Bank peer-to-peer loan risk assessment method and system based on multi-data analysis and reasoning

Abstract

The invention relates to a bank public loan risk assessment method and system based on multi-data analysis reasoning, and the scheme comprises the steps: collecting the loan data of bank historical public loan customers, and carrying out the preprocessing of the loan data, and obtaining a first feature matrix; taking the first feature matrix as the input of a GCN model, and obtaining a second feature matrix containing spatial information; taking the processed second feature matrix as the input of a Transform model, and obtaining a spatio-temporal feature vector; a GCN model and a Transform model are trained, and a risk prediction model is optimized; and finally, newly added customer data is analyzed, and a risk level is output, so that the current credit risk level of the customer is inferred and analyzed through internal space information and time information implied in the neural network mining data.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for assessing the risk of bank corporate loans based on multi-data analysis and reasoning. Background Art

[0002] Traditionally, bank corporate loan risk assessment mainly relies on statistical inference methods, which analyze customers' financial indicators, such as debt-paying ability, earnings quality, profitability, etc., and establish mathematical models for credit rating and debt-paying ability assessment. However, these methods are usually limited by specific assumptions and model structures, and are difficult to adapt to different types of data or flexibly handle data changes. In addition, they require complex preprocessing steps when dealing with missing values ​​and outliers, which may affect the accuracy of the assessment results.

[0003] Application of deep learning in data analysis: With the improvement of computing resources and the development of deep learning algorithms, deep learning has achieved great success in natural language processing, computer vision and other fields. In the field of big data analysis, deep learning has also been applied to a variety of practical tasks such as recommendation systems, autonomous driving, and medical diagnosis. This shows that deep learning has powerful data analysis and reasoning capabilities, which can assist banks in risk assessment of corporate loans and improve loan quality.

[0004] The application of graph convolutional network (GCN) in graph data processing. GCN is a powerful tool for processing graph data in deep learning. It builds a graph structure, defines entities as nodes of the graph, and edges represent the relationship between nodes. It learns the function mapping that combines node features with adjacent node features to extract the structural information of the graph. GCN successfully generalizes the convolution operation from traditional data to graph data, but GCN has limitations in capturing long-distance dependencies and global temporal information.

[0005] The application of recurrent neural network (RNN) in time series data. RNN is a neural network structure specially designed to process data with time sequence or logical sequence, and its design purpose is to capture the temporal dependencies in the sequence. However, RNN may encounter the problem of gradient disappearance or explosion when processing long sequence data, and it is difficult to capture long-term dependencies.

[0006] The application of the Transformer model in sequence prediction. The Transformer model abandons the sequence processing method of RNN by introducing the attention mechanism, realizes fast parallel computing, and solves some training problems of RNN. Although the Transformer performs well in time series prediction, in the field of bank corporate loan risk assessment, there is a lack of customized Transformer models to fully utilize its ability to capture sequence information over a long period of time.

[0007] Existing technologies have the following limitations in risk assessment of bank loans: Traditional statistical methods cannot flexibly adapt to data changes and are sensitive to outliers. GCN is limited in capturing long-distance dependencies and global time information. RNN has gradient problems when processing long sequence data and has difficulty capturing long-term dependencies. Although the Transformer model performs well in time series prediction, it lacks application in the field of risk assessment of bank loans.

[0008] In summary, the existing technology has not fully utilized the advantages of deep learning in the field of bank corporate loan risk assessment, especially in capturing spatiotemporal information and processing long sequence data. The present invention aims to solve the limitations of the existing technology by combining the advantages of GCN and Transformer models and provide a more accurate and efficient risk assessment method. Summary of the invention

[0009] In order to solve the deficiencies of the prior art, the present invention not only demonstrates innovation in technical implementation, but also achieves a major breakthrough in the application level. Relying on the existing real customer data, this method can deeply mine the spatial and temporal information inherent in the data, thereby accurately assessing the customer's credit risk level. In the field of bank corporate loans, the application of the present invention greatly assists loan review personnel, provides quantitative customer information analysis, reduces the uncertainty of subjective judgment, and effectively improves the quality and accuracy of loan decisions. In addition, it not only optimizes the data processing process and improves the scientificity and accuracy of risk assessment, but also provides strong technical support for the steady development of bank loan business.

[0010] To achieve the above objectives, the technical solutions adopted by the present invention include:

[0011] A method for assessing the risk of corporate bank loans based on multi-data analysis and reasoning, characterized by comprising:

[0012] S1. Collect the historical lending data of corporate loan customers of the bank, and pre-process the lending data to obtain a first feature matrix;

[0013] Specifically, the lending data of corporate customers to whom the bank has issued loans are collected for pre-processing before and during the lending period. The lending data include six categories: debt-paying ability (current ratio, quick ratio, equity ratio, net cash flow from operating activities / liabilities), earnings quality (net income from operating activities / total profit, net income from value changes / total profit, net non-operating income and expenses / total profit), profitability (gross profit margin, net profit margin, net profit margin on total assets), operating capacity (inventory turnover, current asset turnover, fixed asset turnover, total asset turnover), capital structure (asset-liability ratio, current assets / total assets, non-current assets / total assets, current liabilities / total liabilities, non-current liabilities / total liabilities), and industry risk status.

[0014] For the first five categories of numerical data, a normalization method is used to process each data indicator in each category of data into [0,1] data, which accelerates the convergence speed, reduces the sensitivity of the model to small changes in the input data, and improves the generalization ability of the model; for discrete data such as industry risk status, One-Hot encoding is used to expand the values ​​of discrete features to Euclidean space for easy input into the GCN model, such as the dimension of month. Finally, the six categories of data after preprocessing generate 6 feature vectors, which are spliced ​​to obtain the first feature matrix.

[0015] S2. Using the first feature matrix as the input of the GCN model to obtain a second feature matrix containing spatial information;

[0016] Specifically, the first feature matrix obtained after preprocessing is used as the input of the GCN model to obtain the spatial relationship between different indicators of the customer, and the second feature matrix containing spatial information is output.

[0017] S3, using the processed second feature matrix as the input of the Transformer model to obtain a spatiotemporal feature vector;

[0018] Specifically, the second feature matrix output by the GCN model in different months is used as the input of the Transformer model to obtain the risk changes of customers between different months, and the output is a spatiotemporal feature vector containing rich spatial and temporal information;

[0019] S4. Use the loss function constraint to train the risk prediction model that combines the GCN model and the Transformer model, so that the labels generated by the risk prediction model are infinitely close to the actual risk level labels;

[0020] Specifically, the obtained spatiotemporal feature vector is used to infer a customer risk level, wherein the actual risk value of the existing customers is used as a label optimization model to obtain a risk prediction model that infers the actual customer risk level as much as possible;

[0021] S5. Use the risk prediction model obtained through training to evaluate the risk level of the lending data of new customers.

[0022] Specifically, recent data of corporate customers applying for loans is obtained once a month, and normalization and One-Hot encoding are used to process numerical data and discrete data respectively. The processed data is input into the trained risk prediction model for analysis and reasoning, and the risk level of corporate customers is output to assist bank auditors in credit rating customers, assessing their debt repayment ability, improving loan quality, and reducing bad debt risks.

[0023] Furthermore, the S1 preprocesses the lending data to obtain a first feature matrix, including normalizing numerical data in the lending data and processing discrete data using One-Hot encoding.

[0024] Furthermore, S2 uses the first feature matrix as the input of the GCN model to obtain a second feature matrix containing spatial information, including: the GCN model mines spatial relationships through graph convolution and activation function, and the specific expression formula is:

[0025] x L =σ(Wx L-1 A+x L-1 )

[0026] Where W is the convolution parameter, x L-1 is the first characteristic matrix, A is the predefined adjacency matrix, x L is the second feature matrix of the output, and σ(·) is the activation function.

[0027] Furthermore, the S2 uses the first feature matrix as an input of the GCN model to obtain a second feature matrix containing spatial information, and also includes: flattening the second feature matrix and splicing it to obtain graph features for input into the Transformer model for training.

[0028] Furthermore, S3 uses the processed second feature matrix as the input of the Transformer model to obtain a spatiotemporal feature vector, including: splicing the graph features at different times into position vectors in time sequence and performing position encoding. The specific expression formula is:

[0029] Z 0 =[x 0 E 0 ;x 1 E 1 ;…;x T E T ], where x T is the graph feature at time T, ET is the position encoding vector at time T, x 0 is a predefined learnable feature vector, and the obtained Z 0 is the input of the Transformer model;

[0030] Through the encoding layer of the Transformer model, the temporal relationship of graph features at different times is obtained. The specific encoding formula is as follows:

[0031]

[0032] Among them, LN(·) is layer normalization; MLP(·) is multi-layer perceptron; MSA(·) is multi-head self-attention.

[0033] Furthermore, the evaluation method also includes that when the correlation between the numerical indicators in the lending data is weak, the Transformer model can be directly used to directly process the first feature matrix.

[0034] Specifically, in terms of mining spatial relationships, the present invention uses GCN, namely Graph Convolutional Networks, which is a deep learning model specifically used to process graph structure data. Its core idea is to generalize the convolution operation from the traditional Euclidean data space to non-Euclidean data structures such as graphs, and extract the features of the graph by applying convolution operations on nodes. GCN focuses on local spatial relationships and can capture information propagation between local neighbors of nodes, but may be limited in processing global relationships; Transformer captures global spatial relationships through the self-attention mechanism, but is not as good as GCN in information propagation between local neighbors. It is more powerful in processing long-distance dependencies and global features.

[0035] Since the local correlation of different indicators of corporate customer data collected by the present invention is strong, GCN is used to focus on the information propagation between local neighbors; when the different indicators of the collected customer data are not closely related, the pure Transformer model can be considered to simultaneously mine spatial information and temporal information.

[0036] A bank corporate loan risk assessment system based on multi-data analysis and reasoning, characterized by comprising:

[0037] A collection and preprocessing module is used to collect the historical lending data of corporate loan customers of the bank, and preprocess the lending data to obtain a first feature matrix;

[0038] A GCN model processing module, used for taking the first feature matrix as an input of a GCN model to obtain a second feature matrix containing spatial information;

[0039] A Transformer processing module, used for taking the processed second feature matrix as an input of a Transformer model to obtain a spatiotemporal feature vector;

[0040] The training module is used to use the loss function constraint to train the risk prediction model that combines the GCN model and the Transformer model, so that the labels generated by the risk prediction model are infinitely close to the actual risk level labels;

[0041] The risk assessment module is used to assess the risk level of the lending data of new customers using the risk prediction model obtained through training.

[0042] The present invention also provides a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

[0043] The present invention also provides an electronic device, characterized in that it includes a processor and a memory;

[0044] The memory is used to store computer programs;

[0045] The processor is configured to execute the method according to any one of claims 1 to 6 by calling a stored computer program.

[0046] The present invention also relates to a computer program product, comprising a computer program and / or instructions, characterized in that the computer program and / or instructions implement any of the methods of claims 1 to 6 when executed by a processor.

[0047] The beneficial effects of the present invention are as follows: the present invention not only demonstrates innovation in technical implementation, but also achieves a major breakthrough at the application level. Relying on the existing real customer data, this method can deeply mine the spatial and temporal information inherent in the data, thereby accurately assessing the customer's credit risk level. In the field of corporate bank loans, the application of the present invention greatly assists loan reviewers, provides quantitative customer information analysis, reduces the uncertainty of subjective judgment, and effectively improves the quality and accuracy of loan decisions. In addition, it not only optimizes the data processing process and improves the scientificity and accuracy of risk assessment, but also provides strong technical support for the steady development of bank loan business, assists bank-side reviewers to credit rate customers, assess their debt repayment ability, improve loan quality, and reduce bad debt risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flow chart of a method for risk assessment of corporate bank loans based on multi-data analysis and reasoning.

[0049] Figure 2The present invention is a logical flow chart of the method for risk assessment of corporate bank loans based on multi-data analysis and reasoning.

[0050] Figure 3 This is a schematic diagram of the structure of the risk prediction model obtained through training in the present invention.

[0051] Figure 4 Schematic diagram of the coding layer function of the Transformer model of the present invention.

[0052] Figure 5 This is a schematic diagram of the risk assessment structure of corporate bank loans based on multi-data analysis and reasoning in the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0055] The following detailed description of the implementation examples of the present invention is given in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the invention.

[0056] according to Figure 1 and Figure 2 As shown in the figure, the loan data of corporate customers to whom the bank has issued loans before and during the loan period are collected for preprocessing, and the collected loan data are divided into two parts, one part is used to train the risk prediction model of the GCN model combined with the Transformer model, and the other part is used to detect and optimize the risk prediction model.

[0057] Figure 3It is a risk prediction model that combines the GCN model with the Transformer model. First, the node features x input at different times are input into the GCN model. According to the adjacency matrix A, the similarity relationship between the node features in the second feature matrix is ​​learned by convolution. The information between the nodes is transmitted to each other, and the spatial relationship between the node features in the second feature matrix is ​​mined; among them, the activation function can introduce nonlinear factors, so that the GCN model can learn and represent complex nonlinear relationships, improve the network's expression ability, and better mine the complex spatial relationship between node features; after multiple layers of convolution + activation operations, the node features are transmitted to each other, and a representation containing rich spatial information, namely the graph feature, is learned. Then, the graph features output by the GCN model are used as the input of the Transformer model. The Transformer model first encodes the positions of the graph features at different times according to the date to help the network process the correlation between different dates; then, the connection between the graph features is established through the Transformer encoding layer, and the relationship between the features of different dates is mined. The encoding layer of the Transformer model is as follows Figure 4 As shown in the figure, layer normalization and skip connection can promote more stable network training; multi-layer perceptron introduces activation function to facilitate mining complex time relationships; multi-head self-attention will generate multiple query features from the features of each moment, calculate the relationship between the features of each moment and the features of other moments, and integrate multiple correlations to help the network mine the time relationship between features at different moments. Finally, the output is a customer risk spatiotemporal feature vector y containing rich spatial and temporal information.

[0058] The specific steps are as follows:

[0059] Step S1: Collect the monthly data of corporate customers in recent years for pre-processing. The data includes six types of data: debt repayment ability, income quality, profitability, operating capacity, capital structure, and industry risk status. The data of each month is processed by the GCN model and used as the input of the Transformer model. The pre-processing method is as follows:

[0060] For numerical information such as solvency, earnings quality, profitability, operating capacity, and capital structure, a normalization method is used to process them into [0,1] data to improve the generalization ability of the model. For example, solvency, including the current ratio α 1 , quick ratio α 2 , property rights ratio α 3 , Net cash flow from operating activities / Debt α 4 , the normalization formula is as follows:

[0061]

[0062] After normalization Spliced ​​together to form the solvency feature vector x 1 , and so on, we can get the return quality characteristic vector x 2 , profitability characteristic vector x 3 , operating capability characteristic vector x 4 , capital structure characteristic vector x 5 ;

[0063] For discontinuous discrete data such as industry risk status, One-Hot encoding is used to obtain a feature vector that is easy to input into the model. For example, the industry risk status is low risk coded as 001, the industry risk status is medium risk coded as 010, and the industry risk status is high risk coded as 100. At any time, only one bit is 1, and the rest are 0. Through this encoding method, the industry risk status feature vector x is obtained. 6 In order to facilitate model training and the subsequent expansion of the fine-grainedness of indicators, all feature vectors obtained by encoding discrete data and numerical data are supplemented with 0 and expanded into 16-bit feature vectors, and the feature matrix x is obtained by splicing t , where t is the number of months.

[0064] Step S2: Use the first feature matrix obtained after preprocessing as the input of GCN to mine the spatial relationship within the feature matrix. The specific process is as follows:

[0065] Step S201: Customize a learnable adjacency matrix A, the adjacency matrix dimension is 6×6, and is initialized to all 1s;

[0066] Step S202: Perform graph convolution + activation operation. The specific formula is:

[0067] x L =σ(Wx L-1 A+x L-1 )

[0068] Where W is the convolution parameter, x L-1 is the first characteristic matrix of the input, A is the adjacency matrix, x L is the output second feature matrix, σ(·) is the activation function. After L layers of convolution + activation operations, a second feature matrix representation containing rich spatial information is obtained.

[0069] Step S203: Flatten and expand the second feature matrix to obtain the graph feature x T , which is convenient for input into the Transformer model for training.

[0070] Step S3: Input the graph features into the Transformer model to mine the temporal relationship within the features. The specific process is as follows:

[0071] Step S301: Concatenate the graph features at different times into a position vector in chronological order and perform position encoding to better mine time series information. A learnable feature vector x is defined as 0 , called the 0th feature vector, is input into the Transformer model together to obtain an unbiased spatiotemporal feature representation of customer risk. The specific encoding formula is as follows:

[0072] Z 0 =[x 0 E 0 ;x 1 E 1 ;…;x T E T ]

[0073] Among them, x T is the graph feature at time T, E T is the position encoding vector at time T, Z 0 It is the input of the Transformer model.

[0074] Step S302: Through the Transformer encoding layer, the temporal relationship of the graph features at different times is mined. The specific encoding formula is as follows:

[0075]

[0076] Among them, LN(·) is layer normalization, which promotes more stable network training; MLP(·) is a multi-layer perceptron, which facilitates the mining of complex temporal relationships; MSA(·) is multi-head self-attention, which integrates multiple correlations to help the network mine the temporal relationship between features at different moments. After L layers of Transformer encoding, a feature representation containing rich temporal information is obtained. The unbiased customer risk spatiotemporal feature representation x obtained after L layers of Transformer encoding 0 As the output y of the Transformer model.

[0077] Step S4: Train the risk prediction model combining GCN and Transformer model. The specific process is as follows:

[0078] Step S401: Based on the spatiotemporal feature representation y of customer risk, the feature dimension is compressed and the probability distribution Y of customer risk is inferred. pre ;

[0079] Step S402: Using the customer's real risk Y obtained by professional evaluation realAs labels, there are five levels: normal [1,0,0,0,0], attention [0,1,0,0,0], secondary [0,0,1,0,0], suspicious [0,0,0,1,0], and loss [0,0,0,0,1], which guide the risk prediction model to infer Y pre Infinitely close to Y real Specifically, the following loss function is used to constrain and train the model:

[0080] Loss = -log(P(Y real |Y pre ))

[0081] Among them, the smaller the loss, the higher the Y pre The closer the distribution is to the true label Y real .

[0082] Step S5: Customer risk level prediction:

[0083] (1) Simulate business scenario data processing and make predictions: First, collect monthly data of a company in the past year, perform One-Hot encoding on discrete data, and normalize numerical data to obtain the following table data:

[0084]

[0085] (2) The input feature matrix x obtained each month t Input the GCN model to obtain the graph convolution features x T ;

[0086] (3) The graph convolution features x obtained each month T Combined with eigenvector x 0 , after position encoding, it is input into the Transformer model to obtain the unbiased customer risk spatiotemporal feature representation y;

[0087] (4) Based on the y compressed feature dimension, the inferred probability distribution of customer risk is [0.6, 0.1, 0.1, 0.1, 0.1], that is, the customer’s current risk level: normal.

[0088] The embodiment of the present invention also provides a bank public loan risk assessment system based on multi-data analysis and reasoning, such as Figure 5 As shown, including:

[0089] A collection and preprocessing module is used to collect the historical lending data of corporate loan customers of the bank, and preprocess the lending data to obtain a first feature matrix;

[0090] A GCN model processing module, used for taking the first feature matrix as an input of a GCN model to obtain a second feature matrix containing spatial information;

[0091] A Transformer processing module, used for taking the processed second feature matrix as an input of a Transformer model to obtain a spatiotemporal feature vector;

[0092] The training module is used to use the loss function constraint to train the risk prediction model that combines the GCN model and the Transformer model, so that the labels generated by the risk prediction model are infinitely close to the actual risk level labels;

[0093] The risk assessment module is used to assess the risk level of the lending data of new customers using the risk prediction model obtained through training.

[0094] By using this system, the above-mentioned evaluation method can be executed and the corresponding technical effects can be achieved.

[0095] An embodiment of the present invention also provides a computer-readable storage medium capable of implementing all the steps in the method in the above embodiment, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, all the steps in the method in the above embodiment are implemented.

[0096] An embodiment of the present invention also provides an electronic device for executing the above method. As an implementation device of the method, the electronic device has at least a processor and a memory, and in particular, the memory stores data and related computer programs required for executing the method, and the processor calls the data and programs in the memory to execute all steps of the implementation method and obtains the corresponding technical effect.

[0097] Preferably, the electronic device may include a bus architecture, the bus may include any number of interconnected buses and bridges, and the bus will include various circuits linked together by one or more processors and memories. The bus may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are all well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the receiver and transmitter. The receiver and transmitter may be the same component, i.e., a transceiver, providing a unit for communicating with various other systems over a transmission medium. The processor is responsible for managing the bus and general processing, while the memory may be used to store data used by the processor when performing operations.

[0098] In addition, the electronic device may further include components such as a communication module, an input unit, an audio processor, a display, and a power supply. The processor (or controller, operation control) used may include a microprocessor or other processor device and / or a logic device, which receives input and controls the operation of various components of the electronic device; the memory may be one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, or other suitable devices, which may store the above-mentioned related data information, and may also store a program for executing related information, and the processor may execute the program stored in the memory to realize information storage or processing, etc.; the input unit is used to provide input to the processor, for example, it may be a key or a touch input device; the power supply is used to provide power to the electronic device; the display is used to display display objects such as images and text, for example, it may be an LCD display. The communication module is a transmitter / receiver that sends and receives signals via an antenna. The communication module (transmitter / receiver) is coupled to the processor to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal. Based on different communication technologies, multiple communication modules may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless local area network module, etc. The communication module (transmitter / receiver) is also coupled to a speaker and a microphone via an audio processor to provide an audio output via the speaker and receive an audio input from the microphone, thereby realizing a common telecommunication function. The audio processor may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor is also coupled to a central processor, so that recording can be performed on the machine through the microphone, and the sound stored on the machine can be played through the speaker.

[0099] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0101] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction system, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in a box or multiple boxes. Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the attached claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0103] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention 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 by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A risk assessment method for bank corporate loans based on multi-data analysis and reasoning, characterized in that: include: S1. Collect the historical lending data of corporate loan customers of the bank, and pre-process the lending data to obtain a first feature matrix; S2. Using the first feature matrix as the input of the GCN model to obtain a second feature matrix containing spatial information; S3, using the processed second feature matrix as the input of the Transformer model to obtain a spatiotemporal feature vector; S4. Use the loss function constraint to train the risk prediction model that combines the GCN model and the Transformer model, so that the labels generated by the risk prediction model are infinitely close to the actual risk level labels; S5. Use the risk prediction model obtained through training to evaluate the risk level of the lending data of new customers.

2. The method according to claim 1, characterized in that The S1 preprocesses the lending data to obtain a first feature matrix, including normalizing the numerical data in the lending data and processing discrete data using One-Hot encoding.

3. The method according to claim 2, characterized in that The S2 uses the first feature matrix as the input of the GCN model to obtain a second feature matrix containing spatial information, including: the GCN model mines spatial relationships through graph convolution and activation function, and the specific expression formula is: x L =σ(Wx L-1 A+x L-1 ) Where W is the convolution parameter, x L-1 is the first characteristic matrix, A is the predefined adjacency matrix, x L is the second feature matrix of the output, and σ(·) is the activation function.

4. The method according to claim 3, characterized in that The S2 uses the first feature matrix as the input of the GCN model to obtain a second feature matrix containing spatial information, and also includes: flattening the second feature matrix and splicing it to obtain graph features so as to facilitate input into the Transformer model for training.

5. The method according to any one of claims 1 to 4, characterized in that: The S3 uses the processed second feature matrix as the input of the Transformer model to obtain a spatiotemporal feature vector, including: splicing the graph features at different times into position vectors in time sequence and performing position encoding. The specific expression formula is: Z0=[x0E0;x1E1;…;x T E T ], where x T is the graph feature at time T, E T is the position encoding vector at time T, x0 is a predefined learnable feature vector, and the obtained Z0 is the input of the Transformer model; Through the encoding layer of the Transformer model, the temporal relationship of graph features at different times is obtained. The specific encoding formula is as follows: Among them, LN(·) is layer normalization; MLP(·) is multi-layer perceptron; MSA(·) is multi-head self-attention.

6. The method according to claim 1, characterized in that The evaluation method also includes, when the correlation between the numerical indicators in the lending data is weak, directly using the Transformer model to directly process the first feature matrix.

7. A bank corporate loan risk assessment system based on multi-data analysis and reasoning, characterized in that: include: A collection and preprocessing module is used to collect the historical lending data of corporate loan customers of the bank, and preprocess the lending data to obtain a first feature matrix; A GCN model processing module, used for taking the first feature matrix as an input of a GCN model to obtain a second feature matrix containing spatial information; A Transformer processing module, used for taking the processed second feature matrix as an input of a Transformer model to obtain a spatiotemporal feature vector; The training module is used to use the loss function constraint to train the risk prediction model that combines the GCN model and the Transformer model, so that the labels generated by the risk prediction model are infinitely close to the actual risk level labels; The risk assessment module is used to assess the risk level of the lending data of new customers using the risk prediction model obtained through training.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 6 by calling a stored computer program.

10. A computer program product comprising a computer program and / or instructions, characterized in that: When the computer program and / or the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images