Risk control decision method and system, storage medium and electronic equipment

Abstract

The invention provides a risk control decision method and system, a storage medium and electronic equipment, and the method comprises the steps: collecting sample data, and carrying out the preprocessing operation; constructing a correlation graph by taking various sample data as nodes and taking the relationship of the correlated sample data as edges; applying a community discovery algorithm to search for sample data of different risk levels, and performing risk classification on various sample data; performing feature extraction on the sample data through attributes of nodes and edges, and mapping the sample data to a low-dimensional vector space; establishing a LightGBM model, and training the LightGBM model by using the low-dimensional feature vector and superposing a local variable and an external three-way variable as an in-modulus variable, so as to carry out risk prediction on new sample data and carry out risk control decision making; and comparing a predicted result of the sample data with an actual result so as to continuously optimize the LightGBM model. According to the method, risk assessment can be carried out on new data, and model optimization and adjustment can be flexibly carried out along with increase of data scales and types.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk control, and in particular to a risk control decision-making method, system, storage medium and electronic equipment. Background Art

[0002] Among existing technologies, financial institutions mainly use risk control methods based on rules and statistical models. These methods usually establish risk assessment models based on historical data, but cannot effectively process large-scale data and mine complex correlations between data. In addition, traditional methods also have shortcomings in the accuracy and real-time nature of risk prediction. Summary of the invention

[0003] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a risk control decision method, which aims to solve the problem that in the prior art, financial institutions mainly use risk control methods based on rules and statistical models. These methods usually establish risk assessment models based on historical data, but cannot effectively process large-scale data and mine complex correlations between data. In addition, traditional methods also have insufficient technical problems in terms of accuracy and real-time performance of risk prediction.

[0004] In order to achieve the above object, the present invention is implemented by the following technical solutions: A risk control decision-making method comprises the following steps: Collecting sample data related to risk control and performing preprocessing operations on the sample data; Constructing a correlation graph using various types of sample data as nodes and relationships between the associated sample data as edges; Applying a community discovery algorithm to find sample data of different risk levels in the association graph, and classifying the risks of various sample data according to the risk levels; Using the association graph, extracting features from the sample data through the attributes of the nodes and the edges, and mapping them into a low-dimensional vector space; Establish a LightGBM model, use the feature vectors in the low-dimensional vector space, and superimpose local variables and external tripartite variables as input variables to train the LightGBM model; Use the trained LightGBM model to predict the risk of new sample data, and make risk control decisions based on the results of risk prediction; In the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

[0005] According to one aspect of the above technical solution, the sample data includes basic user information, transaction records, and credit scores, and the preprocessing operation includes cleaning, deduplication, and outlier processing of the sample data.

[0006] According to one aspect of the above technical solution, the construction of the association graph using various types of sample data as nodes and the relationships between the associated sample data as edges specifically includes: The user's basic information, transaction records and credit scores are used as nodes; Based on big data, the transaction relationship, identity relationship and credit relationship between the user's basic information, transaction records and credit scores are found to construct an association graph.

[0007] According to one aspect of the above technical solution, the application community discovery algorithm searches for sample data of different risk levels in the association graph, and classifies the sample data according to the risk level, specifically including: Find out the behavior habit characteristics of each user in the association graph and establish a behavior habit characteristic matrix; The single user pointed to by all sample data with the highest financial risk is regarded as a high-risk group; Taking the behavior habit feature matrix of the high-risk group as a benchmark, traversing and calculating the similarity between the behavior habit feature matrix of other users and the behavior habit feature matrix of the high-risk group; ; Among them, Vj represents the behavior habit feature matrix of high-risk groups, Vi represents the behavior habit feature matrix of other users, Ij represents the behavior habit feature vector in the behavior habit feature matrix of high-risk groups, Ii represents the behavior habit feature vector in the behavior habit feature matrix of other users, k represents a single behavior habit feature element in the behavior habit feature vector, m represents the number of types of single behavior habit feature elements, and n represents the number of users; Setting a multi-level similarity interval, calculating the similarity interval in which the similarity between the behavior habit feature matrix of each user and the behavior habit feature matrix of the high-risk group is located, so as to classify the risk of the user itself; All sample data related to the user are equated with the risk classification of the user itself.

[0008] According to one aspect of the above technical solution, the method of using the association graph to extract features from the sample data through the attributes of the nodes and the edges and mapping them to a low-dimensional vector space specifically includes: Extracting multiple texts from the information contained in each of the nodes and the edges, and selecting a keyword for each of the texts to construct a feature matrix; ; Calculate the covariance matrix R of the feature matrix A , and calculate the eigenvalues ​​of the covariance matrix and the corresponding eigenvector e 1 , e 2 ,…,e v , where u represents the number of texts and v represents the number of keywords; Select the dimension z to which the eigenvector is to be mapped, and select the best eigenvector corresponding to the z largest eigenvalues ​​of the covariance matrix to form the best transformation matrix W=[e 1 , e 2 ,…,e z ] T , T represents transpose; The optimal conversion matrix is ​​subjected to dimensionality reduction processing and converted into a dimensionality reduction feature matrix Y=[g 1 , g 2 , …, g z ].

[0009] According to one aspect of the above technical solution, the LightGBM model is established, and the characteristic vectors in the low-dimensional vector space are used, and local variables and external tripartite variables are superimposed as input variables to train the LightGBM model, specifically including: Establish a LightGBM model, and use the eigenvectors in the dimension-reduced feature matrix as eigenvectors in the low-dimensional vector space; Obtain local risk variables and external third-party risk variables based on big data; Inputting the feature vector, local risk variables and external third-party risk variables in the low-dimensional vector space as model input variables into the LightGBM model for training; The LightGBM model collects input variables according to different risk classifications, substitutes them into the linear regression algorithm for analysis, and then obtains the health score of the LightGBM model; ; Among them, θ represents the regression coefficient, g represents the eigenvector in the low-dimensional vector space, bd represents the local risk variable, l represents the number of local risk variables, wb represents the external third-party risk variable, and q represents the number of external third-party risk variables.

[0010] According to one aspect of the above technical solution, the trained LightGBM model is used to predict the risk of new sample data, and risk control decisions are made based on the results of risk prediction, specifically including: Set the health score interval, count the health score intervals where the health scores of the input variables of each risk classification are located in the LightGBM model, and then perform secondary risk classification on the sample data carried by the input variables; The similarity of the behavioral habit feature matrix of the same user is combined with the health score to obtain the risk score of the user; ; Among them, H represents the sum of the health scores of all users by the LightGBM model, and H i Represents the health score of the LightGBM model for the i-th user; Based on the risk score P of different users i , select the corresponding risk control decision.

[0011] The present invention also provides a risk control decision system, comprising: Collection module: used to collect sample data related to risk control and perform preprocessing operations on the sample data; the sample data includes basic user information, transaction records, and credit scores, and the preprocessing operations include cleaning, deduplication, and outlier processing of the sample data; Construction module: used to construct a correlation graph using various types of sample data as nodes and the relationships between the associated sample data as edges; The construction module is specifically used to: use the user basic information, transaction records and credit scores as nodes; Based on big data, find out the transaction relationship, identity relationship and credit relationship between the basic information, transaction records and credit scores of the user in pairs, so as to construct a correlation graph; Classification module: used to apply community discovery algorithm to find sample data of different risk levels in the association graph, and classify the sample data according to the risk level; The classification module is specifically used to: find out the behavioral habit characteristics of each user in the association map and establish a behavioral habit characteristic matrix; The single user pointed to by all sample data with the highest financial risk is regarded as a high-risk group; Taking the behavior habit feature matrix of the high-risk group as a benchmark, traversing and calculating the similarity between the behavior habit feature matrix of other users and the behavior habit feature matrix of the high-risk group; ; Among them, Vj represents the behavior habit feature matrix of high-risk groups, Vi represents the behavior habit feature matrix of other users, Ij represents the behavior habit feature vector in the behavior habit feature matrix of high-risk groups, Ii represents the behavior habit feature vector in the behavior habit feature matrix of other users, k represents a single behavior habit feature element in the behavior habit feature vector, m represents the number of types of single behavior habit feature elements, and n represents the number of users; Setting a multi-level similarity interval, calculating the similarity interval in which the similarity between the behavior habit feature matrix of each user and the behavior habit feature matrix of the high-risk group is located, so as to classify the risk of the user itself; Equalize all sample data related to the user to the risk classification of the user itself; Extraction module: used to extract features of the sample data by using the association graph through the attributes of the nodes and the edges, and map them into a low-dimensional vector space; The extraction module is specifically used to: extract multiple texts from the information contained in each of the nodes and the edges, and select a keyword for each of the texts to construct a feature matrix; ; Calculate the covariance matrix R of the feature matrix A , and calculate the eigenvalues ​​of the covariance matrix and the corresponding eigenvector e 1 , e 2 ,…,e v , where u represents the number of texts and v represents the number of keywords; Select the dimension z to which the eigenvector is to be mapped, and select the best eigenvector corresponding to the z largest eigenvalues ​​of the covariance matrix to form the best transformation matrix W=[e 1 , e 2 ,…,e z ] T , T represents transpose; The optimal conversion matrix is ​​subjected to dimensionality reduction processing and converted into a dimensionality reduction feature matrix Y=[g 1 , g 2 , …, g z ]; Training module: establish a LightGBM model, use the feature vectors in the low-dimensional vector space, and superimpose local variables and external tripartite variables as input variables to train the LightGBM model; The training module is specifically used to: establish a LightGBM model, and use the eigenvectors in the reduced-dimensional feature matrix as eigenvectors in the low-dimensional vector space; Obtain local risk variables and external third-party risk variables based on big data; Inputting the feature vector, local risk variables and external third-party risk variables in the low-dimensional vector space as model input variables into the LightGBM model for training; The LightGBM model collects input variables according to different risk classifications, substitutes them into the linear regression algorithm for analysis, and then obtains the health score of the LightGBM model; ; Among them, θ represents the regression coefficient, g represents the eigenvector in the low-dimensional vector space, bd represents the local risk variable, l represents the number of local risk variables, wb represents the external third-party risk variable, and q represents the number of external third-party risk variables; Prediction module: Use the trained LightGBM model to predict the risk of new sample data, and make risk control decisions based on the results of risk prediction; The prediction module is specifically used to: set the health score interval, count the health score interval where the health score of the input variable of each risk classification is located in the LightGBM model, and then perform risk secondary classification on the sample data carried by the input variable; The similarity of the behavioral habit feature matrix of the same user is combined with the health score to obtain the risk score of the user; ; Among them, H represents the sum of the health scores of all users by the LightGBM model, and H i Represents the health score of the LightGBM model for the i-th user; Based on the risk score P of different users i , select the corresponding risk control decision; Optimization module: In the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

[0012] The present invention also provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the risk control decision method as described above is implemented.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the risk control decision method as described above when executing the computer program.

[0014] Compared with the prior art, the present invention has the following beneficial effects: Through the association graph, large-scale data can be effectively represented and processed, and complex associations between data can be mined. It can more comprehensively consider the impact of various factors on risk and extract more representative features, which is conducive to improving the accuracy and reliability of risk prediction; through the community discovery algorithm, high-risk groups can be found, and corresponding rules can be extracted through experience to capture high-risk customer groups; by extracting features from the data in the association graph and mapping them to a low-dimensional vector space, it is convenient for subsequent risk prediction; by building a LightGBM model, the machine learning algorithm based on the gradient boosting decision tree has the characteristics of high efficiency and scalability. Using the feature vector extracted from the association graph, superimposing local variables and external tripartite variables to train the LightGBM model, the probability value or score with risk discrimination can be obtained, and then the rules and thresholds for risk control decisions can be obtained. Use the trained model to predict the risk of new data to determine whether the user has high risk. It is efficient and scalable. Through the LightGBM model training, high-precision risk prediction results can be obtained, and finally risk control decisions are made based on the prediction results of the LightGBM model. If the user is judged to be high risk, corresponding risk control measures will be taken, such as loan rejection, account closure, transaction restriction, margin increase, etc. At the same time, the user's actual risk situation will be compared with the predicted results, and the model will be continuously optimized and adjusted to further improve the accuracy and reliability of risk control decisions; The risk control decision-making method of the present invention is real-time and can quickly perform risk assessment on new data. As the scale and type of data increase, the method of the present invention can flexibly optimize and adjust the model. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 This is a flow chart of the risk control decision-making method in the first embodiment of the present invention; Figure 2 It is a structural block diagram of the association map in the first embodiment of the present invention; Figure 3 This is a structural block diagram of the risk control decision system in the second embodiment of the present invention; Figure 4 is a structural block diagram of an electronic device in a third embodiment of the present invention; The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0015] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0016] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0018] See also Figure 1 , which shows a risk control decision method in the first embodiment of the present invention, comprising the following steps: S10, collecting sample data related to risk control, and performing preprocessing operations on the sample data; S20, constructing a correlation graph using various types of sample data as nodes and relationships between the associated sample data as edges; S30, applying a community discovery algorithm to search for sample data of different risk levels in the association graph, and classifying the risks of various sample data according to the risk levels; S40, using the association graph, extracting features from the sample data through the attributes of the nodes and the edges, and mapping them into a low-dimensional vector space; S50, establishing a LightGBM model, using the feature vectors in the low-dimensional vector space, and superimposing local variables and external tripartite variables as input variables to train the LightGBM model; S60, using the trained LightGBM model to perform risk prediction on new sample data, and making risk control decisions based on the results of the risk prediction; S70, in the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

[0019] It is understandable that through the association graph, large-scale data can be effectively represented and processed, and complex associations between data can be mined. It can more comprehensively consider the impact of various factors on risk and extract more representative features, which is conducive to improving the accuracy and reliability of risk prediction; through the community discovery algorithm, high-risk groups can be found, and corresponding rules can be extracted through experience to capture high-risk customer groups; by extracting features from the data in the association graph and mapping them to a low-dimensional vector space, it is convenient for subsequent risk prediction; by building the LightGBM model, the machine learning algorithm based on the gradient boosting decision tree has the characteristics of high efficiency and scalability. Using the feature vector extracted from the association graph, superimposing local variables and external tripartite variables to train the LightGBM model, the probability value or score with risk discrimination can be obtained, and then the rules and thresholds for risk control decisions can be obtained. Use the trained model to predict the risk of new data to determine whether the user has high risk. It is efficient and scalable. Through the LightGBM model training, high-precision risk prediction results can be obtained, and finally risk control decisions are made based on the prediction results of the LightGBM model. If the user is judged to be high risk, corresponding risk control measures will be taken, such as loan rejection, account closure, transaction restriction, margin increase, etc. At the same time, the user's actual risk situation will be compared with the predicted results, and the model will be continuously optimized and adjusted to further improve the accuracy and reliability of risk control decisions; The risk control decision-making method of the present invention is real-time and can quickly perform risk assessment on new data. As the scale and type of data increase, the method of the present invention can flexibly optimize and adjust the model.

[0020] Specifically, the sample data includes basic user information, transaction records, and credit scores, and the preprocessing operation includes cleaning, deduplication, and outlier processing of the sample data.

[0021] Furthermore, the step S20 specifically includes: The user's basic information, transaction records and credit scores are used as nodes; Based on big data, the transaction relationship, identity relationship and credit relationship between the user's basic information, transaction records and credit scores are found to construct an association graph.

[0022] For example, see Figure 2, the circle represents a node, the connection line represents an edge. The user's basic information includes the user's ID card, denoted by "body", bank card, denoted by "card", mobile phone number, denoted by "number", device, denoted by "device", transaction records and credit scores are sub-information implicit in the user's basic information and carried by the nodes. Then, the relationships between each node are found based on big data and used as edges. The relationships represented by the edges include transaction relationships (borrowing), identity relationships (contacts), and credit relationships are implicit relationships carried by the edges.

[0023] Further, the specific steps of step S30 include: Find out the behavior habit characteristics of each user in the association graph and establish a behavior habit characteristic matrix; Take the single user pointed to by all the sample data with the highest financial risk as the high-risk gang; Based on the behavior habit characteristic matrix of the high-risk gang, traverse and calculate the similarity between the behavior habit characteristic matrix of other users and the behavior habit characteristic matrix of the high-risk gang; ; Among them, Vj represents the behavior habit characteristic matrix of the high-risk gang, Vi represents the behavior habit characteristic matrix of other users, Ij represents the behavior habit characteristic vector in the behavior habit characteristic matrix of the high-risk gang, Ii represents the behavior habit characteristic vector in the behavior habit characteristic matrix of other users, k represents a single behavior habit characteristic element in the behavior habit characteristic vector, m represents the number of types of single behavior habit characteristic elements, and n represents the number of users; Set multiple similarity intervals, calculate the similarity interval where the similarity between the behavior habit characteristic matrix of each user and the behavior habit characteristic matrix of the high-risk gang is located, so as to classify the risk of the user himself; Equate all the sample data related to the user to the risk classification of the user himself.

[0024] It can be understood that by finding out the behavior habit characteristics of each user, the similarity between multiple related users can be obtained, and taking the single user with the highest financial risk (i.e., the worst behavior) as the benchmark, according to the behavior habit similarity between other users and it, the financial risk situation of other users can be evaluated. By setting multiple similarity intervals, for example, users with a similarity between 0 and 0.3 are classified as low-risk user groups, those between 0.3 and 0.6 are classified as medium-risk user groups, and those between 0.6 and 1.0 are classified as high-risk user groups, so as to classify the risk of each user himself, and the sample data corresponding to the user himself is also classified as the same risk category.

[0025] It should be noted that Figure 2In the figure, “body 1”, “card 1”, “number 1” and the corresponding edges all point to the same user 1.

[0026] Furthermore, the step S40 specifically includes: Extracting multiple texts from the information contained in each of the nodes and the edges, and selecting a keyword for each of the texts to construct a feature matrix; ; Calculate the covariance matrix R of the feature matrix A , and calculate the eigenvalues ​​of the covariance matrix and the corresponding eigenvector e 1 , e 2 ,…,e v , where u represents the number of texts and v represents the number of keywords; Select the dimension z to which the eigenvector is to be mapped, and select the best eigenvector corresponding to the z largest eigenvalues ​​of the covariance matrix to form the best transformation matrix W=[e 1 , e 2 ,…,e z ] T , T represents transpose; The optimal conversion matrix is ​​subjected to dimensionality reduction processing and converted into a dimensionality reduction feature matrix Y=[g 1 , g 2 , …, g z ].

[0027] It can be understood that by using the association graph, the relevant features of the sample data are extracted as text, and keywords are extracted from the text. The features of the sample data can be represented by constructing a feature matrix, and then dimensionality reduction is performed to map the features into a low-dimensional vector space, so as to facilitate subsequent risk prediction and capture high-risk groups.

[0028] Furthermore, the step S50 specifically includes: Establish a LightGBM model, and use the eigenvectors in the dimension-reduced feature matrix as eigenvectors in the low-dimensional vector space; Obtain local risk variables and external third-party risk variables based on big data; Inputting the feature vector, local risk variables and external third-party risk variables in the low-dimensional vector space as model input variables into the LightGBM model for training; The LightGBM model collects input variables according to different risk classifications, substitutes them into the linear regression algorithm for analysis, and then obtains the health score of the LightGBM model; ; Among them, θ represents the regression coefficient, g represents the eigenvector in the low-dimensional vector space, bd represents the local risk variable, l represents the number of local risk variables, wb represents the external third-party risk variable, and q represents the number of external third-party risk variables.

[0029] It can be understood that LightGBM is a machine learning algorithm based on gradient boosting decision trees, which is efficient and scalable. By using the feature vectors extracted from the association graph, local risk variables and external third-party risk variables are superimposed to train the LightGBM model to obtain a probability value or score with risk differentiation, and then the rules and thresholds for risk control decisions can be obtained. Use the trained model to predict the risk of new data to determine whether the user is high-risk. Local risk variables include adjustments to the activity strategies of financial institutions, relaxation or tightening of the judgment criteria for high-risk groups, and cooperation between financial institutions and external parties. External third-party risk variables include variables other than financial behavior of users, such as civil liability, criminal liability, etc., which can all be considered as factors.

[0030] When training the LightGBM model, the value of the regression coefficient is set according to the impact of the variable on the health status of the server (such as increasing the health status or decreasing the health status). When the variable causes the server health score to increase, the regression coefficient θ is greater than 0. When the variable causes the output health score to decrease, the regression coefficient θ is less than 0.

[0031] Furthermore, the step S60 specifically includes: Set the health score interval, count the health score intervals where the health scores of the input variables of each risk classification are located in the LightGBM model, and then perform secondary risk classification on the sample data carried by the input variables; The similarity of the behavioral habit feature matrix of the same user is combined with the health score to obtain the risk score of the user; ; Among them, H represents the sum of the health scores of all users by the LightGBM model, and H i Represents the health score of the LightGBM model for the i-th user; Based on the risk score P of different users i , select the corresponding risk control decision.

[0032] It can be understood that by using the similarity between a user and a high-risk group calculated in the previous step as a weight factor to adjust the LightGBM model's health score for the user and then calculate the user's true risk score, it can avoid the LightGBM model falling into the local optimum or worst due to incorrect selection of text and keywords during feature extraction, thereby affecting the user's true risk score. It can greatly affect the user's true risk score based on the determined high-risk groups, but can also take into account the input variables queried from big data, which not only improves the authenticity of the results, but also makes the results more predictive and real-time.

[0033] In summary, the risk control decision-making method in the above-mentioned embodiment of the present invention is real-time and can quickly perform risk assessment on new data. As the scale and type of data increase, the method of the present invention can flexibly optimize and adjust the model. Please refer to Figure 3 to Figure 4 , which shows a risk control decision system in a second embodiment of the present invention, including: Collection module 11: used to collect sample data related to risk control and perform preprocessing operations on the sample data; the sample data includes basic user information, transaction records, and credit scores, and the preprocessing operations include cleaning, deduplication, and outlier processing of the sample data; Construction module 12: used to construct a correlation graph using various types of sample data as nodes and relationships between the associated sample data as edges; The construction module 12 is specifically used to: use the user basic information, transaction records and credit scores as nodes; Based on big data, find out the transaction relationship, identity relationship and credit relationship between the basic information, transaction records and credit scores of the user in pairs, so as to construct a correlation graph; Classification module 13: used to apply community discovery algorithm to find sample data of different risk levels in the association graph, and classify the sample data according to the risk level; The classification module 13 is specifically used to: find out the behavioral habit characteristics of each user in the association map and establish a behavioral habit characteristic matrix; The single user pointed to by all sample data with the highest financial risk is regarded as a high-risk group; Taking the behavior habit feature matrix of the high-risk group as a benchmark, traversing and calculating the similarity between the behavior habit feature matrix of other users and the behavior habit feature matrix of the high-risk group; ; Among them, Vj represents the behavior habit feature matrix of high-risk groups, Vi represents the behavior habit feature matrix of other users, Ij represents the behavior habit feature vector in the behavior habit feature matrix of high-risk groups, Ii represents the behavior habit feature vector in the behavior habit feature matrix of other users, k represents a single behavior habit feature element in the behavior habit feature vector, m represents the number of types of single behavior habit feature elements, and n represents the number of users; Setting a multi-level similarity interval, calculating the similarity interval in which the similarity between the behavior habit feature matrix of each user and the behavior habit feature matrix of the high-risk group is located, so as to classify the risk of the user itself; Equalize all sample data related to the user to the risk classification of the user itself; Extraction module 14: used to extract features of the sample data by using the association graph through the attributes of the nodes and the edges, and map them into a low-dimensional vector space; The extraction module 14 is specifically used to: extract multiple texts from the information contained in each of the nodes and the edges, and select a keyword for each of the texts to construct a feature matrix; ; Calculate the covariance matrix R of the feature matrix A , and calculate the eigenvalues ​​of the covariance matrix and the corresponding eigenvector e 1 , e 2 ,…,e v , where u represents the number of texts and v represents the number of keywords; Select the dimension z to which the eigenvector is to be mapped, and select the best eigenvector corresponding to the z largest eigenvalues ​​of the covariance matrix to form the best transformation matrix W=[e 1 , e 2 ,…,e z ] T , T represents transpose; The optimal conversion matrix is ​​subjected to dimensionality reduction processing and converted into a dimensionality reduction feature matrix Y=[g 1 , g 2 , …, g z ]; Training module 15: Establishing a LightGBM model, using the feature vectors in the low-dimensional vector space, and superimposing local variables and external tripartite variables as input variables to train the LightGBM model; The training module 15 is specifically used to: establish a LightGBM model, and use the eigenvectors in the reduced-dimensional feature matrix as eigenvectors in the low-dimensional vector space; Obtain local risk variables and external third-party risk variables based on big data; Inputting the feature vector, local risk variables and external third-party risk variables in the low-dimensional vector space as model input variables into the LightGBM model for training; The LightGBM model collects input variables according to different risk classifications, substitutes them into the linear regression algorithm for analysis, and then obtains the health score of the LightGBM model; ; Among them, θ represents the regression coefficient, g represents the eigenvector in the low-dimensional vector space, bd represents the local risk variable, l represents the number of local risk variables, wb represents the external third-party risk variable, and q represents the number of external third-party risk variables; Prediction module 16: Use the trained LightGBM model to predict the risk of new sample data, and make risk control decisions based on the results of risk prediction; The prediction module 16 is specifically used to: set a health score interval, count the health score intervals where the health scores of the input variables of each risk classification are located in the LightGBM model, and then perform risk secondary classification on the sample data carried by the input variables; The similarity of the behavioral habit feature matrix of the same user is combined with the health score to obtain the risk score of the user; ; Among them, H represents the sum of the health scores of all users by the LightGBM model, and H i Represents the health score of the LightGBM model for the i-th user; Based on the risk score P of different users i , select the corresponding risk control decision; Optimization module 17: In the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

[0034] The present invention also provides an electronic device, see Figure 4 , shown is an electronic device in the third embodiment of the present invention, including a memory 10, a processor 20, and a computer program 30 stored in the memory 10 and executable on the processor 20, and the processor 20 implements the above-mentioned risk control decision method when executing the computer program 30.

[0035] Among them, the memory 10 includes at least one type of storage medium, and the storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 10 can be an internal storage unit of an electronic device, such as a hard disk of the electronic device. In other embodiments, the memory 10 can also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Further, the memory 10 can also include both an internal storage unit of the electronic device and an external storage device. The memory 10 can not only be used to store application software and various types of data installed in the electronic device, but also can be used to temporarily store data that has been output or is to be output.

[0036] Among them, in some embodiments, the processor 20 can be an electronic control unit (Electronic Control Unit, abbreviated as ECU, also known as a vehicle computer), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run the program code stored in the memory 10 or process data, such as executing access restriction programs, etc.

[0037] It should be pointed out that Figure 4 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than those shown in the figure, or combine certain components, or arrange the components differently.

[0038] An embodiment of the present invention further provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the risk control decision method as described above is implemented.

[0039] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0040] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0041] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0042] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0043] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A risk control decision-making method, characterized in that: The steps include: Collecting sample data related to risk control and performing preprocessing operations on the sample data; Constructing a correlation graph using various types of sample data as nodes and relationships between the associated sample data as edges; Applying a community discovery algorithm to find sample data of different risk levels in the association graph, and classifying the risks of various sample data according to the risk levels; Using the association graph, extracting features from the sample data through the attributes of the nodes and the edges, and mapping them into a low-dimensional vector space; Establish a LightGBM model, use the feature vectors in the low-dimensional vector space, and superimpose local variables and external tripartite variables as input variables to train the LightGBM model; Use the trained LightGBM model to predict the risk of new sample data, and make risk control decisions based on the results of risk prediction; In the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

2. The risk control decision-making method according to claim 1, characterized in that: The sample data includes basic user information, transaction records, and credit scores, and the preprocessing operations include cleaning, deduplication, and outlier processing of the sample data.

3. The risk control decision-making method according to claim 2, characterized in that: The construction of the association graph using various types of sample data as nodes and the relationships between the associated sample data as edges specifically includes: The user's basic information, transaction records and credit scores are used as nodes; Based on big data, the transaction relationship, identity relationship and credit relationship between the user's basic information, transaction records and credit scores are found to construct an association graph.

4. The risk control decision-making method according to claim 3, characterized in that: The application community discovery algorithm searches for sample data of different risk levels in the association graph, and classifies the sample data according to the risk level, specifically including: Find out the behavior habit characteristics of each user in the association graph and establish a behavior habit characteristic matrix; The single user pointed to by all sample data with the highest financial risk is regarded as a high-risk group; Taking the behavior habit feature matrix of the high-risk group as a benchmark, traversing and calculating the similarity between the behavior habit feature matrix of other users and the behavior habit feature matrix of the high-risk group; ； Among them, Vj represents the behavior habit feature matrix of high-risk groups, Vi represents the behavior habit feature matrix of other users, Ij represents the behavior habit feature vector in the behavior habit feature matrix of high-risk groups, Ii represents the behavior habit feature vector in the behavior habit feature matrix of other users, k represents a single behavior habit feature element in the behavior habit feature vector, m represents the number of types of single behavior habit feature elements, and n represents the number of users; Setting a multi-level similarity interval, calculating the similarity interval in which the similarity between the behavior habit feature matrix of each user and the behavior habit feature matrix of the high-risk group is located, so as to classify the risk of the user itself; All sample data related to the user are equated with the risk classification of the user itself.

5. The risk control decision-making method according to claim 4, characterized in that: The method of utilizing the association graph to extract features from the sample data through the attributes of the nodes and the edges and mapping the features to a low-dimensional vector space specifically includes: Extracting multiple texts from the information contained in each of the nodes and the edges, and selecting a keyword for each of the texts to construct a feature matrix; ； Calculate the covariance matrix R of the feature matrix A , and calculate the eigenvalues ​​of the covariance matrix and the corresponding eigenvector e 1 , e 2 ,…,e v , where u represents the number of texts and v represents the number of keywords; Select the dimension z to which the eigenvector is to be mapped, and select the best eigenvector corresponding to the z largest eigenvalues ​​of the covariance matrix to form the best transformation matrix W=[e 1 , e 2 ,…,e z ] T , T represents transpose; The optimal conversion matrix is ​​subjected to dimensionality reduction processing and converted into a dimensionality reduction feature matrix Y=[g 1 , g 2 , …, g z ].

6. The risk control decision-making method according to claim 5, characterized in that: The LightGBM model is established, and the characteristic vectors in the low-dimensional vector space are used, and local variables and external tripartite variables are superimposed as input variables to train the LightGBM model, specifically including: Establish a LightGBM model, and use the eigenvectors in the dimension-reduced feature matrix as eigenvectors in the low-dimensional vector space; Obtain local risk variables and external third-party risk variables based on big data; Inputting the feature vector, local risk variables and external third-party risk variables in the low-dimensional vector space as model input variables into the LightGBM model for training; The LightGBM model collects input variables according to different risk classifications, substitutes them into the linear regression algorithm for analysis, and then obtains the health score of the LightGBM model; ； Among them, θ represents the regression coefficient, g represents the eigenvector in the low-dimensional vector space, bd represents the local risk variable, l represents the number of local risk variables, wb represents the external third-party risk variable, and q represents the number of external third-party risk variables.

7. The risk control decision-making method according to claim 6, characterized in that: The method uses the trained LightGBM model to predict the risk of new sample data and makes risk control decisions based on the results of risk prediction, specifically including: Set the health score interval, count the health score intervals where the health scores of the input variables of each risk classification are located in the LightGBM model, and then perform secondary risk classification on the sample data carried by the input variables; The similarity of the behavioral habit feature matrix of the same user is combined with the health score to obtain the risk score of the user; ； Among them, H represents the sum of the health scores of all users by the LightGBM model, and H i Represents the health score of the LightGBM model for the i-th user; Based on the risk score P of different users i , select the corresponding risk control decision.

8. A risk control decision system, characterized in that: include: Collection module: used to collect sample data related to risk control and perform preprocessing operations on the sample data; Construction module: used to construct a correlation graph using various types of sample data as nodes and the relationships between the associated sample data as edges; Classification module: used to apply community discovery algorithm to find sample data of different risk levels in the association graph, and classify the sample data according to the risk level; Extraction module: used to extract features of the sample data by using the association graph through the attributes of the nodes and the edges, and map them into a low-dimensional vector space; Training module: establish a LightGBM model, use the feature vectors in the low-dimensional vector space, and superimpose local variables and external tripartite variables as input variables to train the LightGBM model; Prediction module: Use the trained LightGBM model to predict the risk of new sample data, and make risk control decisions based on the results of risk prediction; Optimization module: In the subsequent process, the predicted results of the application sample data are compared with the actual results to continuously optimize the LightGBM model.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the risk control decision-making method as described in any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the risk control decision-making method according to any one of claims 1 to 7 is implemented.

Images