Risk prediction method, device, equipment, medium and program product

Abstract

The invention provides a risk prediction method which can be applied to the technical field of artificial intelligence and the technical field of financial science and technology. The risk prediction method comprises the following steps: acquiring enterprise credit data; converting the enterprise credit data into a feature data set; extracting an enterprise potential expression in the feature data set; extracting an enterprise relationship potential expression in the feature data set; and predicting a target enterprise credit risk through a preset machine learning model based on the enterprise potential expression and the enterprise relationship potential expression. The invention further provides a risk prediction device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence technology and financial technology, and specifically to a risk prediction method, device, equipment, medium and program product. Background Art

[0002] With the development of financial markets, corporate credit business, as the main business of banks, has brought profits to the banking industry. However, with the "double increase" of commercial banks' non-performing loan balances and non-performing loan ratios, the quality of bank assets has declined, forcing banks to take measures such as setting aside provisions, which in turn affects the bank's profits and financial conditions. Excessively high non-performing loan ratios may even damage the bank's reputation and brand image. The credit risks faced by banks are gradually increasing, which means that it is necessary to strengthen the management and control of bank credit risks.

[0003] In the existing technology, the prediction of corporate credit business by banks is often based on the financial status of the enterprise itself, and risk prediction is initiated in terms of default, credit and assets, etc., which is used as a basis to judge whether the enterprise will have overdue behavior in the future. However, the existing technology still has the disadvantage of insufficient accuracy in risk prediction results. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a risk prediction method, apparatus, device, medium and program product for improving the accuracy of risk prediction.

[0005] According to the first aspect of the present disclosure, a risk prediction method is provided, comprising: obtaining enterprise credit data; converting the enterprise credit data into a feature data set; taking the enterprise potential expression in the feature data set; extracting the enterprise relationship potential expression in the feature data set; and predicting the target enterprise credit risk through a preset machine learning model based on the enterprise potential expression and the enterprise relationship potential expression.

[0006] According to an embodiment of the present disclosure, converting the enterprise credit data into a feature data set includes: preprocessing the enterprise credit data to obtain standard credit data; establishing an intermediate feature data set based on the standard credit data; and based on a dual-track screening method, screening some features in the intermediate feature data set as a feature data set.

[0007] According to an embodiment of the present disclosure, the preprocessing of the enterprise credit data to obtain standard credit data includes: deleting abnormal values ​​and null values ​​in the enterprise credit data to obtain first enterprise credit data; and randomly sampling the first enterprise credit data to obtain the standard credit data.

[0008] According to an embodiment of the present disclosure, establishing an intermediate feature data set based on the standard credit data includes: extracting target dimension features in the standard credit data to obtain a basic feature set; forming a derived feature set based on the basic feature set; and converting the basic feature set and the derived feature set into the form of a time series to obtain the intermediate feature data set.

[0009] According to an embodiment of the present disclosure, the dual-track screening method screens some features in the intermediate feature data set as a feature data set, including: screening some features of the intermediate feature data set as a first feature data set based on an information value feature screening model; screening some features of the intermediate feature data set as a second feature data set based on a preset decision tree model; and taking the union of the first feature data set and the second feature data set as the feature data set.

[0010] According to an embodiment of the present disclosure, the information value feature screening model is based on screening some features of the intermediate feature data set as the first feature data set, including: binning the feature data in the intermediate feature data set by evidence weight to obtain N bins of evidence weights, where N is a positive integer; calculating N information values ​​based on the evidence weights of the N bins; and retaining the feature data of the N information values ​​that is greater than a first preset threshold to obtain the first feature data set.

[0011] According to an embodiment of the present disclosure, the method of screening some features of the intermediate feature data set as the second feature data set based on a preset decision tree model includes: calculating the ranking of feature data in the intermediate feature data set based on the preset decision tree model to obtain K rankings, where K is a positive integer; and retaining feature data in the K rankings that is greater than a second preset threshold to obtain the second feature data set.

[0012] According to an embodiment of the present disclosure, the extracting the potential expression of the enterprise from the feature data set includes: using the feature data set as input data of a preset recurrent neural network model, and outputting the potential expression of the enterprise.

[0013] According to an embodiment of the present disclosure, the extracting potential expressions of enterprise relationships in the feature data set includes: establishing an enterprise multi-relationship graph based on the feature data set; and extracting the potential expressions of enterprise relationships through a graph sample aggregation algorithm based on the enterprise multi-relationship graph.

[0014] According to an embodiment of the present disclosure, establishing an enterprise multi-relationship graph based on the feature data set includes: determining the enterprise guarantee relationship, equity penetration relationship and capital flow relationship based on the feature data set; taking any enterprise in the feature data set as a node, and forming the enterprise multi-relationship graph through the guarantee relationship, the equity penetration relationship and the capital flow relationship.

[0015] According to an embodiment of the present disclosure, based on the enterprise multi-relationship graph, the potential expression of the enterprise relationship is extracted through a graph sample aggregation algorithm, including: calculating the potential expression of the enterprise relationship between the target node and the domain node based on the guarantee relationship, the equity penetration relationship, the capital flow relationship and the corresponding preset weights.

[0016] The second aspect of the present disclosure provides a risk prediction device, comprising: the device comprises: a credit data acquisition module, used to acquire enterprise credit data; a feature conversion module, used to convert the enterprise credit data into a feature data set; a feature learning module, used to extract enterprise potential expressions in the feature data set; a relationship representation module, used to extract enterprise relationship potential expressions in the feature data set; and a model prediction module, used to predict the target enterprise credit risk through a preset machine learning model based on the enterprise potential expressions and the enterprise relationship potential expressions.

[0017] According to an embodiment of the present disclosure, the feature conversion module includes: a preprocessing unit, an intermediate data establishment unit and a screening unit, wherein the preprocessing unit is used to preprocess the enterprise credit data to obtain standard credit data; the intermediate data establishment unit is used to establish an intermediate feature data set based on the standard credit data; and the screening unit is used to screen some features in the intermediate feature data set as a feature data set based on a dual-track screening method.

[0018] According to an embodiment of the present disclosure, the preprocessing unit is specifically used to delete abnormal values ​​and null values ​​in the enterprise credit data to obtain first enterprise credit data; and randomly sample the first enterprise credit data to obtain the standard credit data.

[0019] According to an embodiment of the present disclosure, the intermediate data establishment unit is specifically used to extract target dimension features in the standard credit data to obtain a basic feature set; form a derived feature set based on the basic feature set; and convert the basic feature set and the derived feature set into the form of a time series to obtain the intermediate feature data set.

[0020] According to an embodiment of the present disclosure, the screening unit includes: an information value screening subunit, a decision tree screening subunit and a feature data set forming subunit, the information value screening subunit is used to screen some features of the intermediate feature data set as the first feature data set based on the information value feature screening model; the decision tree screening subunit is used to screen some features of the intermediate feature data set as the second feature data set based on a preset decision tree model; and the feature data set forming subunit is used to take the union of the first feature data set and the second feature data set as the feature data set.

[0021] According to an embodiment of the present disclosure, the information value screening subunit is specifically used to perform evidence weight binning on the feature data in the intermediate feature data set to obtain evidence weights of N bins, where N is a positive integer; calculate N information values ​​based on the evidence weights of the N bins; and retain the feature data among the N information values ​​that is greater than a first preset threshold to obtain the first feature data set.

[0022] According to an embodiment of the present disclosure, the decision tree screening subunit is specifically used to calculate the ranking of feature data in the intermediate feature data set based on the preset decision tree model to obtain K rankings, where K is a positive integer; and retain the feature data in the K rankings that is greater than a second preset threshold to obtain the second feature data set.

[0023] According to an embodiment of the present disclosure, the feature learning module is specifically used to use the feature data set as input data of a preset recurrent neural network model, and output the potential expression of the enterprise.

[0024] According to an embodiment of the present disclosure, the relationship representation module includes: a relationship graph establishment unit and a potential relationship extraction unit, the relationship graph establishment unit is used to establish an enterprise multi-relationship graph based on the feature data set; and the potential relationship extraction unit is used to extract the enterprise relationship potential expression through a graph sample aggregation algorithm based on the enterprise multi-relationship graph.

[0025] According to an embodiment of the present disclosure, the relationship graph establishing unit is specifically used to determine the enterprise guarantee relationship, equity penetration relationship and capital flow relationship based on the feature data set; any enterprise in the feature data set is used as a node to form the enterprise multi-relationship graph through the guarantee relationship, the equity penetration relationship and the capital flow relationship.

[0026] According to an embodiment of the present disclosure, the potential relationship extraction unit is specifically used to calculate the potential expression of the enterprise relationship between the target node and the domain node based on the guarantee relationship, the equity penetration relationship, the capital flow relationship and the corresponding preset weights.

[0027] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0028] The fourth aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the above computer program or instructions are executed by a processor.

[0029] The fifth aspect of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when the above computer program or instructions are executed by a processor.

[0030] In the embodiments of the present disclosure, in order to improve the accuracy of risk prediction, the embodiments of the present disclosure respectively extract the potential expression of the enterprise itself and the potential expression of the enterprise relationship between enterprises through two deep learning models, which can parse the potential information of the enterprise and the potential information between enterprise relationships, and integrate these two expressions carrying potential information to comprehensively predict the credit risk of the target enterprise. The embodiments of the present disclosure can capture the complex potential dependencies of enterprises and the multiple correlations of risks between enterprises, greatly improving the ability of enterprises to determine whether high credit scores will occur in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above contents and other purposes, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0032] Figure 1 The application scenario diagram of the risk prediction method according to the embodiment of the present disclosure is schematically shown;

[0033] Figure 2 A flow chart of a risk prediction method according to an embodiment of the present disclosure is schematically shown;

[0034] Figure 3 A flowchart schematically shows a method for establishing a feature data set according to an embodiment of the present disclosure.

[0035] Figure 4 A flowchart schematically shows a dual-track screening method according to an embodiment of the present disclosure.

[0036] Figure 5 A flowchart schematically shows a potential relationship extraction method according to an embodiment of the present disclosure.

[0037] Figure 6 A flowchart schematically shows a method for establishing data features according to an embodiment of the present disclosure.

[0038] Figure 7A flowchart schematically shows a method for screening data features according to an embodiment of the present disclosure.

[0039] Figure 8 A flowchart schematically shows a method for representing a relationship according to an embodiment of the present disclosure.

[0040] Fig. 9 Schematically shows a flow chart of a feature fusion prediction method according to an embodiment of the present disclosure

[0041] Fig.10 A schematic diagram of a risk prediction device according to an embodiment of the present disclosure is shown; and

[0042] Fig.11 A block diagram of an electronic device suitable for implementing a risk prediction method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0043] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0044] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0045] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0046] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0047] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0048] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating a person's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs, and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0049] Before the embodiments of the present disclosure are disclosed in detail, the key technical terms involved in the embodiments of the present disclosure are explained one by one, as follows:

[0050] IV information value is: (Information Value, referred to as IV), which is an important statistical indicator, especially in the fields of credit scoring, risk assessment and feature selection of machine learning. Through the IV value, the contribution of the feature to the prediction of the target variable can be quantified, so as to make more reasonable decisions.

[0051] Weight of Evidence (WOE) is a statistic used in statistics to describe the relationship between a variable and a target variable (usually a binary variable, such as good customers and bad customers). WOE is an important tool for evaluating the predictive ability of variables, especially in the fields of credit scoring models and risk assessment. WOE is closely related to IV (Information Value), which is calculated based on WOE and is used to measure the overall predictive ability of variables. The calculation of IV involves the standardized processing of WOE, taking into account the probability of occurrence and non-occurrence of events in each group.

[0052] Binning: is a technique in data analysis and statistical modeling, especially in the fields of credit scoring, risk assessment, and machine learning. The purpose of binning is to discretize continuous variables and divide the data into several intervals or "boxes" in order to better understand the data distribution, simplify the model complexity, and improve the interpretability of the model.

[0053] Inductive Graph Representation Learning Models refer to a type of graph learning model that can infer patterns from existing data and apply these patterns to new, unseen data. The core feature of this type of model is generalization, that is, the model can not only process nodes seen during training, but also nodes that did not appear during the training phase.

[0054] When banks are also faced with credit forecasts, they use cutting-edge technologies such as big data, machine learning, and deep learning to analyze the characteristics of companies with overdue loans, design corresponding algorithms and solutions to predict the possibility of target companies defaulting after lending, and then help bank credit practitioners accurately determine customers' future overdue behavior, so that they can take measures in advance such as increasing or decreasing loan limits, refusing loans, etc., to reduce the bank's non-performing loan balance and non-performing loan ratio.

[0055] However, the applicant found that most existing methods only use machine learning methods to model based on the characteristics of the company's financial status, default, credit and assets, which will ignore the deeper relationship characteristics of the company's capital flow, corporate relationships, equity penetration, etc. These relationship characteristics will largely affect whether the company will have overdue behavior in the future, and ultimately lead to large errors in the results of the model prediction.

[0056] In order to solve the technical problems existing in the prior art, an embodiment of the present disclosure provides a risk prediction method, including obtaining enterprise credit data; converting the enterprise credit data into a feature data set; extracting enterprise potential expressions in the feature data set; extracting enterprise relationship potential expressions in the feature data set; and predicting the target enterprise credit risk through a preset machine learning model based on the enterprise potential expressions and the enterprise relationship potential expressions.

[0057] In the embodiments of the present disclosure, in order to improve the accuracy of risk prediction, the embodiments of the present disclosure respectively extract the potential expression of the enterprise itself and the potential expression of the enterprise relationship between enterprises through two deep learning models, which can parse the potential information of the enterprise and the potential information between enterprise relationships, and integrate these two expressions carrying potential information to comprehensively predict the credit risk of the target enterprise. The embodiments of the present disclosure can capture the complex potential dependencies of enterprises and the multiple correlations of risks between enterprises, greatly improving the ability of enterprises to determine whether high credit scores will occur in the future.

[0058] Figure 1 The application scenario diagram of the risk prediction method according to the embodiment of the present disclosure is schematically shown.

[0059] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0060] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0061] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0062] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0063] It should be noted that the risk prediction method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the risk prediction device provided in the embodiment of the present disclosure can generally be set in the server 105. The risk prediction method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the risk prediction device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0064] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0065] The following will be based on Figure 1 The scene described by Figure 2~Figure 9 The risk prediction method of the disclosed embodiment is described in detail.

[0066] Figure 2 The flowchart of the risk prediction method according to the embodiment of the present disclosure is schematically shown.

[0067] like Figure 2 As shown, the risk prediction method of this embodiment includes operations S210 to S250, and the risk prediction method can be executed by the server 105.

[0068] In operation S210, enterprise credit data is acquired.

[0069] In an embodiment of the present disclosure, before obtaining the user's information, the user's consent or authorization may be obtained. For example, before operation S210, a request to obtain the user's information may be issued to the user. When the user agrees or authorizes that the user's information may be obtained, the operation S210 is performed.

[0070] Among them, corporate credit data at least contains macro information and micro information for the enterprise. At the macro level, macro information such as key industry development strategies, social public opinion and stock prices are selected for subsequent feature extraction. At the micro level, micro information such as basic information of corporate customers, customer potential, product holdings, risks, financial statements and loan information is used for subsequent feature extraction.

[0071] In operation S220, the enterprise credit data is converted into a feature data set.

[0072] In the embodiments of the present disclosure, a corresponding operation entry may be provided for the user to choose to agree or reject the automated decision result. That is, before the risk prediction processing / decision is performed on the user information, an instruction to agree or reject the processing / decision input by the user through the corresponding operation entry may be obtained. If the user agrees to perform the processing / decision, the risk prediction processing / decision is performed on the user information, that is, step S220 is executed. If the user refuses to perform the processing / decision, the expert decision process is entered.

[0073] Specifically, data preprocessing operations, feature creation operations, and feature screening operations are performed on the enterprise credit data to convert the enterprise credit data into a feature data set. The feature data set is not used for direct credit prediction later, but for extracting potential information based on the feature data set.

[0074] The following is a detailed description of the method for establishing the feature dataset, as follows:

[0075] Figure 3 The flowchart of the method for establishing a feature data set according to an embodiment of the present disclosure is schematically shown.

[0076] Figure 6 A flow chart of a method for establishing data features according to an embodiment of the present disclosure is schematically shown.

[0077] like Figure 3 As shown, the method for establishing a feature data set in this embodiment includes operations S310 to S330, which are specifically as follows:

[0078] In operation S310, the enterprise credit data is preprocessed to obtain standard credit data.

[0079] According to an embodiment of the present disclosure, the preprocessing of the enterprise credit data to obtain standard credit data includes: deleting abnormal values ​​and null values ​​in the enterprise credit data to obtain first enterprise credit data; and randomly sampling the first enterprise credit data to obtain the standard credit data.

[0080] Specifically, outliers refer to data points that deviate significantly from other values ​​in the data set, and null values ​​refer to missing data points in the data set. Outliers and controls are deleted, and the remaining data is retained as standard credit data.

[0081] In the embodiment of the present disclosure, the loan delinquency prediction modeling problem is a long-term time series prediction scenario, that is, in the process of customer loan financing and credit approval, based on the customer's historical behavior, predict whether a loan will be overdue in the next year. Among them, customers who have loan overdue are defined as negative samples, and corporate customers who will not have loan overdue behavior in the next year are defined as positive samples. Since there is a significant difference in the proportion of positive and negative samples in the loan overdue scenario, in order to better model the inherent dependence of the scenario, a random sampling method is used to increase the concentration of negative samples. And based on the historical sequence distribution, features are established.

[0082] In operation S320, an intermediate feature data set is established based on the standard credit data.

[0083] According to an embodiment of the present disclosure, establishing an intermediate feature data set based on the standard credit data includes: extracting target dimension features in the standard credit data to obtain a basic feature set; forming a derived feature set based on the basic feature set; and converting the basic feature set and the derived feature set into the form of a time series to obtain the intermediate feature data set.

[0084] Specifically, combined Figure 6 As shown, a large number of features are established in advance, and the internal factors that affect the problem are modeled as much as possible, starting from both macro and micro aspects. At the macro level, the characteristics of key industry development strategies, social public opinion and stock prices are selected. At the micro level, the basic feature set is based on dimensions such as basic information of corporate customers, customer potential, product holdings, risks, financial statements and loan information.

[0085] In addition to basic features, feature derivation is also an important part of feature engineering. This is because the basic information of the customer's assets and loan amount over a period of time cannot fully reflect the customer's behavior pattern and financial risk. The risk of whether the customer has overdue behavior depends more on the changes in behavior or operating characteristics over a certain period of time. Derivative features such as the company's historical long-term operating benefits, revenue-expenditure ratio, and short-term financial fluctuations are calculated to further improve the model's predictive ability for risk overdue scenarios in loans. Not only that, feature derivation can also effectively improve the model's generalization ability and can be used for this modeling. As shown in Table 1 below, Table 1 schematically shows the construction method of derived features:

[0086] Table 1

[0087] Build method illustrate Targeting risk points Find the proportion Calculate the proportion of income and expenditure in different time windows and different combination dimensions to the total amount of transactions of this type, such as the cumulative number of payments / cumulative number of income in the past 30 days and other derivative features Performance Ability Find the variance Calculate the variance of daily transaction amount and number of transactions in different combination dimensions over a period of time, such as the variance of daily transaction number within 90 days Stable situation combination Combine certain features, such as: the stock price of a certain customer's industry has experienced drastic changes, and the number of transactions in the past month has dropped sharply compared to the previous month. Used to identify certain high-risk groups Automatic extraction of time series features According to the time dimension, features such as the difference in the number of transactions this month compared to last month are derived based on customer transaction information data and customer balance information data. Features such as the difference in the account balance this month compared to last month are derived based on customer account balance information. Historical changes

[0088] In operation S330, based on a dual-track screening method, some features in the intermediate feature data set are screened as feature data sets.

[0089] Specifically, the dual-track screening method uses two different screening rules / screening models to screen the feature data in the intermediate feature data set, which has achieved the purpose of data mining exploration. For example, multiple methods are used to screen features from different dimensions, including IV information volume and tree model feature importance analysis methods, to fully explore and mine the relationship between basic data features and others, in order to find out the real reasons that affect customer loan delinquency behavior. The dual-track screening method is as follows:

[0090] Figure 4 The flowchart of the dual-track screening method according to an embodiment of the present disclosure is schematically shown.

[0091] Figure 7 A flowchart of a data feature screening method according to an embodiment of the present disclosure is schematically shown.

[0092] like Figure 4 As shown, the dual-track screening method of this embodiment includes operations S410 to S430, which are specifically as follows:

[0093] In operation S410, some features of the intermediate feature data set are screened as a first feature data set based on an information value feature screening model.

[0094] Among them, the information value feature screening model is a method of screening through IV value. It can evaluate each feature data in the intermediate feature data set and screen out the feature data that meets the standards.

[0095] According to an embodiment of the present disclosure, the information value feature screening model is based on screening some features of the intermediate feature data set as the first feature data set, including: binning the feature data in the intermediate feature data set by evidence weight to obtain N bins of evidence weights, where N is a positive integer; calculating N information values ​​based on the evidence weights of the N bins; and retaining the feature data of the N information values ​​that is greater than a first preset threshold to obtain the first feature data set.

[0096] Specifically, combined Figure 7 As shown, for the intermediate feature data set containing basic features and derived features, the IV information method is used to screen and retain the main features according to the specified threshold. The information screening method is: WOE binning of each variable, and then calculating the IV value related to it. Among them, for each variable, the calculation of each group of woe values ​​is shown in the following formula 1:

[0097] Formula 1

[0098] in, is the total number of bad samples, is the total number of good samples, is a bad sample and X is the number of the i-th group, is a good sample and X is the number of the i-th group.

[0099] The IV value is calculated as shown in formula 2:

[0100] Formula 2

[0101] After calculating the IV values ​​of each group of a variable, the IV value of the entire variable can be calculated as shown in Formula 3 below:

[0102] Formula 3

[0103] Since the IV value indicates the predictive ability of the variable, the value of the IV value is closely related to the predictive ability. The evaluation criteria of the IV value are shown in Table 2 below:

[0104] Table 2

[0105] IV value [0.02,0.1) [0.1,0.3) [0.3,0.5) [0.5,+∞) The predictive power of the representative Low middle high Very high

[0106] In the embodiment of the present disclosure, the threshold value may be set to 0.02, IV values ​​are calculated for all features, features with IV values ​​less than 0.02 are removed, and the remaining ones are retained.

[0107] In operation S420, some features of the intermediate feature data set are screened based on a preset decision tree model as a second feature data set.

[0108] According to an embodiment of the present disclosure, the method of screening some features of the intermediate feature data set as the second feature data set based on a preset decision tree model includes: calculating the ranking of feature data in the intermediate feature data set based on the preset decision tree model to obtain K rankings, where K is a positive integer; and retaining feature data in the K rankings that is greater than a second preset threshold to obtain the second feature data set.

[0109] Specifically, combined Figure 7As shown, in addition to mathematical statistical analysis methods, data intelligence thinking can also be used to define the problem between features and prediction targets as a binary classification problem. By introducing a decision tree (such as the LightGBM model) algorithm, the trained model is directly used to output the importance of features, and a threshold is set to remove unimportant features. High feature correlation will waste computing resources and affect the interpretability of the model. The Pearson correlation coefficient and collinearity diagnosis are used to weaken the correlation between features. Since the embodiments of the present disclosure are essentially long-term time series prediction problems, and time series problems are related to historical moments, in addition to completing feature screening, it is also necessary to use data values ​​from historical moments in recent days, weeks, and months according to the time period dimension as support for the model in the time dimension, and as the basis for constructing multi-dimensional time series features.

[0110] In operation S430, a union of the first feature data set and the second feature data set is taken as the feature data set.

[0111] Specifically, combined Figure 7 As shown, the features retained in two different ways are combined to form a feature data set.

[0112] In operation S230, potential expressions of the enterprise in the feature dataset are extracted.

[0113] According to an embodiment of the present disclosure, the extracting the potential expression of the enterprise from the feature data set includes: using the feature data set as input data of a preset recurrent neural network model, and outputting the potential expression of the enterprise.

[0114] Specifically, through the above operations S410 to S430, the main characteristic factors that affect whether the enterprise will have overdue behavior in the future are obtained. In order to fully learn the potential impact of the established characteristics on the enterprise's overdue behavior, a deep learning algorithm recurrent neural network model can be introduced, such as the GRU model. Compared with traditional machine learning algorithms, GRU has a more powerful complex dependency modeling capability. Compared with convolutional neural networks such as CNN, it is more suitable for capturing complex time dependencies in time series prediction problems. Compared with other recurrent neural network variants such as LSTM, it can better alleviate the problem of gradient explosion and improve model training efficiency.

[0115] Specifically, define the input features , put in In the deep GRU network obtained by stacking L layers of GRU, after being processed by the L-layer GRU network, a highly modeled latent feature expression is obtained. , which is the potential expression of the enterprise. It is shown in the following equations 4 to 7:

[0116] ) Formula 4

[0117] Formula 5

[0118] Formula 6

[0119] Formula 7

[0120] in, for Input features at time , The value of , is a hyperparameter.

[0121] In operation S240, potential expressions of enterprise relationships in the feature dataset are extracted.

[0122] Specifically, massive features are established based on the company's historical data, and the GRU algorithm is used to capture complex potential dependencies to model the possibility of overdue credit risk of the company. However, the historical behavioral characteristics of the company cannot fully reflect whether the company will be overdue in the future. This is because the company has a variety of complex relationships, such as the company's loan guarantees, capital flows, and equity penetration relationships. The complex relationships of the company will affect the company's ability to repay in the future, and how to effectively extract the potential feature expressions implied in the relationship graph is a difficulty in predicting whether future customers will have overdue behavior. For this reason, the inventor has designed a relationship representation method that embeds multiple graph relationships to model the impact of potential relationships on the future overdue behavior of the target company. The details are as follows:

[0123] Figure 5 The flowchart of the potential relationship extraction method according to the embodiment of the present disclosure is schematically shown.

[0124] Figure 8 The flowchart of the relationship characterization method according to the embodiment of the present disclosure is schematically shown.

[0125] like Figure 5 As shown, the potential relationship extraction method of this embodiment includes operations S510 to S520, which are specifically as follows:

[0126] In operation S510, an enterprise multi-relationship graph is established based on the feature data set.

[0127] According to an embodiment of the present disclosure, establishing an enterprise multi-relationship graph based on the feature data set includes: determining the enterprise guarantee relationship, equity penetration relationship and capital flow relationship based on the feature data set; taking any enterprise in the feature data set as a node, and forming the enterprise multi-relationship graph through the guarantee relationship, the equity penetration relationship and the capital flow relationship.

[0128] Specifically, combined Figure 7 As shown, for the loan guarantee relationship, the embodiment of the present disclosure includes the enterprises with which the target enterprise has a mortgage guarantee relationship into the nodes of the multi-graph relationship, and records the guarantee relationship between the enterprises as a relationship ; For equity penetration, we focus on the upstream and downstream links of the target corporate legal person, extract data such as the proportion of equity holdings and the number of capital transactions, and comprehensively select The enterprise is included in the multi-relationship graph and is represented as a vertex in the graph with equity penetration relationship and is represented as ; For the relationship between capital flow , set the threshold , and use the following calculation formula to define whether there is an inflow and outflow relationship of capital flow between enterprises, and include enterprises with capital flow relationship into the multi-graph relationship. The calculation method of is shown in the following formula 8:

[0129] Formula 8

[0130] In addition, in the embodiments of the present disclosure, basic characteristics such as the company's assets, number of financings, amount, customer star rating, and industry can be extracted together with the company's cumulative number and amount of guarantee relationships, number and amount of inflow and outflow relationships, equity penetration quantity and level in the constructed multi-relationship graph as the company's own attributes in the multi-relationship graph.

[0131] In operation S520, based on the enterprise multi-relationship graph, the enterprise relationship potential expressions are extracted by a graph sample aggregation algorithm.

[0132] According to an embodiment of the present disclosure, based on the enterprise multi-relationship graph, the potential expression of the enterprise relationship is extracted through a graph sample aggregation algorithm, including: calculating the potential expression of the enterprise relationship between the target node and the domain node based on the guarantee relationship, the equity penetration relationship, the capital flow relationship and the corresponding preset weights.

[0133] Specifically, combined Figure 7 As shown in the figure, through the above data processing, a complex multi-relationship graph between enterprises is modeled and the attributes of the enterprises themselves are integrated into the graph. At the same time, an inductive graph representation learning model can be introduced, such as the GraghSAGE algorithm. The core idea of ​​the algorithm is to generate the potential relationship expression of the target node by learning the aggregation function between the neighborhood nodes and the target node. The calculation process in the GraghSAGE algorithm is obtained by repeated iterative calculation based on the selected k-order neighbors. The calculation process of each order remains the same. The calculation formula for the v-th vertex of the k-th order is shown in the following formulas 9-11:

[0134] Formula 9

[0135] Formula 10

[0136] Formula 11

[0137] It is worth noting that since there are multiple relationships in the established multi-relationship graph, the original GraghSAGE algorithm calculation process needs to be optimized in the embodiment of the present disclosure. When using the aggregation function to calculate and extract neighborhood information, different relationships will be extracted and calculated separately. , and in the next step, the potential expression is calculated and generated, different weights are assigned to different relationships to capture the influence of different relationships, and the potential relationship expression vector of the target node is generated. The optimized formula is shown in Equation 12-Equation 15:

[0138] Formula 12

[0139] Formula 13

[0140] Formula 14

[0141] Formula 15

[0142] After repeating the calculation K times, the final target enterprise relationship potential expression is obtained .

[0143] In operation S250, based on the enterprise potential expression and the enterprise relationship potential expression, the credit risk of the target enterprise is predicted by a preset machine learning model.

[0144] Fig. 9 The flowchart of the feature fusion prediction method according to an embodiment of the present disclosure is schematically shown.

[0145] Specifically, combined Fig. 9 As shown in Figure 2, highly nonlinear potential expressions are obtained by learning the feature datasets through deep GRU and improved GraphSAGE. Then, they are stacked and input into the fully connected neural network to get the output of the model. The output of the model is mapped to the range of 0 to 1 through the sigmoid activation function, and finally the probability value of the target enterprise's overdue payment and whether the overdue behavior occurs are output, as shown in the following formulas 16 and 17:

[0146] Formula 16

[0147] Formula 17

[0148] The embodiments of the present disclosure design and construct a multi-relationship graph of an enterprise based on modeling the enterprise's own financial status, default, credit, assets and other dimensional characteristics, form the enterprise's association network, and use the GraghSAGE algorithm to capture the enterprise's potential association characteristics. At the same time, in order to fully model the impact of various dimensional characteristics on the enterprise's future overdue behavior, a deep GRU network is established and used to obtain potential feature expressions. Finally, the deep potential association relationship expression and the potential feature expression are stacked and input into the fully connected neural network together to obtain an accurate prediction result of whether the enterprise will be overdue in the future.

[0149] In the embodiments of the present disclosure, in order to improve the accuracy of risk prediction, the embodiments of the present disclosure respectively extract the potential expression of the enterprise itself and the potential expression of the enterprise relationship between enterprises through two deep learning models, which can parse the potential information of the enterprise and the potential information between enterprise relationships, and integrate these two expressions carrying potential information to comprehensively predict the credit risk of the target enterprise. The embodiments of the present disclosure can capture the complex potential dependencies of enterprises and the multiple correlations of risks between enterprises, greatly improving the ability of enterprises to determine whether high credit scores will occur in the future.

[0150] Based on the above risk prediction method, the present disclosure also provides a risk prediction device. Fig.10 The device is described in detail.

[0151] Fig.10 The structural block diagram of the risk prediction device according to an embodiment of the present disclosure is schematically shown.

[0152] like Fig.10 As shown, the risk prediction device 1000 of this embodiment includes a credit data acquisition module 1010 , a feature conversion module 1020 , a feature learning module 1030 , a relationship representation module 1040 and a model prediction module 1050 .

[0153] The credit data acquisition module 1010 is used to acquire the enterprise credit data. In one embodiment, the credit data acquisition module 1010 can be used to perform the operation S210 described above, which will not be described in detail here.

[0154] The feature conversion module 1020 is used to convert the enterprise credit data into a feature data set. In one embodiment, the feature conversion module 1020 can be used to perform the operation S220 described above, which will not be described in detail here.

[0155] The feature learning module 1030 is used to extract the enterprise potential expressions in the feature data set. In one embodiment, the feature learning module 1030 can be used to perform the operation S230 described above, which will not be described in detail here.

[0156] The relationship representation module 1040 is used to extract the potential expressions of the enterprise relationship in the feature data set. In one embodiment, the relationship representation module 1040 can be used to perform the operation S240 described above, which will not be described in detail here.

[0157] The model prediction module 1050 is used to predict the target enterprise credit risk through a preset machine learning model based on the enterprise potential expression and the enterprise relationship potential expression. In one embodiment, the model prediction module 1050 can be used to perform the operation S250 described above, which will not be repeated here.

[0158] In the embodiments of the present disclosure, in order to improve the accuracy of risk prediction, the embodiments of the present disclosure respectively extract the potential expression of the enterprise itself and the potential expression of the enterprise relationship between enterprises through two deep learning models, which can parse the potential information of the enterprise and the potential information between enterprise relationships, and integrate these two expressions carrying potential information to comprehensively predict the credit risk of the target enterprise. The embodiments of the present disclosure can capture the complex potential dependencies of enterprises and the multiple correlations of risks between enterprises, greatly improving the ability of enterprises to determine whether high credit scores will occur in the future.

[0159] According to an embodiment of the present disclosure, the feature conversion module includes: a preprocessing unit, an intermediate data establishment unit and a screening unit, wherein the preprocessing unit is used to preprocess the enterprise credit data to obtain standard credit data; the intermediate data establishment unit is used to establish an intermediate feature data set based on the standard credit data; and the screening unit is used to screen some features in the intermediate feature data set as a feature data set based on a dual-track screening method.

[0160] According to an embodiment of the present disclosure, the preprocessing unit is specifically used to delete abnormal values ​​and null values ​​in the enterprise credit data to obtain first enterprise credit data; and randomly sample the first enterprise credit data to obtain the standard credit data.

[0161] According to an embodiment of the present disclosure, the intermediate data establishment unit is specifically used to extract target dimension features in the standard credit data to obtain a basic feature set; form a derived feature set based on the basic feature set; and convert the basic feature set and the derived feature set into the form of a time series to obtain the intermediate feature data set.

[0162] According to an embodiment of the present disclosure, the screening unit includes: an information value screening subunit, a decision tree screening subunit and a feature data set forming subunit, the information value screening subunit is used to screen some features of the intermediate feature data set as the first feature data set based on the information value feature screening model; the decision tree screening subunit is used to screen some features of the intermediate feature data set as the second feature data set based on a preset decision tree model; and the feature data set forming subunit is used to take the union of the first feature data set and the second feature data set as the feature data set.

[0163] According to an embodiment of the present disclosure, the information value screening subunit is specifically used to perform evidence weight binning on the feature data in the intermediate feature data set to obtain evidence weights of N bins, where N is a positive integer; calculate N information values ​​based on the evidence weights of the N bins; and retain the feature data among the N information values ​​that is greater than a first preset threshold to obtain the first feature data set.

[0164] According to an embodiment of the present disclosure, the decision tree screening subunit is specifically used to calculate the ranking of feature data in the intermediate feature data set based on the preset decision tree model to obtain K rankings, where K is a positive integer; and retain the feature data in the K rankings that is greater than a second preset threshold to obtain the second feature data set.

[0165] According to an embodiment of the present disclosure, the feature learning module is specifically used to use the feature data set as input data of a preset recurrent neural network model, and output the potential expression of the enterprise.

[0166] According to an embodiment of the present disclosure, the relationship representation module includes: a relationship graph establishment unit and a potential relationship extraction unit, the relationship graph establishment unit is used to establish an enterprise multi-relationship graph based on the feature data set; and the potential relationship extraction unit is used to extract the enterprise relationship potential expression through a graph sample aggregation algorithm based on the enterprise multi-relationship graph.

[0167] According to an embodiment of the present disclosure, the relationship graph establishing unit is specifically used to determine the enterprise guarantee relationship, equity penetration relationship and capital flow relationship based on the feature data set; any enterprise in the feature data set is used as a node to form the enterprise multi-relationship graph through the guarantee relationship, the equity penetration relationship and the capital flow relationship.

[0168] According to an embodiment of the present disclosure, the potential relationship extraction unit is specifically used to calculate the potential expression of the enterprise relationship between the target node and the domain node based on the guarantee relationship, the equity penetration relationship, the capital flow relationship and the corresponding preset weights.

[0169] According to an embodiment of the present disclosure, any multiple modules of the credit data acquisition module 1010, the feature conversion module 1020, the feature learning module 1030, the relationship representation module 1040, and the model prediction module 1050 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the credit data acquisition module 1010, the feature conversion module 1020, the feature learning module 1030, the relationship representation module 1040, and the model prediction module 1050 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the credit data acquisition module 1010, the feature conversion module 1020, the feature learning module 1030, the relationship representation module 1040 and the model prediction module 1050 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is executed.

[0170] Fig.11 A block diagram of an electronic device suitable for implementing a risk prediction method according to an embodiment of the present disclosure is schematically shown.

[0171] like Fig.11 As shown, the electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage part 1108 to a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include an onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0172] In RAM 1103, various programs and data required for the operation of electronic device 1100 are stored. Processor 1101, ROM 1102 and RAM 1103 are connected to each other through bus 1104. Processor 1101 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 1102 and / or RAM 1103. It should be noted that the program can also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0173] According to an embodiment of the present disclosure, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to the bus 1104. The electronic device 1100 may further include one or more of the following components connected to the input / output (I / O) interface 1105: an input portion 1106 including a keyboard, a mouse, etc.; an output portion 1107 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 1108 including a hard disk, etc.; and a communication portion 1109 including a network interface card such as a LAN card, a modem, etc. The communication portion 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed, so that a computer program read therefrom is installed into the storage portion 1108 as needed.

[0174] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0175] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus, or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 1102 and / or RAM 1103 described above and / or one or more memories other than ROM 1102 and RAM 1103.

[0176] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present disclosure.

[0177] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 1101. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0178] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 1109, and / or installed from a removable medium 1111. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0179] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0180] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0181] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0182] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0183] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A risk prediction method, characterized in that: The method comprises: Obtain corporate credit data; Converting the enterprise credit data into a feature data set; Extracting enterprise potential expressions from the feature data set; Extracting potential expressions of enterprise relationships in the feature data set; and Based on the enterprise potential expression and the enterprise relationship potential expression, the credit risk of the target enterprise is predicted by a preset machine learning model.

2. The method according to claim 1, characterized in that The converting the enterprise credit data into a feature data set includes: Preprocessing the enterprise credit data to obtain standard credit data; Establishing an intermediate feature data set based on the standard credit data; and Based on the dual-track screening method, some features in the intermediate feature data set are screened as the feature data set.

3. The method according to claim 2, characterized in that The preprocessing of the enterprise credit data to obtain standard credit data includes: Deleting abnormal values ​​and null values ​​in the enterprise credit data to obtain first enterprise credit data; and The first enterprise credit data is randomly sampled to obtain the standard credit data.

4. The method according to claim 2, characterized in that: The establishing of an intermediate feature data set based on the standard credit data comprises: Extracting target dimension features from the standard credit data to obtain a basic feature set; forming a derived feature set based on the base feature set; and The basic feature set and the derived feature set are converted into a time series form to obtain the intermediate feature data set.

5. The method according to claim 4, characterized in that The dual-track screening method, screening some features in the intermediate feature data set as the feature data set, includes: Based on the information value feature screening model, some features of the intermediate feature data set are screened as the first feature data set; Filtering some features of the intermediate feature data set as a second feature data set based on a preset decision tree model; and A union of the first feature data set and the second feature data set is taken as the feature data set.

6. The method according to claim 5, characterized in that The step of selecting some features of the intermediate feature data set as the first feature data set based on the information value feature selection model includes: Binning the feature data in the intermediate feature data set with evidence weights to obtain evidence weights of N bins, where N is a positive integer; Calculating N information values ​​based on the evidence weights of the N bins; and The feature data whose values ​​are greater than a first preset threshold value among the N information values ​​are retained to obtain the first feature data set.

7. The method according to claim 5, characterized in that The screening of some features of the intermediate feature data set as the second feature data set based on a preset decision tree model includes: Calculating the ranking of feature data in the intermediate feature data set based on the preset decision tree model to obtain K rankings, where K is a positive integer; and The feature data that is greater than a second preset threshold in the K rankings is retained to obtain the second feature data set.

8. The method according to any one of claims 1 to 7, characterized in that: The step of extracting the enterprise potential expressions from the feature data set comprises: The feature data set is used as input data of a preset recurrent neural network model, and the potential expression of the enterprise is obtained as output.

9. The method according to any one of claims 1 to 7, characterized in that: The step of extracting potential expressions of enterprise relationships from the feature data set includes: Establishing an enterprise multi-relationship graph based on the feature data set; and Based on the enterprise multi-relationship graph, the potential expressions of the enterprise relationships are extracted through a graph sample aggregation algorithm.

10. The method according to claim 9, characterized in that The step of establishing an enterprise multi-relationship graph based on the feature data set includes: Determine the enterprise guarantee relationship, equity penetration relationship and capital flow relationship based on the feature data set; Any enterprise in the feature data set is taken as a node, and the enterprise multi-relationship graph is formed through the guarantee relationship, the equity penetration relationship and the capital flow relationship.

11. The method according to claim 10, characterized in that The step of extracting the enterprise relationship potential expressions based on the enterprise multi-relationship graph by using a graph sample aggregation algorithm includes: The potential expression of the enterprise relationship between the target node and the domain node is calculated based on the guarantee relationship, the equity penetration relationship, the capital flow relationship and the corresponding preset weights.

12. A risk prediction device, characterized in that: The device comprises: Credit data acquisition module, used to obtain enterprise credit data; A feature conversion module, used to convert the enterprise credit data into a feature data set; A feature learning module, used for extracting enterprise potential expressions from the feature data set; A relationship representation module, used to extract potential expressions of enterprise relationships in the feature data set; and The model prediction module is used to predict the credit risk of the target enterprise through a preset machine learning model based on the enterprise potential expression and the enterprise relationship potential expression.

13. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

15. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images