Small and micro enterprise credit risk prediction method and device, storage medium and processor

Abstract

The invention discloses a small and micro enterprise credit risk prediction method and device, a storage medium and a processor. In the scheme, based on acquired enterprise information of a target enterprise, a plurality of credit risk indexes of the target enterprise are converted into a plurality of index vectors; based on the plurality of converted index vectors, constructing a graph structure reflecting a dependency relationship between the indexes; in the graph structure, nodes represent credit risk indexes, and edges represent dependency relationships among the indexes; and then inputting the graph structure into a credit assessment model, performing prediction by the credit assessment model according to the learned graph structure information, and outputting a credit risk assessment level of the target enterprise. The method is different from the prior art which mainly depends on financial index analysis, historical repayment records and qualitative judgment and has limitation in the aspects of processing complex nonlinear relations and deep time sequence information. Compared with the prior art in which the credit risk of the enterprise is difficult to accurately capture, the method has obvious advantages.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, storage medium and processor for predicting credit risk of small and micro enterprises. Background Art

[0002] The credit risk of small and micro enterprises refers to the risk that small and micro enterprises may be unable to repay their debts on time due to changes in their own financial conditions, market environment, management capabilities and other factors during the course of their business operations.

[0003] Traditional credit risk assessment methods mainly rely on financial indicator analysis, historical repayment records and qualitative judgments, etc. These methods have limitations in dealing with complex nonlinear relationships and deep time series information. Especially for small and micro enterprises, due to their unstable operations and opaque information, traditional methods are difficult to accurately capture their credit risks.

[0004] Faced with the complexity of credit risk assessment for small and micro enterprises, how to extract in-depth information from a large amount of information to accurately predict credit risk is a technical problem that needs to be solved urgently. Summary of the invention

[0005] Based on the above problems, the present application provides a method, device, storage medium and processor for predicting credit risk of small and micro enterprises, aiming to improve the accuracy of credit risk prediction of small and micro enterprises.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] The first aspect of the present application provides a method for predicting credit risk of small and micro enterprises, the method comprising:

[0008] Obtain corporate information of target companies;

[0009] Based on the enterprise information of the target enterprise, converting multiple credit risk indicators of the target enterprise into multiple indicator vectors; the credit risk indicators are constructed by performing credit risk assessment based on the enterprise information of the enterprise;

[0010] Based on the converted multiple indicator vectors, a graph structure is constructed; the graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators;

[0011] The graph structure is input into a credit assessment model, and the credit assessment model makes predictions based on the learned graph structure information, and outputs a credit risk assessment grade of a target enterprise; the credit assessment model is obtained after training; the credit assessment model comprises a graph convolutional network and a bidirectional long short-term memory network connected in sequence; the graph convolutional network is used to learn information in the graph structure, and output a feature vector to the bidirectional long short-term memory network; the bidirectional long short-term memory network is used to extract time features and make predictions based on the feature vector, and output a credit risk assessment grade of a target enterprise.

[0012] In an optional implementation, the constructing of the graph structure includes:

[0013] An undirected weighted graph is used to construct a graph structure; the undirected weighted graph includes nodes, edges and edge weights.

[0014] In an optional implementation, the edge and the edge weight are determined by:

[0015] Respectively calculate the similarity between the indicator vector corresponding to the credit risk indicator i and the indicator vector corresponding to other credit risk indicators; i is a positive integer;

[0016] The calculated similarities are arranged from large to small, and the first K credit risk indicators are selected; K is a positive integer;

[0017] Build edges between credit risk indicator i and the first K credit risk indicators;

[0018] The similarity between credit risk indicator i and the first K credit risk indicators is used as the weight of the corresponding edge.

[0019] In an optional implementation, the graph convolutional network learns information in the graph structure and outputs a feature vector to the bidirectional long short-term memory network, including:

[0020] According to the feature sequence composed of multiple indicator vectors and the adjacency matrix of the undirected weighted graph, the graph convolution network updates the representation of the node by aggregating the information of the node and its neighboring nodes, and outputs a one-dimensional feature vector of a fixed length to the long short-term memory network; the adjacency matrix of the undirected weighted graph is obtained by converting the relationship between the edges in the undirected weighted graph.

[0021] In an optional implementation, the bidirectional long short-term memory network extracts the time feature and makes a prediction based on the feature vector, and outputs the credit risk assessment level of the target enterprise, including:

[0022] Based on the one-dimensional feature vector, the time-dependent features in the one-dimensional feature vector are extracted by a forward long short-term memory network and a reverse long short-term memory network respectively, and the extraction results of the forward long short-term memory network and the extraction results of the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

[0023] In an optional implementation, the training process of the credit assessment model includes:

[0024] Get sample data;

[0025] Preprocessing the sample data;

[0026] Dividing the sample data into a training set and a test set according to a preset ratio;

[0027] Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

[0028] The second aspect of the present application provides a small and micro enterprise credit risk prediction device, which includes:

[0029] Enterprise information acquisition module, used to acquire enterprise information of target enterprises;

[0030] A credit risk indicator conversion module, used to convert multiple credit risk indicators of the target enterprise into multiple indicator vectors based on the enterprise information of the target enterprise; the credit risk indicators are constructed by the bank through credit risk assessment based on the enterprise information of the enterprise;

[0031] A graph construction module, for constructing a graph structure based on the converted multiple indicator vectors; the graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators;

[0032] A credit risk prediction module is used to input the graph structure into a credit assessment model, and the credit assessment model makes predictions based on the learned graph structure information and outputs the credit risk assessment level of the target enterprise; the credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network.

[0033] In an optional implementation, the device further includes: a training module; the training module is used to train the credit assessment model in the following manner:

[0034] Get sample data;

[0035] Preprocessing the sample data;

[0036] Dividing the sample data into a training set and a test set according to a preset ratio;

[0037] Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

[0038] A third aspect of the present application provides a computer-readable storage medium, in which a computer program is stored. When the program is executed by a processor, a method for predicting credit risk of small and micro enterprises as provided in any implementation of the first aspect is implemented.

[0039] A fourth aspect of the present application provides a processor for running a computer program, which, when running, executes a method for predicting credit risk of small and micro enterprises as provided in any implementation of the first aspect.

[0040] Compared with the prior art, this application has the following beneficial effects:

[0041] The credit risk prediction method for small and micro enterprises provided in this application converts the credit risk-related indicators of the enterprise into indicator vectors based on the relevant information of the enterprise obtained, so as to capture the characteristics reflected by each indicator in a digital form in the future. Based on the converted multiple indicator vectors, a graph structure reflecting the dependency relationship between indicators is constructed. In this graph structure, nodes represent credit risk indicators, and edges represent the dependency relationship between indicators. Then, the graph structure constructed is processed by the graph convolution network (GCN) in the credit assessment model to learn and extract the dimensional features and dependency patterns between indicators, and then the learned graph structure information is input into the bidirectional long short-term memory network (Bi-LSTM) for processing to extract the time characteristics and long-term dependencies of the data. This model can not only reflect the interaction between credit risk indicators, but also capture the long-term dependencies in time series data. Therefore, the credit risk prediction method for small and micro enterprises provided in this application realizes accurate prediction of the credit risk of small and micro enterprises by using the credit assessment model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] 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 labor.

[0043] Figure 1 A flowchart of a method for predicting credit risk of small and micro enterprises provided in an embodiment of the present application;

[0044] Figure 2 A flowchart of another small and micro enterprise credit risk prediction method provided in an embodiment of the present application;

[0045] Figure 3 A training flow chart of a credit assessment model (GC-BL model) provided in an embodiment of the present application;

[0046] Figure 4 A schematic diagram of the structure of a credit assessment model (GC-BL model) provided in an embodiment of the present application;

[0047] Figure 5 The overall framework diagram of the small and micro enterprise credit risk prediction method provided in the embodiment of the present application;

[0048] Figure 6 A schematic diagram of the structure of a small and micro enterprise credit risk prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] As described above, current credit risk assessment methods mainly rely on financial indicator analysis, historical repayment records and qualitative judgments, etc. These methods have limitations in dealing with complex nonlinear relationships and deep time series information. Especially for small and micro enterprises, due to their unstable operations and opaque information, it is difficult for current credit risk assessment methods to accurately capture their credit risks.

[0050] In view of the above problems, the inventor has proposed a method, device, storage medium and processor for predicting the credit risk of small and micro enterprises after research. Based on the enterprise information of the target enterprise obtained, multiple credit risk indicators of the target enterprise are converted into multiple indicator vectors, so as to capture the characteristics reflected by each indicator in a digital form in the future. Based on the converted multiple indicator vectors, a graph structure reflecting the dependency relationship between indicators is constructed. In this graph structure, the nodes represent credit risk indicators, and the edges represent the dependency relationship between indicators. Then, the constructed graph structure is processed by the graph convolution network (GCN) in the credit assessment model to learn and extract the dimensional features and dependency patterns between indicators, and then the learned graph structure information is input into the bidirectional long short-term memory network (Bi-LSTM) for processing to extract the time characteristics and long-term dependencies of the data. The model can not only reflect the interaction between credit risk indicators, but also capture the long-term dependencies in time series data. Therefore, the credit risk prediction method for small and micro enterprises provided by this application realizes accurate prediction of the credit risk of small and micro enterprises by using the credit assessment model.

[0051] 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.

[0052] See also Figure 1 , which is a flow chart of a method for predicting credit risk of small and micro enterprises provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0053] S101. Obtain enterprise information of a target enterprise.

[0054] The target company's corporate information can be obtained through a variety of channels and tools to obtain the company's operating conditions, tax information, loan information, equity information, etc. The company's corporate information is obtained in a legal and compliant manner.

[0055] S102: Based on the enterprise information of the target enterprise, convert multiple credit risk indicators of the target enterprise into multiple indicator vectors.

[0056] The credit risk index is constructed by conducting credit risk assessment based on the enterprise information of the enterprise.

[0057] This embodiment introduces an embedding vector representing the characteristics of each credit risk indicator. The embedding vector technology is a technology that converts categorical data (such as words, labels, IDs, etc.) into points in a continuous vector space. Through this conversion, the semantic information of the original data can be preserved in the vector space and can be used in various machine learning models to improve processing efficiency and model performance.

[0058] S103. Construct a graph structure based on the converted multiple indicator vectors.

[0059] The graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators.

[0060] The constructing of the graph structure includes:

[0061] An undirected weighted graph is used to construct a graph structure; the undirected weighted graph includes nodes, edges and edge weights.

[0062] Graph structure is a graph representation of data learned through algorithms, where data entities are nodes and relationships between entities are edges. This method can reveal hidden structures and dependencies in data, and is particularly suitable for data that naturally forms a network structure, such as social networks, knowledge graphs, protein interaction networks, etc.

[0063] The edge and the edge weight are determined by:

[0064] Respectively calculate the similarity between the indicator vector corresponding to the credit risk indicator i and the indicator vector corresponding to other credit risk indicators; i is a positive integer;

[0065] The calculated similarities are arranged from large to small, and the first K credit risk indicators are selected; K is a positive integer;

[0066] Build edges between credit risk indicator i and the first K credit risk indicators;

[0067] The similarity between credit risk indicator i and the first K credit risk indicators is used as the weight of the corresponding edge.

[0068] Credit risk indicators with similar functions may have similar characteristic change curves, that is, they may have similar change trends, and similar indicator data will show strong correlation.

[0069] S104: Input the graph structure into a credit assessment model, and the credit assessment model performs prediction based on the learned graph structure information and outputs a credit risk assessment grade of the target enterprise.

[0070] Among them, the credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network connected in sequence; the graph convolutional network is used to learn the information in the graph structure and output a feature vector to the bidirectional long short-term memory network; the bidirectional long short-term memory network is used to extract time features and make predictions based on the feature vector, and output the credit risk assessment level of the target enterprise.

[0071] The graph convolutional network learns information in the graph structure and outputs a feature vector to the bidirectional long short-term memory network, including:

[0072] According to the feature sequence composed of multiple indicator vectors and the adjacency matrix of the undirected weighted graph, the graph convolution network updates the representation of the node by aggregating the information of the node and its neighboring nodes, and outputs a one-dimensional feature vector of a fixed length to the long short-term memory network; the adjacency matrix of the undirected weighted graph is obtained by converting the relationship between the edges in the undirected weighted graph.

[0073] Graph convolutional network is a neural network architecture that operates on graph structured data and can learn node representations directly on the graph. Graph convolutional network can capture the dependencies between nodes by performing local neighbor aggregation and feature combination on the nodes of the graph, and is suitable for tasks such as node classification, graph classification, and link prediction.

[0074] The bidirectional long short-term memory network extracts the time feature based on the feature vector and makes predictions, and outputs the credit risk assessment level of the target enterprise, including:

[0075] Based on the one-dimensional feature vector, the time-dependent features in the one-dimensional feature vector are extracted by a forward long short-term memory network and a reverse long short-term memory network respectively, and the extraction results of the forward long short-term memory network and the extraction results of the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

[0076] Bi-LSTM is a special type of recurrent neural network (RNN) that can capture long-term dependencies in sequence data. Compared with traditional LSTM, Bi-LSTM considers both past and future information when processing sequence data, updating its memory from two directions, which makes Bi-LSTM particularly suitable for tasks that require understanding the entire sequence data, such as text processing and speech recognition.

[0077] The credit risk prediction method for small and micro enterprises provided in the embodiment of the present application, based on the relevant information of the target enterprise obtained, converts multiple credit risk indicators of the target enterprise into an indicator vector, so as to capture the characteristics reflected by each indicator in a digital form later. Based on the converted multiple indicator vectors, a graph structure reflecting the dependency relationship between indicators is constructed. In this graph structure, nodes represent credit risk indicators, and edges represent the dependency relationship between indicators. Then, the graph structure constructed is processed by the graph convolution network (GCN) in the credit assessment model to learn and extract the dimensional features and dependency patterns between indicators, and then the learned graph structure information is input into the bidirectional long short-term memory network (Bi-LSTM) for processing to extract the time characteristics and long-term dependencies of the data. This model can not only reflect the interaction between credit risk indicators, but also capture the long-term dependencies in time series data. Therefore, the credit risk prediction method for small and micro enterprises provided in the present application, by using the credit assessment model, realizes the accurate prediction of the credit risk of small and micro enterprises.

[0078] The credit assessment model in the small and micro enterprise credit risk prediction method described in the above embodiment is obtained after training. The following embodiment introduces the training process of the credit assessment model in detail.

[0079] See also Figure 2, which is a flowchart of another small and micro enterprise credit risk prediction method provided by the embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0080] S201. Train a credit assessment model.

[0081] like Figure 3 As shown, Figure 3 The training process of the credit assessment model (GC-BL model) is shown, including:

[0082] Get sample data;

[0083] Preprocessing the sample data;

[0084] Preprocessing the sample data includes data labeling and data cleaning. Data labeling refers to adding labels or annotations to the original data to make it the training data of the supervised learning algorithm. Common data labeling types include classification labels, bounding boxes, key points, etc. Data cleaning refers to discovering and correcting errors, incompleteness or inconsistencies in the data to improve the quality of the data. Common data cleaning tasks include missing value processing, outlier processing, duplicate data processing, etc.

[0085] By labeling and cleaning sample data, the quality of data and the performance of the model can be significantly improved. During the labeling process, the consistency and accuracy of the labeling should be ensured. During the data cleaning process, attention should be paid to handling missing values, outliers and duplicate data, and data format standardization and type conversion should be performed at the same time.

[0086] Dividing the sample data into a training set and a test set according to a preset ratio;

[0087] Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

[0088] S202: Obtain the enterprise information of the target enterprise.

[0089] Obtain the target company's operating conditions, tax information, loan information, equity information, etc.

[0090] S203: Based on the enterprise information of the target enterprise, convert multiple credit risk indicators of the target enterprise into multiple indicator vectors.

[0091] The credit risk index is constructed by conducting credit risk assessment based on the enterprise information of the enterprise.

[0092] In this embodiment, a customer scoring system is constructed from both quantitative and qualitative analysis, and five primary indicators, 24 secondary indicators, and 89 tertiary indicators, including basic enterprise quality, enterprise business vitality, enterprise repayment ability, enterprise performance willingness, and enterprise growth ability, are established. Most of the enterprise's risk indicators, such as historical sales, tax information, loan information, and equity information, will continue to change with the enterprise's operating conditions, and the enterprise's credit risk level should also be adjusted accordingly.

[0093] This embodiment introduces an embedding vector for each indicator to represent its characteristics:

[0094]

[0095] Among them, v i represents the learned credit indicator embedding vector, and d represents the dimension of the embedding vector. These embedding vectors are randomly initialized at first and then determined through training.

[0096] S204: construct a graph structure based on the converted multiple indicator vectors.

[0097] The graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators.

[0098] This embodiment uses an undirected weighted graph to represent the dependency relationship between credit risk indicators, and the undirected weighted graph includes nodes, edges and edge weights. In the absence of prior information, the candidate relationships of indicator i are only all indicators, not itself.

[0099]

[0100] By calculating the embedding vector of node i and its candidate node j∈C i The similarity between the embedding vectors of is used to select the mutual spatial dependencies of indicator i in these candidate nodes:

[0101]

[0102] First calculate e ji , that is, the embedding vector of index i and candidate relation j∈C i Then, the first K such standardized dot products are selected, and edges are constructed between credit risk indicator i and the first K credit risk indicators respectively, and the similarity between credit risk indicator i and the first K credit risk indicators is used as the weight K of the corresponding edge. The value of K can be selected according to the desired sparsity level.

[0103] Use the adjacency matrix A to represent this undirected graph, where A ijIndicates the existence of an edge between node i and node j.

[0104] Embedded vector technology is used to convert various credit risk-related indicators into vectors in high-dimensional space. Through graph structure learning methods, the interdependence between indicators can be reflected.

[0105] S205: Input the graph structure into a credit assessment model, and the credit assessment model performs prediction based on the learned graph structure information and outputs a credit risk assessment grade of the target enterprise.

[0106] Among them, the credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network connected in sequence; the graph convolutional network is used to learn the information in the graph structure and output a feature vector to the bidirectional long short-term memory network; the bidirectional long short-term memory network is used to extract time features and make predictions based on the feature vector, and output the credit risk assessment level of the target enterprise.

[0107] The structure of the credit assessment model (GC-BL model) is as follows: Figure 4 As shown, it includes graph convolutional network (GCN) and bidirectional long short-term memory network (Bi-LSTM).

[0108] The graph convolutional network learns information in the graph structure and outputs a feature vector to the bidirectional long short-term memory network, including:

[0109] According to the feature sequence composed of multiple indicator vectors and the adjacency matrix of the undirected weighted graph, the graph convolution network updates the representation of the node by aggregating the information of the node and its neighboring nodes, and outputs a one-dimensional feature vector of a fixed length to the long short-term memory network; the adjacency matrix of the undirected weighted graph is obtained by converting the relationship between the edges in the undirected weighted graph.

[0110] In order to capture the relationship between risk indicators, the graph convolutional network is used to learn the characteristic function on the graph structure data G = (V, E), and the credit risk assessment and prediction problem of small and micro enterprises is expressed as the mapping relationship between the credit risk indicator network G and the characteristic sequence. train And the adjacency matrix A learned from the previous graph structure is used as input:

[0111] X=(s train ,A)

[0112]

[0113] where s trainIt is an N×T feature sequence composed of the feature description data of each node i, where N is the number of nodes (i.e. the number of samples of various indicators) and T is the length of all time series data. The adjacency matrix represents the spatial dependency relationship between nodes in matrix form. The output layer of the graph convolution is a compressed one-dimensional vector X ′ , each graph convolution layer can be written as a nonlinear function:

[0114] H (k+1) =f(H (k) ,A),k∈{0,1,…,L}

[0115] Where L is the number of graph convolutional layers. When k is equal to 0, H (0) Same as X; when k equals L, H (L) With X ′ The specific graph convolutional layers differ only in the choice and parameters of the activation function f(·).

[0116] f(H (k) ,A)=σ(AH (k) W (k) )

[0117] Where σ is a nonlinear activation function, such as the ReLU function, and W (k) is the weight matrix in the kth graph convolutional layer.

[0118] The graph convolutional network can retain rich spatial features in the data during the convolution process. The one-dimensional feature vector output by the network retains the most significant features of the original data for subsequent model training. The feature extraction capability of the graph convolutional network is used to reduce the collinearity problem of the input variables. It helps the model automatically learn the specific spatial relationship between various credit indicators.

[0119] The bidirectional long short-term memory network extracts the time feature and makes predictions based on the feature vector, and outputs the credit risk assessment grade of the target enterprise.

[0120] For each specific training, x (t) The sliding window size is w, x (t) =[s (t-w) ,s (t-w+1) ,…,s (t-1) ], and x (t) With the adjacency matrix A learned from the graph structure as input, the calculation process of each graph convolution can be expressed as:

[0121]

[0122] The calculation result of the graph convolutional network is compressed into a one-dimensional vector by the fully connected layer as the input of the next stage, and then connected to the Bi-LSTM network. Based on the one-dimensional feature vector, the forward long short-term memory network and the reverse long short-term memory network respectively extract the time-dependent features in the one-dimensional feature vector, and the extraction results of the forward long short-term memory network and the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

[0123] like Figure 5 As shown, Figure 5 The overall framework diagram of the credit risk prediction method for small and micro enterprises is shown. The GC-BL in the figure is the credit assessment model in this method. The constructed graph structure is input into the credit risk assessment model, and the hidden layer extracts useful features from the node and its adjacent nodes through convolution operation. The features of each node include not only its own features, but also the features of its neighboring nodes. This local aggregation operation enables the model to capture the local dependencies in the graph structure. Then, after nonlinear activation (such as ReLU), the model can learn more complex feature representations, and overfitting is prevented by discarding Dropout on the nodes. The softmax function converts the feature vector after Dropout into a probability distribution, that is, the sum of all elements is 1. The flatten operation is used to flatten the multi-dimensional feature vector after the softmax function into a one-dimensional feature vector, and the one-dimensional feature vector is used as the input of the Bi-LSTM network. Based on the one-dimensional feature vector, the time-dependent features in the one-dimensional feature vector are extracted by the forward long short-term memory network and the reverse long short-term memory network, respectively, and the extraction results of the forward long short-term memory network and the extraction results of the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

[0124] Another method for predicting the credit risk of small and micro enterprises provided in this embodiment takes into account the obvious lag effect and accumulation process between the changes in the credit risk of small and micro enterprises and their current and historical operating status. First, the unique characteristics between the credit risk indicators are flexibly captured using the embedded vector technology, and the abstract dependency relationship between these indicators is learned through the graph structure construction method, and then the dependency relationship is encoded as the edge in the graph. Then, a credit assessment model based on the time series correlation between credit risk indicators is established, aiming to learn the characteristics between credit risk indicators through the graph convolution network, and further extract the characteristics of time series data such as tax and transaction amount by using Bi-LSTM, so as to achieve accurate prediction of the credit risk of small and micro enterprises.

[0125] Based on the small and micro enterprise credit risk prediction method introduced in the previous embodiment, the present application also provides a small and micro enterprise credit risk prediction device. Figure 6 Figure 1 is a schematic diagram of the structure of the device. Figure 6As shown, the small and micro enterprise credit risk prediction device includes:

[0126] The enterprise information acquisition module 301 is used to acquire the enterprise information of the target enterprise;

[0127] A credit risk indicator conversion module 302 is used to convert multiple credit risk indicators of the target enterprise into multiple indicator vectors based on the enterprise information of the target enterprise; the credit risk indicators are constructed by the bank through credit risk assessment based on the enterprise information of the enterprise;

[0128] A graph construction module 303 is used to construct a graph structure based on the converted multiple indicator vectors; the graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators;

[0129] In an optional implementation, the constructing of the graph structure includes:

[0130] An undirected weighted graph is used to construct a graph structure; the undirected weighted graph includes nodes, edges and edge weights.

[0131] In an optional implementation, the edge and the edge weight are determined by:

[0132] Respectively calculate the similarity between the indicator vector corresponding to the credit risk indicator i and the indicator vector corresponding to other credit risk indicators; i is a positive integer;

[0133] The calculated similarities are arranged from large to small, and the first K credit risk indicators are selected; K is a positive integer;

[0134] Build edges between credit risk indicator i and the first K credit risk indicators;

[0135] The similarity between credit risk indicator i and the first K credit risk indicators is used as the weight of the corresponding edge.

[0136] The credit risk prediction module 304 is used to input the graph structure into the credit assessment model, and the credit assessment model makes predictions based on the learned graph structure information and outputs the credit risk assessment level of the target enterprise; the credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network.

[0137] In an optional implementation, the graph convolutional network learns information in the graph structure and outputs a feature vector to the bidirectional long short-term memory network, including:

[0138] According to the feature sequence composed of multiple indicator vectors and the adjacency matrix of the undirected weighted graph, the graph convolution network updates the representation of the node by aggregating the information of the node and its neighboring nodes, and outputs a one-dimensional feature vector of a fixed length to the long short-term memory network; the adjacency matrix of the undirected weighted graph is obtained by converting the relationship between the edges in the undirected weighted graph.

[0139] In an optional implementation, the bidirectional long short-term memory network extracts the time feature and makes a prediction based on the feature vector, and outputs the credit risk assessment level of the target enterprise, including:

[0140] Based on the one-dimensional feature vector, the time-dependent features in the one-dimensional feature vector are extracted by a forward long short-term memory network and a reverse long short-term memory network respectively, and the extraction results of the forward long short-term memory network and the extraction results of the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

[0141] Optionally, the small and micro enterprise credit risk prediction device further includes a training module, which is used to train the credit assessment model in the following manner:

[0142] Obtaining sample data; the sample data is divided into a training set and a test set according to a preset ratio;

[0143] Preprocessing the sample data;

[0144] Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

[0145] In addition, an embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. When the program is executed by a processor, a method for predicting credit risk of small and micro enterprises as described in any of the method embodiments is implemented.

[0146] In addition, an embodiment of the present application also provides a processor, which is used to run a computer program. When the program is running, it executes the small and micro enterprise credit risk prediction method introduced in any implementation method of the aforementioned method embodiment.

[0147] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0148] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting credit risk of small and micro enterprises, characterized in that: include: Obtain corporate information of target companies; Based on the enterprise information of the target enterprise, converting multiple credit risk indicators of the target enterprise into multiple indicator vectors; The credit risk indicators are constructed based on the credit risk assessment of the enterprise's corporate information; Based on the converted multiple indicator vectors, a graph structure is constructed; the graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators; The graph structure is input into a credit assessment model, and the credit assessment model makes predictions based on the learned graph structure information and outputs a credit risk assessment grade of the target enterprise; The credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network connected in sequence; the graph convolutional network is used to learn the information in the graph structure and output a feature vector to the bidirectional long short-term memory network; the bidirectional long short-term memory network is used to extract time features and make predictions based on the feature vector, and output the credit risk assessment level of the target enterprise.

2. The method according to claim 1, characterized in that: The constructing of the graph structure includes: An undirected weighted graph is used to construct a graph structure; the undirected weighted graph includes nodes, edges and edge weights.

3. The method according to claim 2, characterized in that The edge and the edge weight are determined by: Respectively calculate the similarity between the indicator vector corresponding to the credit risk indicator i and the indicator vector corresponding to other credit risk indicators; i is a positive integer; The calculated similarities are arranged from large to small, and the first K credit risk indicators are selected; K is a positive integer; Build edges between credit risk indicator i and the first K credit risk indicators; The similarity between credit risk indicator i and the first K credit risk indicators is used as the weight of the corresponding edge.

4. The method according to claim 2, characterized in that: The graph convolutional network learns information in the graph structure and outputs a feature vector to the bidirectional long short-term memory network, including: According to the feature sequence composed of multiple indicator vectors and the adjacency matrix of the undirected weighted graph, the graph convolution network updates the representation of the node by aggregating the information of the node and its neighboring nodes, and outputs a one-dimensional feature vector of a fixed length to the long short-term memory network; the adjacency matrix of the undirected weighted graph is obtained by converting the relationship between the edges in the undirected weighted graph.

5. The method according to claim 4, characterized in that Based on the feature vector, the bidirectional long short-term memory network extracts the time feature and makes predictions, and outputs the credit risk assessment level of the target enterprise, including: Based on the one-dimensional feature vector, the time-dependent features in the one-dimensional feature vector are extracted by a forward long short-term memory network and a reverse long short-term memory network respectively, and the extraction results of the forward long short-term memory network and the extraction results of the reverse long short-term memory network are combined to output the credit risk assessment level of the target enterprise.

6. The method according to claim 1, characterized in that The training process of the credit assessment model includes: Get sample data; Preprocessing the sample data; Dividing the sample data into a training set and a test set according to a preset ratio; Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

7. A credit risk prediction device for small and micro enterprises, characterized in that: include: Enterprise information acquisition module, used to acquire enterprise information of target enterprises; A credit risk indicator conversion module, used to convert multiple credit risk indicators of the target enterprise into multiple indicator vectors based on the enterprise information of the target enterprise; The credit risk index is constructed by the bank through credit risk assessment based on the enterprise’s corporate information; A graph construction module, for constructing a graph structure based on the converted multiple indicator vectors; the graph structure includes nodes and edges; the nodes represent the credit risk indicators of the target enterprise, and the edges represent the dependency relationship between the credit risk indicators; A credit risk prediction module, used to input the graph structure into a credit assessment model, wherein the credit assessment model makes predictions based on the learned graph structure information and outputs a credit risk assessment grade of the target enterprise; The credit assessment model is obtained after training; the credit assessment model includes a graph convolutional network and a bidirectional long short-term memory network.

8. A credit risk prediction device for small and micro enterprises, characterized in that: The device further comprises: a training module; the training module is used to train the credit assessment model in the following manner: Get sample data; Preprocessing the sample data; Dividing the sample data into a training set and a test set according to a preset ratio; Based on the preprocessed sample data, the credit assessment model is trained using the training set to determine the model parameters of the credit assessment model, and the trained credit assessment model is tested using the test set to verify the accuracy of the assessment result; the model parameters include the dimension of the indicator vector and the number of graph convolution layers.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the program is executed by the processor, the method for predicting credit risk of small and micro enterprises as described in any one of claims 1 to 6 is implemented.

10. A processor, characterized in that: Used to run a computer program, which, when running, executes the small and micro enterprise credit risk prediction method as described in any one of claims 1-6.

Images