Risk control decision-making method and system based on direct push type semi-supervised learning

Abstract

The invention relates to a risk control decision method and system based on direct push type semi-supervised learning, and the method comprises the following steps: collecting labeled user data, and collecting unlabeled user data; according to the purpose of the credit scoring model, multiple data attribute indexes are set, and different weights are given to the multiple data attribute indexes; acquiring multiple data attributes of the user data in the first data sample according to the multiple data attribute indexes, and generating a screening condition according to a set strategy; screening in the basic database, adjusting the number of screening results according to the weight of the data attribute index, and finally obtaining a second data sample; the first data sample is used as a sample with a class label, the second data sample is used as a sample without a class label, and the class label of the second data sample is calculated and predicted through a direct push type semi-supervised learning algorithm; by applying the method, the data processing efficiency can be effectively improved, the number and quality of the training sample data can be effectively improved, and the reliability of the credit scoring model is improved.

Description

Technical Field

[0001] The present invention relates to the field of smart banking technology, and more specifically, to a risk control decision-making method and system based on direct semi-supervised learning. Background Art

[0002] The credit scoring model is a model for defining personal financial rights that has emerged in recent years to ensure the financial security of banks and other financial institutions. This model refers to the use of a certain credit scoring model to obtain different levels of credit scores based on the customer's credit history information, and to determine the amount of money the customer can hold based on the customer's credit score, thereby ensuring the security of repayment and other services. As the role of loans and credit cards becomes increasingly prominent in modern society and companies, the development prospects of the credit scoring model are immeasurable.

[0003] At present, in the credit risk control process, the samples for training the credit scorecard model are mostly based on labeled data. However, the amount of such labeled data is very small in weight and it is also very difficult to obtain, which ultimately leads to great limitations in the credit scorecard model. There is a need for a risk control decision-making method and system based on direct semi-supervised learning that can reasonably solve the above defects. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a risk control decision-making method based on direct induction semi-supervised learning and a risk control decision-making system based on direct induction semi-supervised learning in response to the above-mentioned defects of the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problem is:

[0006] A risk control decision-making method based on transductive semi-supervised learning is constructed, wherein the method comprises the following steps:

[0007] Collect labeled user data as the first data sample, and collect unlabeled user data as the basic database;

[0008] According to the purpose of the credit scoring model, multiple data attribute indicators are set and different weights are assigned to the multiple data attribute indicators;

[0009] Acquire multiple data attributes of user data in the first data sample according to multiple data attribute indicators, and generate screening conditions according to the set strategy;

[0010] Screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally a second data sample is obtained;

[0011] The first data sample is taken as a sample with a class label, and the second data sample is taken as a sample without a class label. The class label of the second data sample is predicted by a direct inference semi-supervised learning algorithm.

[0012] The risk control decision-making method based on direct induction semi-supervised learning of the present invention further comprises the steps of:

[0013] The credit score card model is trained using the first data sample and the predicted second data sample as labeled samples.

[0014] The risk control decision-making method based on direct induction semi-supervised learning of the present invention, wherein the method of obtaining multiple data attributes of user data in the first data sample according to multiple data attribute indicators and generating screening conditions according to the set strategy adopts the following method:

[0015] According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data;

[0016] The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted;

[0017] Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a filtering condition.

[0018] The risk control decision-making method based on direct induction semi-supervised learning of the present invention, wherein the screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally the second data sample is obtained by adopting the method:

[0019] Traverse the data in the basic database according to the data attributes to obtain a data set containing the data;

[0020] Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes;

[0021] According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted;

[0022] The set of user data corresponding to the multiple data sets finally obtained is used as the second data sample.

[0023] The risk control decision-making method based on direct induction semi-supervised learning of the present invention, wherein the method of further adjusting the amount of data in the data set according to the ratio of the weights of multiple data attribute indicators is as follows:

[0024] Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark;

[0025] Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks;

[0026] Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

[0027] A risk control decision system based on direct induction semi-supervised learning, wherein the system comprises: a data collection unit, a data storage unit, a data preprocessing unit and a data processing unit;

[0028] The data collection unit is used to collect labeled user data as the first data sample, collect unlabeled user data as the basic database, and store them in the data storage unit;

[0029] The data preprocessing unit is used to set a plurality of data attribute indicators and assign different weights to the plurality of data attribute indicators according to the purpose of the credit scoring model; obtain a plurality of data attributes of the user data in the first data sample according to the plurality of data attribute indicators, and generate screening conditions according to the set strategy; perform screening in the basic database according to the screening conditions, and adjust the number of screening results according to the weights of the data attribute indicators, and finally obtain the second data sample;

[0030] The data processing unit is used to use the first data sample as a sample with a class label and the second data sample as a sample without a class label, and predict the class label of the second data sample through a direct inference semi-supervised learning algorithm.

[0031] The risk control decision-making system based on direct induction semi-supervised learning of the present invention, wherein the system further includes a model training unit;

[0032] The model training unit is used to train the credit score card model using the first data sample and the predicted second data sample as labeled samples.

[0033] The risk control decision-making system based on direct induction semi-supervised learning of the present invention, wherein the method of obtaining multiple data attributes of user data in the first data sample according to multiple data attribute indicators and generating screening conditions according to the set strategy adopts the following method:

[0034] According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data;

[0035] The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted;

[0036] Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a filtering condition.

[0037] The risk control decision-making system based on direct induction semi-supervised learning of the present invention, wherein the screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally the second data sample is obtained by adopting the method:

[0038] Traverse the data in the basic database according to the data attributes to obtain a data set containing the data;

[0039] Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes;

[0040] According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted;

[0041] The set of user data corresponding to the multiple data sets finally obtained is used as the second data sample.

[0042] The risk control decision-making system based on direct induction semi-supervised learning of the present invention, wherein the method for further adjusting the amount of data in the data set according to the ratio of the weights of multiple data attribute indicators is as follows:

[0043] Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark;

[0044] Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks;

[0045] Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

[0046] The beneficial effects of the present invention are as follows: by applying the method of the present application, guided by the purpose of the credit scoring model, a data screening mechanism that is more in line with the actual application scenario is added to pre-process the unlabeled user data based on conditional weights, and the pre-processed data is combined with the labeled data through a direct push semi-supervised learning method to finally obtain optimized training sample data, which can effectively improve the data processing efficiency, as well as the quantity and quality of the training sample data, and enhance the reliability of the credit scoring model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work:

[0048] Figure 1 It is a flow chart of a risk control decision method based on direct induction semi-supervised learning according to a preferred embodiment of the present invention;

[0049] Figure 2 It is a principle block diagram of a risk control decision system based on direct induction semi-supervised learning in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the following will be described clearly and completely in combination with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0051] The risk control decision-making method based on direct induction semi-supervised learning in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the following steps are included:

[0052] S01: Collect labeled user data as a first data sample, and collect unlabeled user data as a basic database;

[0053] S02: According to the purpose of the credit scoring model, multiple data attribute indicators are set and different weights are assigned to the multiple data attribute indicators;

[0054] S03: obtaining multiple data attributes of user data in the first data sample according to multiple data attribute indicators, and generating screening conditions according to a set strategy;

[0055] S04: screening in the basic database according to the screening conditions, and adjusting the number of screening results according to the weight of the data attribute index, and finally obtaining a second data sample;

[0056] S05: taking the first data sample as a sample with a class label and the second data sample as a sample without a class label, and predicting the class label of the second data sample by using a direct induction semi-supervised learning algorithm;

[0057] S06: Using the first data sample and the predicted second data sample as labeled samples, training a credit score card model.

[0058] By applying the method and approach of this application, guided by the purpose of the credit scoring model, a data screening mechanism that is more in line with the actual application scenario is added to pre-process the unlabeled user data based on conditional weights, and the pre-processed data is combined with the labeled data through a direct semi-supervised learning method to finally obtain optimized training sample data, which can effectively improve data processing efficiency, as well as the quantity and quality of training sample data, and enhance the reliability of the credit scoring model.

[0059] According to different purposes (application scenarios), different weighted attributes can be selected, and the weights of these attributes can be specifically distinguished. The setting of such conditions is the premise for implementing the method of this application; on this basis, the collected unidentified data is screened and optimized, and the composition structure is adjusted so that the overall composition of the training data is closer to the expected ideal training data, which is also the key to the final trained model being more in line with the actual scenario;

[0060] It should be noted that for different application scenarios or different bank advantages, different emphasis attributes can usually be customized to obtain a more personalized model;

[0061] Preferably, a plurality of data attributes of the user data in the first data sample are obtained according to a plurality of data attribute indicators, and a screening condition is generated according to a set strategy using the following method:

[0062] According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data;

[0063] The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted;

[0064] Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a screening condition;

[0065] Segmenting the data is the basis for optimizing the data structure. Only when such segmentation is established can the subsequent operations of increasing or decreasing the amount of data be facilitated.

[0066] Preferably, screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally the second data sample is obtained using the method:

[0067] Traverse the data in the basic database according to the data attributes to obtain a data set containing the data;

[0068] Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes;

[0069] According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted;

[0070] Taking the set of user data corresponding to the multiple data sets finally obtained as the second data sample;

[0071] The second data sample obtained after further screening and adjustment of the data structure can be more close to the ideal training data.

[0072] Preferably, according to the ratio of the weights of multiple data attribute indicators, the method of further adjusting the number of data in the data set is as follows:

[0073] Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark;

[0074] Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks;

[0075] Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

[0076] A risk control decision system based on direct semi-supervised learning, such as Figure 2 As shown, the system includes: a data collection unit 100, a data storage unit 101, a data pre-processing unit 102 and a data processing unit 103;

[0077] The data collection unit 100 is used to collect labeled user data as a first data sample, collect unlabeled user data as a basic database, and store them in the data storage unit 101;

[0078] The data preprocessing unit 102 is used to set a plurality of data attribute indicators and assign different weights to the plurality of data attribute indicators according to the purpose of the credit scoring model; obtain a plurality of data attributes of the user data in the first data sample according to the plurality of data attribute indicators, and generate a screening condition according to the set strategy; perform screening in the basic database according to the screening condition, and adjust the number of screening results according to the weight of the data attribute indicator, and finally obtain a second data sample;

[0079] The data processing unit 103 is used to use the first data sample as a sample with a class label and the second data sample as a sample without a class label, and predict the class label of the second data sample by using a direct induction semi-supervised learning algorithm;

[0080] The system of the present application is guided by the purpose of the credit scoring model and adds a data screening mechanism that is more in line with the actual application scenario to pre-process the unlabeled user data based on conditional weights, and combines the pre-processed data with the labeled data through a direct semi-supervised learning method to finally obtain optimized training sample data, which can effectively improve the data processing efficiency, as well as the quantity and quality of the training sample data, and enhance the reliability of the credit scoring model.

[0081] Preferably, the system further comprises a model training unit 104;

[0082] The model training unit 104 is used to train the credit score card model using the first data sample and the predicted second data sample as labeled samples.

[0083] Preferably, a plurality of data attributes of the user data in the first data sample are obtained according to a plurality of data attribute indicators, and a screening condition is generated according to a set strategy using the following method:

[0084] According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data;

[0085] The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted;

[0086] Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a screening condition;

[0087] Segmenting the data is the basis for optimizing the data structure. Only when such segmentation is established can the subsequent operations of increasing or decreasing the amount of data be facilitated.

[0088] Preferably, screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally the second data sample is obtained using the method:

[0089] Traverse the data in the basic database according to the data attributes to obtain a data set containing the data;

[0090] Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes;

[0091] According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted;

[0092] Taking the set of user data corresponding to the multiple data sets finally obtained as the second data sample;

[0093] The second data sample obtained after further screening and adjustment of the data structure can be more close to the ideal training data.

[0094] Preferably, according to the ratio of the weights of multiple data attribute indicators, the method of further adjusting the number of data in the data set is as follows:

[0095] Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark;

[0096] Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks;

[0097] Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

[0098] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A risk control decision-making method based on direct induction semi-supervised learning, characterized in that: The method comprises the following steps: Collect labeled user data as the first data sample, and collect unlabeled user data as the basic database; According to the purpose of the credit scoring model, multiple data attribute indicators are set and different weights are assigned to the multiple data attribute indicators; Acquire multiple data attributes of user data in the first data sample according to multiple data attribute indicators, and generate screening conditions according to the set strategy; Screening is performed in the basic database according to the screening conditions, and the number of screening results is adjusted according to the weight of the data attribute index, and finally a second data sample is obtained; The first data sample is taken as a sample with a class label, and the second data sample is taken as a sample without a class label. The class label of the second data sample is predicted by a direct inference semi-supervised learning algorithm.

2. The risk control decision-making method based on transductive semi-supervised learning according to claim 1 is characterized in that: The method further comprises the steps of: The credit score card model is trained using the first data sample and the predicted second data sample as labeled samples.

3. The risk control decision-making method based on direct induction semi-supervised learning according to claim 1 is characterized in that: The method of obtaining multiple data attributes of user data in the first data sample according to multiple data attribute indicators and generating screening conditions according to the set strategy adopts: According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data; The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted; Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a filtering condition.

4. The risk control decision-making method based on direct induction semi-supervised learning according to claim 3 is characterized in that: The method of screening in the basic database according to the screening conditions and adjusting the number of screening results according to the weight of the data attribute index to finally obtain the second data sample is as follows: Traverse the data in the basic database according to the data attributes to obtain a data set containing the data; Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes; According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted; The set of user data corresponding to the multiple data sets finally obtained is used as the second data sample.

5. The risk control decision-making method based on transductive semi-supervised learning according to claim 4 is characterized in that: The method for further adjusting the number of data in the data set based on the weight ratio of multiple data attribute indicators is as follows: Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark; Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks; Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

6. A risk control decision system based on direct induction semi-supervised learning, characterized in that: The system comprises: a data collection unit, a data storage unit, a data preprocessing unit and a data processing unit; The data collection unit is used to collect labeled user data as the first data sample, collect unlabeled user data as the basic database, and store them in the data storage unit; The data preprocessing unit is used to set a plurality of data attribute indicators and assign different weights to the plurality of data attribute indicators according to the purpose of the credit scoring model; obtain a plurality of data attributes of the user data in the first data sample according to the plurality of data attribute indicators, and generate screening conditions according to the set strategy; perform screening in the basic database according to the screening conditions, and adjust the number of screening results according to the weights of the data attribute indicators, and finally obtain the second data sample; The data processing unit is used to use the first data sample as a sample with a class label and the second data sample as a sample without a class label, and predict the class label of the second data sample through a direct inference semi-supervised learning algorithm.

7. The risk control decision system based on transductive semi-supervised learning according to claim 6 is characterized in that: The system also includes a model training unit; The model training unit is used to train the credit score card model using the first data sample and the predicted second data sample as labeled samples.

8. The risk control decision system based on transductive semi-supervised learning according to claim 6 is characterized in that: The method of obtaining multiple data attributes of user data in the first data sample according to multiple data attribute indicators and generating screening conditions according to the set strategy adopts: According to each data attribute indicator, the data attribute of the user data in the first data sample is matched, and a data attribute set is established for the matched qualified user data; The data attribute set is segmented respectively, and the corresponding label information is added to each segment and stored in the label of the user data, and the percentage of each segment data in the total data is counted; Based on the percentage of each segment of data in the total data, adjust the value range of data attributes and use it as a filtering condition.

9. The risk control decision system based on transductive semi-supervised learning according to claim 8 is characterized in that: The method of screening in the basic database according to the screening conditions and adjusting the number of screening results according to the weight of the data attribute index to finally obtain the second data sample is as follows: Traverse the data in the basic database according to the data attributes to obtain a data set containing the data; Filter out the parts of the data set that do not meet the requirements based on the value range of the data attributes; According to the ratio of the weights of multiple data attribute indicators, the amount of data in the data set is further adjusted; The set of user data corresponding to the multiple data sets finally obtained is used as the second data sample.

10. The risk control decision system based on transductive semi-supervised learning according to claim 9 is characterized in that: The method for further adjusting the number of data in the data set based on the weight ratio of multiple data attribute indicators is as follows: Taking the smallest weight of multiple data attribute indicators as a benchmark, determining the benchmark quantity of the data set corresponding to the benchmark; Determine the number of data sets corresponding to other data attribute indicators based on the ratio of the weight value of other data attribute indicators to the value of the benchmark and the number of benchmarks; Increase or decrease the data of the corresponding data set according to the number of data sets corresponding to other data attribute indicators.

Images