training data for training the business data model. The data filtering unit is configured to obtain the difference distance by using the quotient of the data stability values ​​of the test sample set and the data stability values ​​of the training sample set; the difference distance is used to characterize the degree of dispersion between the two data stability values; the magnitude of the difference distance is negatively correlated with the stability of the business data model. Based on the relationship between the difference distance and the preset distance limit, at least a portion of the training data is selected for training the business data model; The data filtering unit is configured to calculate the absolute value of the difference between the quotient and a preset constant, and determine the difference distance. When selecting at least a portion of the training data to train the business data model based on the relationship between the difference distance and the preset distance limit, if the difference distance is less than or equal to the distance limit, all training data in the training sample set are selected to train the business data model.

10. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to execute the executable instructions to implement the training data screening method as described in any one of claims 1 to 7 and / or the risk assessment method as described in claim 8.

11. A computer-readable storage medium comprising program data, characterized in that, When the program data is executed by the processor of the electronic device, the electronic device is able to perform the training data screening method as described in any one of claims 1 to 7 and / or the risk assessment method as described in claim 8.

12. A computer program product, the computer program product comprising program instructions, characterized in that, When the program instructions are executed by the processor of the electronic device, the electronic device is enabled to perform the training data screening method as described in any one of claims 1 to 7 and / or the risk assessment method as described in claim 8.

Images