Credit risk assessment method, device and equipment for small and medium-sized enterprises and storage medium

Abstract

The invention discloses a middle and small-sized enterprise credit risk assessment method and device, equipment and a storage medium, and the method comprises the steps: obtaining the multi-source associated data of a middle and small-sized enterprise, and constructing a financial credit risk association map of the middle and small-sized enterprise; extracting credit risk association map embedding indexes from the financial credit risk association maps of the small and medium-sized enterprises; extracting industry and commerce information risk indexes from the industry and commerce registration data; extracting an enterprise historical loan information risk index from the loan business contract information data; and inputting the credit risk association graph embedding index, the industry and commerce information risk index and the enterprise historical loan information risk index into a credit risk assessment model for credit risk assessment to obtain a credit risk assessment result of the small and medium-sized enterprises. The technical problem of low accuracy of credit risk assessment of small and medium-sized enterprises due to insufficient company audit and financial data of an existing model is solved, and particularly, a reliable assessment effect is achieved for the small and medium-sized enterprises lacking financial data and historical credit data.

Description

Technical Field

[0001] The present application relates to the technical field of credit risk assessment, and in particular to a method, device, equipment and storage medium for credit risk assessment of small and medium-sized enterprises. Background Art

[0002] Small and medium-sized enterprises have long faced financing difficulties, and their financing difficulties are accompanied by the risk control problems of financial institutions. The financing risks of small and medium-sized enterprises are difficult to assess, mainly for the following reasons:

[0003] (1) Poor credit data quality and little historical data. SMEs lack effective regulatory systems and strict information disclosure requirements. Financial institutions do not have much assessment data to collect when approving their loans. Most SMEs lack mature and complete financial management systems, so the credit assessment model based on traditional financial data is no longer applicable. Financial institutions also find it difficult to verify the accuracy and authenticity of data proactively reported by enterprises. In addition, the average lifespan of SMEs is short and there is too little historical data, which makes it difficult to build an assessment model.

[0004] (2) The dynamic impact of risk is complex. In addition to its internal operating conditions and willingness to repay, the credit risk of SMEs is also related to the external macro-environment and contagion effect. Compared with large enterprises, SMEs have lower risk resistance, which leads to the spread of risks among SMEs. The contagion effect is often ignored in traditional models, and the root cause of contagion is the complex relationship between enterprises.

[0005] The traditional corporate financial risk assessment model based on company audit and financial data is only applicable to large companies with standardized financial systems. However, the credit data quality of small and medium-sized enterprises is poor and there is little historical data, so the effectiveness of this assessment model is greatly limited. Summary of the invention

[0006] The present application provides a method, device, equipment and storage medium for evaluating the credit risk of small and medium-sized enterprises, which are used to improve the technical problem that the existing model has low accuracy in evaluating the credit risk of small and medium-sized enterprises due to insufficient company audit and financial data.

[0007] In view of this, the first aspect of the present application provides a method for assessing the credit risk of small and medium-sized enterprises, comprising:

[0008] Acquire multi-source related data of small and medium-sized enterprises, wherein the multi-source related data of small and medium-sized enterprises includes business registration data, shareholder and executive data, judicial litigation data, and loan business contract information data of small and medium-sized enterprises;

[0009] Constructing a financial credit risk association map for SMEs through the multi-source association data of SMEs;

[0010] Extracting graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators;

[0011] Extracting business information risk indicators from the business registration data;

[0012] Extracting the enterprise's historical loan information risk index from the loan business contract information data;

[0013] The credit risk association graph embedding index, the industrial and commercial information risk index and the enterprise historical loan information risk index are input into a credit risk assessment model to perform credit risk assessment and obtain credit risk assessment results for small and medium-sized enterprises.

[0014] Optionally, the SME financial credit risk association graph includes a SME industrial and commercial social network subgraph and a SME lending business network subgraph;

[0015] The entities in the SME business social network subgraph include enterprises and natural persons, and the relationships between entities include directors, supervisors, managers, investors and legal persons;

[0016] The entities in the SME lending business network subgraph include enterprises and lending contracts, and the relationships between entities include borrowing and lending.

[0017] Optionally, the step of extracting graph embedding features from the SME financial credit risk association graph to obtain a credit risk association graph embedding index includes:

[0018] Extracting centrality measures of borrowing enterprises from the SME financial credit risk association graph, wherein the centrality measures include degree centrality, betweenness centrality, closeness centrality and PageRank centrality;

[0019] Based on the graph calculation method, the number of small loan platforms with industrial and commercial association paths within 6 degrees of the borrowing enterprise and the length of the industrial and commercial association path of the small loan platform closest to the borrowing enterprise are extracted from the SME financial credit risk association graph;

[0020] Node representation vectors are extracted from the SME financial credit risk association graph based on the Node2vec algorithm or graph convolutional neural network.

[0021] Optionally, the business information risk indicators include: the type of borrowing enterprise, the industry type of the borrowing enterprise, and the industry to which the borrowing enterprise belongs.

[0022] Optionally, the enterprise's historical loan information risk indicators include: the natural logarithm of the last loan amount, the annualized interest rate of the last loan, whether the last loan was in default, whether the last loan was overdue, the number of days between the last loan and the current loan, the ratio of the current loan amount to the historical average amount, the number of historical borrowings, and the number of days between the first loan and the current loan.

[0023] Optionally, the method further includes:

[0024] extracting corporate judicial risk indicators from the judicial litigation data;

[0025] The credit risk association graph embedding indicator, the industrial and commercial information risk indicator and the enterprise historical loan information risk indicator are input into the credit risk assessment model to perform credit risk assessment, and obtain the credit risk assessment results of small and medium-sized enterprises, including:

[0026] The credit risk association map embedding index, the industrial and commercial information risk index, the enterprise historical loan information risk index and the enterprise judicial risk index are input into the credit risk assessment model to conduct credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

[0027] Optionally, the corporate judicial risk indicators include: whether the corporate entity was an abnormal taxpayer within one year before the date of this loan, whether it owed taxes, whether it was restricted from high consumption, whether it was identified as a deadbeat, whether it was publicly executed, and the ratio of the current loan amount to the total amount of execution against the corporate entity within one year.

[0028] Optionally, the training process of the credit risk assessment model includes:

[0029] Extracting risk indicators of small and medium-sized enterprises with known credit risk assessment results to obtain a data set, wherein the risk indicators include credit risk association graph embedding indicators, industrial and commercial information risk indicators, and enterprise historical loan information risk indicators;

[0030] Dividing the data set into a training set and a test set, training the machine learning model with the training set, and testing the trained machine learning model with the test set until the machine learning model converges to obtain a trained machine learning model;

[0031] A credit risk assessment model is obtained based on the trained machine learning model.

[0032] Optionally, the number of the machine learning models is 1 or more;

[0033] The step of obtaining a credit risk assessment model based on the trained machine learning model includes:

[0034] When the number of the machine learning models is 1, using the trained machine learning model as a credit risk assessment model;

[0035] When there are multiple machine learning models, the machine learning model with the highest evaluation accuracy is selected from the multiple trained machine learning models as the credit risk assessment model, or multiple trained machine learning models are used as credit risk assessment models, and the average value or voting result of the credit risk assessment results output by the multiple trained machine learning models is used as the final credit risk assessment result.

[0036] The second aspect of the present application provides a credit risk assessment device for small and medium-sized enterprises, comprising:

[0037] A data acquisition unit, used to acquire multi-source related data of small and medium-sized enterprises, wherein the multi-source related data of small and medium-sized enterprises includes business registration data, shareholder and executive data, judicial litigation data, and loan business contract information data of small and medium-sized enterprises;

[0038] A graph construction unit, used to construct a SME financial credit risk association graph through the SME multi-source association data;

[0039] A risk indicator extraction unit is used to extract graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators; extract business information risk indicators from the business registration data; and extract enterprise historical loan information risk indicators from the loan business contract information data;

[0040] The risk assessment unit is used to input the credit risk association map embedding index, the industrial and commercial information risk index and the enterprise historical loan information risk index into the credit risk assessment model to perform credit risk assessment and obtain the credit risk assessment result of small and medium-sized enterprises.

[0041] A third aspect of the present application provides an electronic device, the device comprising a processor and a memory;

[0042] The memory is used to store program code and transmit the program code to the processor;

[0043] The processor is used to execute any one of the small and medium-sized enterprise credit risk assessment methods of the first aspect according to the instructions in the program code.

[0044] A fourth aspect of the present application provides a computer-readable storage medium, which is used to store program codes. When the program codes are executed by a processor, the method for assessing the credit risk of small and medium-sized enterprises described in any one of the first aspects is implemented.

[0045] It can be seen from the above technical solutions that this application has the following advantages:

[0046] The credit risk assessment method for SMEs provided in this application constructs a credit risk association map based on the risk contagion theory, supplements the risk information of SMEs with cross-sectional risk association information, and solves the information asymmetry problem from another perspective, so as to improve the technical problem that the credit risk assessment of SMEs is not accurate due to insufficient data.

[0047] Furthermore, this application introduces risk association graphs and complex network theory, utilizes the public information and short-term information that are highly accessible to SMEs, identifies and models credit risks from the level of complex association structures, and thus constructs a reliable SME credit risk assessment model;

[0048] Furthermore, this application uses multiple indicators such as credit risk association graph embedding indicators, industrial and commercial information risk indicators, and enterprise historical loan information risk indicators to assess the credit risk of SMEs, considering the impact of multiple factors on the credit risk of SMEs from different perspectives, which helps to improve the accuracy of credit risk assessment of SMEs;

[0049] Furthermore, this application is based on the idea of ​​risk modeling of complex network structures, utilizing the commonalities in the local neighborhood topological structure of nodes to achieve the identification and evaluation of risk nodes, and further explore the complex dependencies between enterprises and potential risk transmission paths, thereby achieving a good modeling of the global financial risk contagion phenomenon, which helps to improve the accuracy of credit risk assessment for small and medium-sized enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A flow chart of a method for assessing the credit risk of small and medium-sized enterprises provided in an embodiment of the present application;

[0052] Figure 2 The financial risk association map and credit assessment model framework for small and medium-sized enterprises provided in the embodiments of this application;

[0053] Figure 3 A graph of the financial credit risk association of small and medium-sized enterprises provided in the embodiment of this application;

[0054] Figure 4 The credit risk association graph provided in the embodiment of the present application is embedded in the indicator graph;

[0055] Figure 5 The enterprise business information risk indicator diagram provided for the embodiment of this application;

[0056] Figure 6 A risk indicator diagram of the enterprise's historical loan information provided in the embodiment of the present application;

[0057] Figure 7 Another flowchart of a method for assessing credit risk of small and medium-sized enterprises provided in an embodiment of the present application;

[0058] Figure 8 An enterprise judicial risk indicator diagram provided for an embodiment of the present application;

[0059] Fig. 9 Feature subsets and model feature dimensions provided for embodiments of the present application;

[0060] Fig.10 The feature subset experimental results provided for the embodiments of the present application;

[0061] Fig.11 The model feature importance ranking result obtained based on the SHAP method provided in the embodiment of the present application;

[0062] Fig.12 A schematic diagram of the structure of a credit risk assessment device for small and medium-sized enterprises provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] For easier understanding, please refer to Figure 1 and Figure 2 The present application embodiment provides a method for assessing the credit risk of small and medium-sized enterprises, including:

[0065] Step 110: Obtain multi-source related data of small and medium-sized enterprises.

[0066] The multi-source related data of small and medium-sized enterprises include the industrial and commercial registration data, shareholder and senior executive data, judicial litigation data and loan business contract information data of small and medium-sized enterprises; among them, the industrial and commercial registration data of small and medium-sized enterprises include the basic industrial and commercial information of the enterprises, such as enterprise name, credit code, scale, industry, etc.; the shareholder and senior executive data record the information of the shareholders, legal persons, senior executives, etc. of the enterprises; the judicial litigation data record the event information such as judicial decisions and penalties for violations involving the enterprises; the loan business contract information data includes the business data of several Internet micro-loan platforms.

[0067] Step 120: Construct a financial credit risk correlation map for SMEs through multi-source correlation data of SMEs.

[0068] After preprocessing the acquired multi-source association data of SMEs, a financial credit risk association graph of SMEs is constructed. The financial credit risk association graph of SMEs consists of a subgraph of the SME industrial and commercial social network and a subgraph of the SME lending business network. Please refer to Figure 3 The SME financial credit risk association graph describes the association between entities. Due to different data sources, the entities are also heterogeneous.

[0069] In the SME business social network subgraph, there are two main entities: enterprises and natural persons, and the relationships include directors, supervisors, supervisors, investors, and legal persons. The SME lending business network subgraph describes the lending business association relationship between entities. The entities include enterprises and lending contracts, and the relationship types are mainly borrowing and lending. By mining the node relationships in the risk association graph, the financial risk transmission relationship between different enterprises can be established. For example, if a company has extremely poor credit risk and the company's legal person has multiple financial loan defaults and high consumption restrictions, then the N-degree associated entities of the corresponding enterprise entity will be affected.

[0070] Step 130: extract graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators.

[0071] The SME financial credit risk association graph represents the relationship between entities in the form of a graph, but as a non-Euclidean data, it cannot be directly used as the input of the machine learning model. Graph embedding encodes the graph structure and can generate an embedding vector as the input of the model. This application uses three methods to extract graph embedding features from the SME financial credit risk association graph based on the risk contagion theory:

[0072] The first is an indicator construction method based on complex network theory, which mainly selects four centrality measures, namely degree centrality, betweenness centrality, closeness centrality and PageRank centrality. Degree centrality is the most direct measure of node importance. The higher the value, the more important the node in the graph; betweenness centrality measures the degree of control of a node on important paths. Closeness centrality measures the central position of a given node in the graph; PageRank centrality measures the influence of a node in the graph.

[0073] The second method is the feature extraction method of graph computing. Using the graph computing method, the topological structure features of the evaluated node (i.e. the evaluated enterprise) in the credit risk association graph are extracted. Here, the positional relationship between the evaluated node and the risk node, as well as the positional relationship between the borrowing enterprise and the loan platform are mainly considered. Specifically, the embodiment of the present application extracts the number of micro-loan platforms with industrial and commercial association paths within 6 degrees of the borrowing enterprise, and the length of the industrial and commercial association path of the micro-loan platform closest to the borrowing enterprise.

[0074] The third method is a vector representation method based on random walk sampling. The node2vec embedding algorithm can be used to extract node representation vectors from the SME financial credit risk association graph. The algorithm trains a low-dimensional vector representation of each graph node through random walks. The vector of this node can well reflect the local and global structure and semantic similarity information of the graph. Node embedding learning methods such as graph convolutional neural networks can also be used to extract node representation vectors from the SME financial credit risk association graph.

[0075] The credit risk association graph extracted by the above three methods is embedded in the indicators such as Figure 4 As shown in the figure, the first part is the node importance index constructed based on complex network theory, the second part is the risk feature index extracted based on graph calculation, and the last part is the node representation vector obtained based on the Node2vec algorithm or graph convolutional neural network.

[0076] Step 140: extracting business information risk indicators from business registration data.

[0077] After cleaning and sorting the industrial and commercial registration data, the industrial and commercial information risk indicators were extracted. The initial data included the company name, company type, industry, industry, establishment time, paid-in capital, etc. According to the data preprocessing results, the indicators with too many missing values ​​and obviously irrelevant were eliminated, and finally the borrowing company type, borrowing company industry type, and borrowing company industry were extracted as industrial and commercial information risk indicators, such as Figure 5 shown.

[0078] Step 150: extract the enterprise's historical loan information risk index from the loan business contract information data.

[0079] The historical loan information of enterprises can be obtained through the off-site supervision system deployed by the financial regulatory authorities in financial institutions. After data cleaning, the information risk indicators of historical loans of enterprises are extracted, such as Figure 6 As shown in the figure, for the time from the first loan to the present, the number of days from the first loan date of the borrowing enterprise in the database to the current loan date is used. For the first-time borrowing enterprise, the indicator value is 0.

[0080] Step 160: Input the credit risk association graph embedding index, the industrial and commercial information risk index and the enterprise historical loan information risk index into the credit risk assessment model to perform credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

[0081] In the embodiment of the present application, the training process of the credit risk assessment model includes:

[0082] Extract risk indicators of small and medium-sized enterprises with known credit risk assessment results to obtain a data set. The risk indicators include credit risk association graph embedding indicators, industrial and commercial information risk indicators, and enterprise historical loan information risk indicators.

[0083] Divide the data set into a training set and a test set, train the machine learning model with the training set, and test the trained machine learning model with the test set until the machine learning model converges to obtain a trained machine learning model;

[0084] Obtain a credit risk assessment model based on the trained machine learning model.

[0085] This application builds a credit risk assessment model for SMEs based on machine learning methods. The credit risk assessment model for SMEs divides credit entities into two categories: defaulting enterprises and non-defaulting enterprises, and estimates the future default probability of the assessed entity. is an m-dimensional feature vector representing the enterprise being evaluated, is a binary variable that distinguishes good loans (y = 0) from bad loans (y = 1). The credit risk assessment model estimates the (posterior) probability p of default event i occurring. i :

[0086]

[0087] Wherein, f(·) is a credit risk assessment model. One or more machine learning models among logistic regression model (LR), decision tree model (DT), K nearest neighbor model (KNN), random forest model (RF), XGBoost model (XGB), support vector machine model (SVM) can be used as the credit risk assessment model.

[0088] The credit risk association graph embedding index, industrial and commercial information risk index and enterprise historical loan information risk index are obtained through the loan enterprise data with known credit risk assessment results to obtain a data set, and the data set is used as input data, and the credit risk assessment results corresponding to the data set are used as training targets to train the above one or more machine learning models, and the trained machine learning models are used as credit risk assessment models. When training one machine learning model, the trained machine learning model is the final credit risk assessment model; when training multiple machine learning models, the machine learning model with the highest assessment accuracy can be selected from multiple trained machine learning models as the final credit risk assessment model, or multiple trained machine learning models can be used as credit risk assessment models. The credit risk association graph embedding index, industrial and commercial information risk index and enterprise historical loan information risk index of the enterprise to be assessed extracted in the above process are input into the credit risk assessment model for credit risk assessment, and the credit risk assessment results of small and medium-sized enterprises are obtained. When the credit risk assessment model is a plurality of trained machine learning models, the credit risk assessment results of the multiple machine learning models are averaged or voted to obtain the final credit risk assessment results.

[0089] Due to information opacity and lack of credit data, the credit risk assessment of small and medium-sized enterprises faces modeling difficulties caused by data limitations. Secondly, due to the weak risk resistance of small and medium-sized enterprises, they are easily affected by their own behavior and risk association, leading to credit risk contagion. This application innovatively proposes a financial risk modeling method for small and medium-sized enterprises based on complex networks. By utilizing the topological structure information of nodes in complex networks, the impact path of risk transmission between enterprises is constructed, thereby constructing a reliable risk assessment model to improve the accuracy of credit risk assessment for small and medium-sized enterprises.

[0090] The above is an embodiment of a method for assessing credit risk of small and medium-sized enterprises provided by the present application. The following is another embodiment of a method for assessing credit risk of small and medium-sized enterprises provided by the present application.

[0091] Please refer to Figure 7 The present application provides a method for assessing the credit risk of small and medium-sized enterprises, including:

[0092] Step 210: Obtain multi-source related data of small and medium-sized enterprises.

[0093] The multi-source related data of small and medium-sized enterprises include the industrial and commercial registration data, shareholder and senior executive data, judicial litigation data and loan business contract information data of small and medium-sized enterprises; among them, the industrial and commercial registration data of small and medium-sized enterprises include the basic industrial and commercial information of the enterprises, such as enterprise name, credit code, scale, industry, etc.; the shareholder and senior executive data record the information of the shareholders, legal persons, senior executives, etc. of the enterprises; the judicial litigation data record the event information such as judicial decisions and penalties for violations involving the enterprises; the loan business contract information data includes the business data of several Internet micro-loan platforms.

[0094] Step 220: Construct a financial credit risk correlation map for SMEs through multi-source correlation data of SMEs.

[0095] After preprocessing the acquired multi-source association data of SMEs, a financial credit risk association graph of SMEs is constructed. The financial credit risk association graph of SMEs consists of the SME industrial and commercial social network subgraph and the SME lending business network subgraph. The financial credit risk association graph of SMEs describes the association relationship between entities. Due to different data sources, the entities are also heterogeneous.

[0096] In the SME business social network subgraph, there are two main entities: enterprises and natural persons, and the relationships include directors, supervisors, supervisors, investors, and legal persons. The SME lending business network subgraph describes the lending business association relationship between entities. The entities include enterprises and lending contracts, and the relationship types are mainly borrowing and lending. By mining the node relationships in the risk association graph, the financial risk transmission relationship between different enterprises can be established. For example, if a company has extremely poor credit risk and the company's legal person has multiple financial loan defaults and high consumption restrictions, then the N-degree associated entities of the corresponding enterprise entity will be affected.

[0097] Step 230: extract graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators.

[0098] The SME financial credit risk association graph represents the relationship between entities in the form of a graph, but as a non-Euclidean data, it cannot be directly used as the input of the machine learning model. Graph embedding encodes the graph structure and can generate an embedding vector as the input of the model. This application uses three methods to extract graph embedding features from the SME financial credit risk association graph based on the risk contagion theory:

[0099] The first is an indicator construction method based on complex network theory, which mainly selects four centrality measures, namely degree centrality, betweenness centrality, closeness centrality and PageRank centrality. Degree centrality is the most direct measure of node importance. The higher the value, the more important the node in the graph; betweenness centrality measures the degree of control of a node on important paths. Closeness centrality measures the central position of a given node in the graph; PageRank centrality measures the influence of a node in the graph.

[0100] The second method is the feature extraction method of graph computing. Using the graph computing method, the topological structure features of the evaluated node (i.e. the evaluated enterprise) in the credit risk association graph are extracted. Here, the positional relationship between the evaluated node and the risk node, as well as the positional relationship between the borrowing enterprise and the loan platform are mainly considered. Specifically, the embodiment of the present application extracts the number of micro-loan platforms with industrial and commercial association paths within 6 degrees of the borrowing enterprise, and the length of the industrial and commercial association path of the micro-loan platform closest to the borrowing enterprise.

[0101] The third method is a vector representation method based on random walk sampling. The node2vec embedding algorithm can be used to extract node representation vectors from the SME financial credit risk association graph. The algorithm trains a low-dimensional vector representation of each graph node through random walks. The vector of this node can well reflect the local and global structure and semantic similarity information of the graph. Node embedding learning methods such as graph convolutional neural networks can also be used to extract node representation vectors from the SME financial credit risk association graph.

[0102] The credit risk association graph embedding indicators extracted by the above three methods include three parts. The first part is the node importance indicator constructed based on complex network theory, the second part is the risk feature indicator extracted based on graph calculation, and the last part is the node representation vector obtained based on the Node2vec algorithm or graph convolutional neural network.

[0103] Step 240: extracting business information risk indicators from business registration data.

[0104] After cleaning and sorting the industrial and commercial registration data, the industrial and commercial information risk indicators were extracted. The initial data included the company name, company type, industry, industry, establishment time, paid-in capital, etc. According to the data preprocessing results, the indicators with too many missing values ​​and obviously irrelevant were eliminated, and finally the borrowing company type, borrowing company industry type, and borrowing company industry were extracted as industrial and commercial information risk indicators.

[0105] Step 250: extract the enterprise's historical loan information risk index from the loan business contract information data.

[0106] The historical loan information of enterprises can be obtained through the off-site supervision system deployed by the financial regulatory authorities in financial institutions. After data cleaning, the information risk indicators of historical loans of enterprises are extracted. For the time indicator from the first loan to the present, the number of days from the first loan date of the borrowing enterprise in the database to the current loan date is used. For the first-time loan enterprise, the indicator value is 0.

[0107] Step 260: extracting corporate judicial risk indicators from judicial litigation data.

[0108] Based on the relevant judicial litigation data of the assessed enterprise, the embodiment of the present application extracts the enterprise judicial risk indicators, such as Figure 8 As shown. The selection window for the corporate judicial risk index is the judicial risk performance within one year before the date of this loan, and it is evaluated according to the type of subject involved, the type of risk involved, and the amount involved. The types of subjects involved include the borrowing enterprise itself, the legal person, senior executives, and shareholders of the enterprise; the types of risks involved include whether it is identified as an abnormal taxpayer, whether there is a record of tax arrears, whether it is restricted from high consumption, and whether it is publicly executed. In the risk disclosure for public execution, the amount of execution is also recorded, so here is constructed a field for the ratio of the amount of public execution to the amount of this loan to evaluate the transmission of liquidity risk caused by the execution penalty to the borrowing enterprise.

[0109] Step 270: Input the credit risk association graph embedding index, the industrial and commercial information risk index, the enterprise historical loan information risk index and the enterprise judicial risk index into the credit risk assessment model to conduct credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

[0110] In the embodiment of the present application, the training process of the credit risk assessment model includes:

[0111] Extract risk indicators of small and medium-sized enterprises with known credit risk assessment results to obtain a data set. The risk indicators include credit risk association graph embedding indicators, industrial and commercial information risk indicators, enterprise historical loan information risk indicators, and enterprise judicial risk indicators.

[0112] Divide the data set into a training set and a test set, train the machine learning model with the training set, and test the trained machine learning model with the test set until the machine learning model converges to obtain a trained machine learning model;

[0113] Obtain a credit risk assessment model based on the trained machine learning model.

[0114] This application builds a credit risk assessment model for SMEs based on machine learning methods. The credit risk assessment model for SMEs divides credit entities into two categories: defaulting enterprises and non-defaulting enterprises, and estimates the future default probability of the assessed entity. is an m-dimensional feature vector representing the enterprise being evaluated, is a binary variable that distinguishes good loans (y = 0) from bad loans (y = 1). The credit risk assessment model estimates the (posterior) probability p of default event i occurring. i :

[0115]

[0116] Wherein, f(·) is a credit risk assessment model. One or more machine learning models among a logistic regression model, a decision tree model, a K nearest neighbor model, a random forest model, an XGBoost model, and a support vector machine model can be used as the credit risk assessment model.

[0117] The credit risk association graph embedding index, industrial and commercial information risk index, enterprise historical loan information risk index and enterprise judicial risk index are obtained through the loan enterprise data with known credit risk assessment results to obtain a data set, and the data set is used as input data, and the credit risk assessment results corresponding to the data set are used as training targets to train the above one or more machine learning models, and the trained machine learning models are used as credit risk assessment models. When training a machine learning model, the trained machine learning model is the final credit risk assessment model; when training multiple machine learning models, the machine learning model with the highest assessment accuracy can be selected from multiple trained machine learning models as the final credit risk assessment model, or multiple trained machine learning models can be used as credit risk assessment models. The credit risk association graph embedding index, industrial and commercial information risk index, enterprise historical loan information risk index and enterprise judicial risk index of the enterprise to be assessed extracted in the above process are input into the credit risk assessment model for credit risk assessment, and the credit risk assessment results of small and medium-sized enterprises are obtained. When the credit risk assessment model is a plurality of trained machine learning models, the credit risk assessment results of the multiple machine learning models are averaged or voted to obtain the final credit risk assessment results.

[0118] In order to evaluate the predictive ability of risk association graph information for the financial credit risk of small and medium-sized enterprises, the embodiment of the present application divides all the features into 4 feature subsets and forms 5 models. The AUC (Area Under Curve) indicator is used to evaluate each model, which mainly examines the model's ability to distinguish between positive and negative samples and is independent of the threshold. The normal value range of the AUC indicator is between 0.5 and 1. The closer the AUC value is to 1, the better the classifier performance is. The closer the AUC value is to 0.5, the closer the classifier is to a random classifier and has no application value. Among them, Model 1 is the most basic model, which only uses the company's own ICBC information risk indicators; Model 2 adds historical loan risk information; Model 3 additionally adds judicial risk information. Model 4 adds credit risk association graph information on the basis of Model 2, and is the model that this application focuses on. Model 5 integrates all risk feature subsets, such as Fig. 9 shown.

[0119] Through the loan enterprise data with known credit risk assessment results, we obtain the credit risk association graph embedding index, industrial and commercial information risk index, enterprise historical loan information risk index and enterprise judicial risk index as the data set. In the experiment, we randomly sampled 75% of the data from the data set as the training set, and the remaining 25% as the test set. After 5 random sampling divisions with different random seeds and 10 repeated experiments, Fig.10 The average performance metrics of each model are shown.

[0120] Judging from the experimental results, Model 4 and Model 5 have the best experimental results, and are significantly stronger than the first three models in terms of the ability to distinguish positive samples. It can be concluded that the credit risk association graph embedding index plays a significant role in improving the model's ability to distinguish. Comparing Model 4 with Model 2 and Model 5 with Model 3, the former adds a credit risk association graph indicator more than the latter, and the AUC value of the model is improved on all classifiers, which proves that the graph indicator is robust to the improvement of the evaluation model. This application aims to solve the problems of information asymmetry and complex risk associations in the credit risk management of small and medium-sized enterprises. Based on the graph data paradigm, this application integrates and manages multi-source heterogeneous data related to the credit risk of small and medium-sized enterprises, and constructs a credit risk association graph for small and medium-sized enterprises to realize the mining and modeling of complex risk associations between small and medium-sized enterprises; the network structure characteristics of the credit risk of small and medium-sized enterprises are characterized by a graph embedding method to provide new features for the evaluation model, which helps to improve the accuracy of credit evaluation results.

[0121] Based on the goal of supervision and penetrating supervision, financial regulatory authorities usually have certain requirements for the interpretability of financial institutions' credit models. Model-agnostic interpretability can analyze the relative importance ranking of all features. Here, the SHAP method, which is more commonly used in the industry, is used for analysis. The results are as follows: Fig.11 As shown. From the results, among the top 10 important features, graph embedding features account for 9. Among them, 4 are from the 5-dimensional features of node2vec. The 4 node centrality embedding indicators proposed by complex network theory, among which closeness, betweenness, and PageRank centrality rank 2nd, 3rd, and 4th respectively. These three are all measures of the importance of nodes in the global network. The degree centrality, which represents the local structure, ranks 9th, which is lower than the previous features, indicating that it is not enough to consider only the local topological structure. At the same time, the graph computing embedding feature LCOMPANY_COUNT (the number of micro-loan platforms with industrial and commercial association paths within 6 degrees of the borrowing enterprise) is also among the important features, indicating that the relationship between the borrowing enterprise and the financial institution is related to the probability of loan default. This result provides inspiration for issues such as multiple borrowing and self-financing supervision.

[0122] In the absence of financial data and insufficient historical credit information of SMEs, this application studies how to identify and model credit risks from the perspective of complex risk correlation structure based on public, short-term credit information, and effectively assess the credit risks of SMEs. This application constructs a credit risk association map based on the risk contagion theory, supplements the risk information of small and medium-sized enterprises with cross-sectional risk association information, solves the information asymmetry problem from another perspective, and improves the technical problem of low accuracy of credit risk assessment of small and medium-sized enterprises due to insufficient data; by introducing risk association maps and complex network theory, using the highly accessible public information and short-term information of small and medium-sized enterprises, credit risks are identified and modeled from the level of complex association structures, thereby constructing a reliable credit risk assessment model for small and medium-sized enterprises; credit risk assessment of small and medium-sized enterprises is performed using multiple indicators such as credit risk association map embedding indicators, industrial and commercial information risk indicators, enterprise historical loan information risk indicators and enterprise judicial risk indicators, considering the impact of multiple factors on the credit risk of small and medium-sized enterprises from different perspectives, which is helpful to improve the accuracy of credit risk assessment of small and medium-sized enterprises; based on the risk modeling idea of ​​complex network structure, the commonality of the local neighborhood topological structure of the node is used to realize the identification and assessment of risk nodes, and further explore the complex dependencies between enterprises and potential risk transmission paths, thereby well realizing the modeling of the global financial risk contagion phenomenon, which is helpful to improve the accuracy of credit risk assessment of small and medium-sized enterprises.

[0123] Please refer to Fig.12 The present application also provides a device for assessing the credit risk of small and medium-sized enterprises, including:

[0124] The data acquisition unit 310 is used to acquire multi-source related data of SMEs, which includes the business registration data, shareholder and executive data, judicial litigation data, and loan business contract information data of SMEs;

[0125] A graph construction unit 320, for constructing a SME financial credit risk association graph through multi-source association data of SMEs;

[0126] The risk index extraction unit 330 is used to extract the graph embedding features from the SME financial credit risk association graph to obtain the credit risk association graph embedding index; extract the business information risk index from the business registration data; extract the enterprise historical loan information risk index from the loan business contract information data;

[0127] The risk assessment unit 340 is used to input the graph embedding features, the business information risk indicators and the enterprise historical loan information risk indicators into the credit risk assessment model to perform credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

[0128] As a further improvement, the risk index extraction unit 330 is also used to extract the enterprise judicial risk index from the judicial litigation data;

[0129] The risk assessment form 340 is specifically used to input the credit risk association map embedding indicators, industrial and commercial information risk indicators, enterprise historical loan information risk indicators and enterprise judicial risk indicators into the credit risk assessment model to conduct credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

[0130] The embodiment of the present application also provides an electronic device, the device comprising a processor and a memory;

[0131] The memory is used to store the program code and transmit the program code to the processor;

[0132] The processor is used to execute the credit risk assessment method for small and medium-sized enterprises in the aforementioned method embodiment according to the instructions in the program code.

[0133] The embodiment of the present application also provides a computer-readable storage medium, which is used to store program codes. When the program codes are executed by a processor, the credit risk assessment method for small and medium-sized enterprises in the aforementioned method embodiment is implemented.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0135] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0136] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0137] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0138] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0140] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0141] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A credit risk assessment method for small and medium-sized enterprises, characterized in that: include: Acquire multi-source related data of small and medium-sized enterprises, wherein the multi-source related data of small and medium-sized enterprises includes business registration data, shareholder and executive data, judicial litigation data, and loan business contract information data of small and medium-sized enterprises; Constructing a financial credit risk association map for SMEs through the multi-source association data of SMEs; Extracting graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators; Extracting business information risk indicators from the business registration data; Extracting the enterprise's historical loan information risk index from the loan business contract information data; The credit risk association graph embedding index, the industrial and commercial information risk index and the enterprise historical loan information risk index are input into a credit risk assessment model to perform credit risk assessment and obtain credit risk assessment results for small and medium-sized enterprises.

2. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The SME financial credit risk association graph includes a SME business social network subgraph and a SME lending business network subgraph; The entities in the SME business social network subgraph include enterprises and natural persons, and the relationships between entities include directors, supervisors, managers, investors and legal persons; The entities in the SME lending business network subgraph include enterprises and lending contracts, and the relationships between entities include borrowing and lending.

3. The credit risk assessment method for small and medium-sized enterprises according to claim 1 or 2, characterized in that: The step of extracting graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators includes: Extracting centrality measures of borrowing enterprises from the SME financial credit risk association graph, wherein the centrality measures include degree centrality, betweenness centrality, closeness centrality and PageRank centrality; Based on the graph calculation method, the number of small loan platforms with industrial and commercial association paths within 6 degrees of the borrowing enterprise and the length of the industrial and commercial association path of the small loan platform closest to the borrowing enterprise are extracted from the SME financial credit risk association graph; Node representation vectors are extracted from the SME financial credit risk association graph based on the Node2vec algorithm or graph convolutional neural network.

4. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The industrial and commercial information risk indicators include: borrowing enterprise type, borrowing enterprise industry type, and the industry to which the borrowing enterprise belongs.

5. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The risk indicators of the enterprise's historical loan information include: the natural logarithm of the last loan amount, the annualized interest rate of the last loan, whether the last loan was in default, whether the last loan was overdue, the number of days between the last loan and the current loan, the ratio of the current loan amount to the historical average amount, the number of historical borrowings, and the number of days between the first loan and the current loan.

6. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The method further comprises: extracting corporate judicial risk indicators from the judicial litigation data; The credit risk association graph embedding indicator, the industrial and commercial information risk indicator and the enterprise historical loan information risk indicator are input into the credit risk assessment model to perform credit risk assessment, and obtain the credit risk assessment results of small and medium-sized enterprises, including: The credit risk association map embedding index, the industrial and commercial information risk index, the enterprise historical loan information risk index and the enterprise judicial risk index are input into the credit risk assessment model to conduct credit risk assessment and obtain the credit risk assessment results of small and medium-sized enterprises.

7. The credit risk assessment method for small and medium-sized enterprises according to claim 6 is characterized in that: The corporate judicial risk indicators include: whether the corporate entity was an abnormal taxpayer within one year before the loan date, whether it owed taxes, whether it was restricted from high consumption, whether it was publicly executed, and the ratio of the current loan amount to the total amount of execution against the corporate entity within one year.

8. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The training process of the credit risk assessment model includes: Extracting risk indicators of small and medium-sized enterprises with known credit risk assessment results to obtain a data set, wherein the risk indicators include credit risk association graph embedding indicators, industrial and commercial information risk indicators, and enterprise historical loan information risk indicators; Dividing the data set into a training set and a test set, training the machine learning model with the training set, and testing the trained machine learning model with the test set until the machine learning model converges to obtain a trained machine learning model; A credit risk assessment model is obtained based on the trained machine learning model.

9. The credit risk assessment method for small and medium-sized enterprises according to claim 1 is characterized in that: The number of the machine learning models is 1 or more; The step of obtaining a credit risk assessment model based on the trained machine learning model includes: When the number of the machine learning models is 1, using the trained machine learning model as a credit risk assessment model; When there are multiple machine learning models, the machine learning model with the highest evaluation accuracy is selected from the multiple trained machine learning models as the credit risk assessment model, or multiple trained machine learning models are used as credit risk assessment models, and the average value or voting result of the credit risk assessment results output by the multiple trained machine learning models is used as the final credit risk assessment result.

10. A credit risk assessment device for small and medium-sized enterprises, characterized in that: include: A data acquisition unit, used to acquire multi-source related data of small and medium-sized enterprises, wherein the multi-source related data of small and medium-sized enterprises includes business registration data, shareholder and executive data, judicial litigation data, and loan business contract information data of small and medium-sized enterprises; A graph construction unit, used to construct a SME financial credit risk association graph through the SME multi-source association data; A risk indicator extraction unit, used for extracting graph embedding features from the SME financial credit risk association graph to obtain credit risk association graph embedding indicators; Extracting business information risk indicators from the business registration data; extracting enterprise historical loan information risk indicators from the loan business contract information data; The risk assessment unit is used to input the credit risk association map embedding index, the industrial and commercial information risk index and the enterprise historical loan information risk index into the credit risk assessment model to perform credit risk assessment and obtain the credit risk assessment result of small and medium-sized enterprises.

11. An electronic device, characterized in that: The device comprises a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the credit risk assessment method for small and medium-sized enterprises according to any one of claims 1 to 9 according to the instructions in the program code.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and when the program codes are executed by a processor, the method for assessing the credit risk of small and medium-sized enterprises according to any one of claims 1 to 9 is implemented.

Images