Credit risk assessment methods, devices, equipment, and media based on computing power enterprises

Abstract

This application relates to the fields of risk assessment technology and artificial intelligence technology. It discloses a method, apparatus, equipment, and medium for credit risk assessment of computing power enterprises. The method includes: obtaining the contrastive loss of a neural network model on the training set using a contrastive loss function; obtaining the regression loss of the neural network model on the training set using a predefined regression loss function; generating the total loss of the neural network model based on the contrastive loss, regression loss, and total loss function; stopping training the neural network model and saving the trained model when the total loss meets a preset condition; inputting the feature vector of the computing power enterprise to be assessed into the trained neural network model; and generating the credit risk assessment result of the computing power enterprise to be assessed through the trained neural network model. This application can improve the efficiency of generating credit risk assessment results for computing power enterprises to be assessed.

Description

Technical Field

[0001] This application relates to the fields of risk assessment technology and artificial intelligence technology, and in particular to credit risk assessment methods, devices, equipment and media based on computing power enterprises. Background Technology

[0002] Corporate bonds are debt financing instruments issued by qualified enterprises to investors in accordance with the law. Essentially, they are debt instruments in which enterprises raise funds from investors and promise to pay interest and repay principal on schedule. In corporate bond-related credit risk management scenarios, accurately obtaining the credit risk assessment results of the enterprise to be evaluated is a crucial prerequisite for investment decisions, credit approval, and risk warning.

[0003] However, the current process for generating credit risk assessment results for companies with computing power to be evaluated is cumbersome, hindering the efficiency of such results generation. This is because existing technologies primarily employ manual assessment methods. These methods require manual collection and organization of company financial data, industry information, and bond maturity details, followed by qualitative analysis and subjective judgment by professionals based on experience. This consumes significant human and time resources, increasing the time required to generate credit risk assessment results and thus hindering efficiency. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for credit risk assessment of computing power enterprises, in order to solve the technical problem that the existing process for generating credit risk assessment results for computing power enterprises to be assessed is cumbersome and not conducive to improving the efficiency of generating credit risk assessment results.

[0005] In a first aspect, embodiments of this application provide a credit risk assessment method based on computing power enterprises, applied to electronic devices, the credit risk assessment method comprising: Obtain multiple default samples and multiple non-default samples, and combine the multiple different default samples and multiple different non-default samples into a training set; The neural network model is trained based on the training set. A predefined contrastive loss function is used to obtain the contrastive loss of the neural network model on the training set, and a predefined regression loss function is used to obtain the regression loss of the neural network model on the training set. The total loss of the neural network model is generated based on the contrastive loss, regression loss, and total loss function of the neural network model on the training set. When the total loss meets the preset conditions, stop training the neural network model and save the trained neural network model. The evaluation data of the computing power companies to be evaluated is obtained. Based on the evaluation data, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power companies to be evaluated is determined. The feature vector of the computing power companies to be evaluated is input into the trained neural network model. The credit risk assessment result of the computing power companies to be evaluated is generated through the trained neural network model.

[0006] In one possible implementation of the first aspect, obtaining multiple default samples and multiple non-default samples, and forming a training set from multiple different default samples and multiple different non-default samples, includes: The evaluation data and default labels of defaulting companies are used to form a default sample. The evaluation data of defaulting companies includes financial data, policy data, technical data, and capital data of defaulting companies. Defaulting companies are computing power companies that have defaulted on credit bonds. The non-default sample is composed of the evaluation data and non-default labels of non-default enterprises. The evaluation data of non-default enterprises includes financial data, policy data, technical data, and capital data of non-default enterprises. Non-default enterprises are computing power enterprises that have not defaulted on credit bonds. The training set consists of multiple different default samples and multiple different non-default samples.

[0007] In one possible implementation of the first aspect, stopping the training of the neural network model and saving the trained neural network model when the total loss meets a preset condition includes: With the goal of minimizing the total loss of the neural network model, an optimizer is used to iteratively update the model parameters of the neural network model through backpropagation; When the total loss of the neural network model is less than the preset loss, stop training the neural network model and save the trained neural network model.

[0008] In one possible implementation of the first aspect, the step of obtaining evaluation data of the computing power enterprise to be evaluated, determining the feature vector of the computing power enterprise to be evaluated based on the evaluation data, the hypergraph construction module, and the hypergraph neural network, inputting the feature vector of the computing power enterprise to be evaluated into the trained neural network model, and generating the credit risk assessment result of the computing power enterprise to be evaluated through the trained neural network model includes: Obtain evaluation data from the computing power companies to be evaluated. The evaluation data includes the financial data, policy data, technical data, and capital data of the computing power companies to be evaluated. Input the financial data, policy data, technical data, and capital data of the computing power enterprise to be evaluated into the hypergraph construction module, and generate the first, second, third, and fourth hypergraphs of the computing power enterprise to be evaluated, respectively. The first, second, third, and fourth hypergraphs of the computing power enterprise to be evaluated are input into the hypergraph neural network, and the feature vectors of the computing power enterprise to be evaluated are generated through the hypergraph neural network. The feature vectors of the computing power companies to be evaluated are input into the trained neural network model. The trained neural network model generates the credit risk assessment results of the computing power companies to be evaluated. The computing power companies to be evaluated are the computing power companies that need to be evaluated at present.

[0009] In one possible implementation of the first aspect, the loss function is defined as follows: ; This represents the contrastive loss of the neural network model on the training set. The larger the contrastive loss of the neural network model, the weaker its ability to distinguish between defaulted and non-defaulted samples; the smaller the contrastive loss of the neural network model, the stronger its ability to distinguish between defaulted and non-defaulted samples. This represents the original features of the i-th default sample under the encoder; This represents the enhanced feature of the i-th default sample under the encoder; This represents the original feature of the j-th non-default sample under the encoder; Represents the cosine similarity function; Temperature parameters used to control the smoothness of the distribution; This represents the size of the current training batch.

[0010] In one possible implementation of the first aspect, the regression loss function is defined as follows: ; This represents the regression loss of the neural network model on the training set; the larger the regression loss of the neural network model, the lower its predictive ability for default samples; the smaller the regression loss of the neural network model, the stronger its predictive ability for default samples. This represents the true label of the i-th default sample; This represents the predicted probability value that the i-th default sample is a default category; This represents the proportion of defaulted samples; This represents the proportion of samples that did not default. To adjust the parameters.

[0011] In one possible implementation of the first aspect, the total loss function is defined as follows: ; in, This represents the total loss value of the neural network model on the training set. The larger the total loss value of the neural network model, the worse its overall performance in terms of prediction ability and feature discrimination ability. The smaller the total loss value of the neural network model, the stronger its overall performance in terms of prediction ability and feature discrimination ability. A set of parameters representing a neural network model; : Represents the regression loss of the neural network model; : Represents the contrastive loss of the neural network model; It is the first weighting coefficient; It is the second weighting coefficient; It is the square of the L2 norm of the model parameters.

[0012] Secondly, embodiments of this application provide a credit risk assessment device based on computing power enterprises, applied to electronic devices, including: The first acquisition module is used to acquire multiple default samples and multiple non-default samples, and to form a training set by combining multiple different default samples and multiple different non-default samples. The second acquisition module is used to train the neural network model based on the training set, using a predefined contrastive loss function to obtain the contrastive loss of the neural network model on the training set, and using a predefined regression loss function to obtain the regression loss of the neural network model on the training set. The first generation module is used to generate the total loss of the neural network model based on the contrastive loss, regression loss and total loss function of the neural network model on the training set. The save module is used to stop training the neural network model and save the trained neural network model when the total loss meets the preset conditions. The second generation module is used to obtain the evaluation data of the computing power enterprise to be evaluated. Based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model. Through the trained neural network model, the credit risk assessment result of the computing power enterprise to be evaluated is generated.

[0013] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the credit risk assessment method described in the first aspect above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the credit risk assessment method described in the first aspect above.

[0015] Fifthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to execute the credit risk assessment method described in the first aspect.

[0016] The beneficial effects of the embodiments of this application are as follows: Firstly, when the total loss meets the preset conditions, the training of the neural network model is stopped, the trained neural network model is saved, the evaluation data of the computing power enterprise to be evaluated is obtained, and based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model, and the credit risk assessment result of the computing power enterprise to be evaluated is generated through the trained neural network model. Since the computing power enterprise to be evaluated does not require manual evaluation, the generation time of the credit risk assessment result of the computing power enterprise to be evaluated is reduced, which is conducive to improving the efficiency of obtaining the credit risk assessment result of the computing power enterprise to be evaluated. Secondly, the smaller the total loss value of the neural network model, the stronger the comprehensive performance of the neural network model in terms of prediction ability and feature discrimination ability. When the total loss meets the preset conditions, it means that the trained neural network model has converged. Stop training the neural network model and save the trained neural network model. Therefore, the credit risk assessment results of the computing power enterprise to be evaluated output by the trained neural network model are more consistent and robust, which can effectively improve the credibility of the credit risk assessment results of the computing power enterprise to be evaluated. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This application scenario diagram illustrates the credit risk assessment method provided in the embodiments of this application. Figure 2 This is a flowchart illustrating the credit risk assessment method provided in the embodiments of this application; Figure 3 A flowchart illustrating the implementation of S205 provided in this application embodiment; Figure 4 A schematic block diagram of a credit risk assessment device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. The terms "comprising," "including," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0022] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0023] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0024] The credit risk assessment method provided in this application can be applied to electronic devices, including but not limited to servers, mobile phones, tablets, wearable devices, vehicle-mounted devices, and laptops. This application does not impose any restrictions on the specific type of electronic device.

[0025] Please see Figure 1 , Figure 1 The application scenario diagram of the credit risk assessment method provided in the embodiments of this application is described in detail below: Electronic devices access a database to retrieve evaluation data for defaulting companies and non-defaulting companies. The evaluation data and default labels of defaulting companies are used to form a default sample. The evaluation data of defaulting companies includes financial data, policy data, technical data, and capital data of defaulting companies. Defaulting companies are computing power companies that have defaulted on credit bonds. The non-default sample is composed of the evaluation data and non-default labels of non-default enterprises. The evaluation data of non-default enterprises includes financial data, policy data, technical data, and capital data of non-default enterprises. Non-default enterprises are computing power enterprises that have not defaulted on credit bonds. The training set consists of multiple different default samples and multiple different non-default samples.

[0026] In this embodiment, the electronic device accesses the database to automatically read the evaluation data of defaulting companies and non-defaulting companies, which can effectively improve the efficiency of data acquisition for both.

[0027] Please see Figure 2 , Figure 2 This is a flowchart illustrating the credit risk assessment method provided in this application embodiment, which can be applied to electronic devices.

[0028] like Figure 2 As shown in the embodiments of this application, the credit risk assessment method includes the following steps, which are detailed below: S201, obtain multiple default samples and multiple non-default samples, and combine multiple different default samples and multiple different non-default samples into a training set; The step of obtaining multiple default samples and multiple non-default samples, and forming a training set from multiple different default samples and multiple different non-default samples, includes: The evaluation data and default labels of defaulting companies are used to form a default sample. The evaluation data of defaulting companies includes financial data, policy data, technical data, and capital data of defaulting companies. Defaulting companies are computing power companies that have defaulted on credit bonds. The non-default sample is composed of the evaluation data and non-default labels of non-default enterprises. The evaluation data of non-default enterprises includes financial data, policy data, technical data, and capital data of non-default enterprises. Non-default enterprises are computing power enterprises that have not defaulted on credit bonds. The training set consists of multiple different default samples and multiple different non-default samples.

[0029] For ease of explanation, the following example is provided: For example, there are 3 defaulting companies, namely the first defaulting company, the second defaulting company, and the third defaulting company; the evaluation data and default label of the first defaulting company constitute the first defaulting sample. The evaluation data and default labels of the second defaulting companies constitute the second default sample; The evaluation data and default labels of the third defaulting companies constitute the third default sample; There are four companies that have not defaulted: the first non-defaulting company, the second non-defaulting company, the third non-defaulting company, and the fourth non-defaulting company. The evaluation data and the non-defaulting label of the first non-defaulting company constitute the first non-defaulting sample. The evaluation data of the second non-defaulting enterprises and the non-defaulting label constitute the second non-defaulting sample; The evaluation data of the third non-defaulting enterprises and the non-defaulting label constitute the third non-defaulting sample; The evaluation data of the fourth non-defaulting enterprise and the non-defaulting label constitute the fourth non-defaulting sample.

[0030] The training set consists of the first default sample, the second default sample, the third default sample, the first non-default sample, the second non-default sample, the third non-default sample, and the fourth non-default sample.

[0031] Specifically, the evaluation data and default labels of defaulting companies form the default sample, and the evaluation data and non-default labels of non-defaulting companies form the non-default sample, including: The evaluation data of defaulting companies is obtained, including their financial, policy, technical, and capital data. These data are then input into a hypergraph construction module, which generates four hypergraphs for each defaulting company. These four hypergraphs are then input into a hypergraph neural network, which generates feature vectors for each defaulting company. Finally, these feature vectors and default labels are combined to form a default sample. The evaluation data of non-defaulting companies is obtained, including their financial, policy, technical, and capital data. These data are then input into a hypergraph construction module, which generates four hypergraphs for each non-defaulting company. These four hypergraphs are then input into a hypergraph neural network, which generates feature vectors for each non-defaulting company. Finally, these feature vectors and non-defaulting labels are combined to form a non-defaulting sample.

[0032] Among them, the hyperedge of the first hypergraph of defaulting companies is used to represent the pairwise relationships of defaulting companies such as financial transactions and fund lending; the hyperedge of the second hypergraph of defaulting companies is used to represent the group relationships of companies such as policy adaptation, policy coordination, and policy application; the hyperedge of the third hypergraph of defaulting companies is used to represent the relationships such as technology cooperation, technology sharing, and technology R&D alliances; and the hyperedge of the fourth hypergraph of defaulting companies is used to represent the relationships such as equity investment and financing, capital linkage, and listing coordination.

[0033] Among them, the node system of the first hypergraph, the second hypergraph, the third hypergraph, and the fourth hypergraph of the defaulting enterprise is consistent, and all adopt a heterogeneous node set, which includes entity nodes and tag nodes corresponding to semantic anchors.

[0034] Among them, entity nodes correspond to the bond-issuing entities and carry their relevant dynamic characteristics; semantic anchors are label-like carriers used in the risk association modeling process, carrying structured prior information such as administrative regions, technological generations, and capital connections, providing clear semantic references for the association between the bond-issuing entity and the external environment, and realizing the classification and anchoring of risk sources. The first, second, third, and fourth hypergraphs of defaulting enterprises do not have ordinary edges, but only connect various nodes through hyperedges. Relying on the characteristic that hyperedges can connect multiple nodes, complex relationships are accurately depicted.

[0035] The architecture of the hypergraph neural network is based on risk semantic constraints and multi-node interaction, and consists of an input layer and an output layer. The input layer is used to input the first hypergraph, the second hypergraph, the third hypergraph, and the fourth hypergraph of defaulting companies.

[0036] The output layer is used to generate the influence weight of each hyperedge of each hypergraph based on the first, second, third, and fourth hypergraphs of the defaulting enterprise and the influence weight generation model. Based on the influence weight of each hyperedge of each hypergraph and the fusion model, the feature vector of the defaulting enterprise is generated. The feature vector of the defaulting enterprise accurately reflects the core features of the defaulting enterprise in various dimensions, including both the basic features of the defaulting enterprise itself and the related features, thus achieving a comprehensive presentation of the features of the defaulting enterprise.

[0037] The influence weight generation model is defined as follows: ; in, For the defaulting company The first hypergraph Superedge pair The influence weight of the node increases with each node. The first hypergraph Superedge pair The stronger the influence of a node, the lower its influence weight; the stronger the influence of a node, the less influence it has. The first hypergraph Superedge pair The weaker the influence of each node; It is a natural exponential function used to map attention scores to non-negative values, ensuring that the weights are non-negative and facilitating normalization; It is a linear rectification activation function; It represents the time step from the moment a specific risk event occurs to the current observation point, and is used to measure the physical time taken for risk propagation; For the first time targeting defaulting companies The learnable time decay coefficient of a hypergraph determines the decay rate of the hypergraph over time, and is used to characterize the differentiated features of long-term policy effects and instantaneous market shocks. For the defaulting company The set of superedges of a hypergraph; The index of the superedge; The defaulting company's first The first hypergraph Superedge pair Attention coefficient of each node; For the defaulting company Attention parameter vectors output by each hypergraph; Transpose of a vector; For the first The first layer super edge eigenvectors For the first The first layer super edge eigenvectors.

[0038] The fusion model is defined as follows: ; This represents the comprehensive feature vector of defaulting enterprises. , , , These are the first hypergraph of the defaulting company, the second hypergraph of the defaulting company, the third hypergraph of the defaulting company, and the fourth hypergraph of the defaulting company.

[0039] This indicates a multi-channel splicing operator; For the defaulting company The set of superedges of a hypergraph; The index of the superedge; The defaulting company's first The first hypergraph Superedge pair Attention coefficient of each node; For the first The first layer super edge eigenvectors; It is a non-linear activation function.

[0040] S202, train the neural network model based on the training set, use a predefined contrastive loss function to obtain the contrastive loss of the neural network model on the training set, and use a predefined regression loss function to obtain the regression loss of the neural network model on the training set; The comparison loss function is defined as follows: ; This represents the contrastive loss of the neural network model on the training set; The larger the contrastive loss of a neural network model, the weaker its ability to distinguish between defaulted and non-defaulted samples; the smaller the contrastive loss of a neural network model, the stronger its ability to distinguish between defaulted and non-defaulted samples. This represents the original features of the i-th default sample under the encoder; This represents the enhanced feature of the i-th default sample under the encoder; This represents the original feature of the j-th non-default sample under the encoder; Represents the cosine similarity function; Temperature parameters used to control the smoothness of the distribution; This represents the size of the current training batch.

[0041] The regression loss function is defined as follows: ; This represents the regression loss of the neural network model on the training set; the larger the regression loss of the neural network model, the lower its predictive ability for default samples; the smaller the regression loss of the neural network model, the stronger its predictive ability for default samples. This represents the true label of the i-th default sample; This represents the predicted probability value that the i-th default sample is a default category; This represents the proportion of defaulted samples; This represents the proportion of samples that did not default. To adjust the parameters.

[0042] S203, Generate the total loss of the neural network model based on the contrastive loss, regression loss, and total loss function of the neural network model on the training set; The total loss function is defined as follows: ; in, This represents the total loss value of the neural network model on the training set. The larger the total loss value of the neural network model, the worse its overall performance in terms of prediction ability and feature discrimination ability. The smaller the total loss value of the neural network model, the stronger its overall performance in terms of prediction ability and feature discrimination ability. A set of parameters representing a neural network model; : Represents the regression loss of the neural network model; : Represents the contrastive loss of the neural network model; It is the first weighting coefficient; It is the second weighting coefficient; It is the square of the L2 norm of the model parameters.

[0043] S204, when the total loss meets the preset condition, stop training the neural network model and save the trained neural network model; Wherein, stopping the training of the neural network model and saving the trained neural network model when the total loss meets a preset condition includes: With the goal of minimizing the total loss of the neural network model, an optimizer is used to iteratively update the model parameters of the neural network model through backpropagation; When the total loss of the neural network model is less than the preset loss, stop training the neural network model and save the trained neural network model.

[0044] S205: Obtain the evaluation data of the computing power enterprise to be evaluated. Based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, determine the feature vector of the computing power enterprise to be evaluated. Input the feature vector of the computing power enterprise to be evaluated into the trained neural network model. Through the trained neural network model, generate the credit risk assessment result of the computing power enterprise to be evaluated.

[0045] Among them, the computing power enterprises to be evaluated are those whose computing power needs to be evaluated at present.

[0046] Among them, computing power companies refer to companies that use computing power as their core to provide computing resources and technical support to other industries or individuals.

[0047] The credit risk assessment method, after acquiring the evaluation data of the computing power enterprise to be evaluated, determining the feature vector of the computing power enterprise based on the evaluation data, the hypergraph construction module, and the hypergraph neural network, inputting the feature vector of the computing power enterprise to be evaluated into the trained neural network model, and generating the credit risk assessment result of the computing power enterprise to be evaluated through the trained neural network model, also includes: When the credit risk assessment result of the computing power enterprise to be evaluated is that there is credit risk, the contribution of the risk output is quantitatively ranked, and the feature with the highest contribution is taken as the risk cause. A display window is created to show the warning message and risk cause of the computing power enterprise to be evaluated. The risk cause is the key factor that leads to the computing power enterprise to be evaluated being judged to have credit risk.

[0048] Displaying early warning messages and corresponding risk causes for computing power companies to be evaluated can identify credit risks and provide explainable, traceable, and actionable evidence for risk control decisions, significantly improving the intelligence and practicality of credit risk management.

[0049] For ease of explanation, the following example is provided: For example, in regulatory and post-investment management scenarios, by clearly identifying the causes of risks, key monitoring can be conducted on the computing power companies to be evaluated, urging them to optimize operational efficiency, improve the utilization rate of computing power resources, and improve their ability to collect payments, thereby helping them to resolve risks in a timely manner and achieve sound development.

[0050] The process involves inputting the feature vectors of the computing power companies to be evaluated into a trained neural network model. The trained neural network model then generates credit risk assessment results for these companies, achieving automation, standardization, and precision in credit risk assessment. This effectively replaces the traditional manual assessment model, significantly improving assessment efficiency, reducing manual operation costs, and minimizing subjective judgment bias, thus ensuring the accuracy of credit risk assessment results for the computing power companies being evaluated.

[0051] For ease of explanation, the following example is provided: For example, when the government guides funding to support new digital infrastructure, it can select computing power companies with strong technical capabilities, standardized operations, and great development potential based on the credit risk assessment results of the companies to be evaluated. This will provide precise support for key areas such as the construction of intelligent computing centers, artificial intelligence research and development, and the deployment of computing power networks, thereby accelerating the implementation of digital economy infrastructure.

[0052] For example, in scenarios such as supply chain cooperation and investment due diligence, the credit risk assessment results of computing power companies awaiting evaluation can provide reliable references for partners, reduce information asymmetry, and promote fair competition and standardized development in the computing power industry. This intelligent evaluation method can guide social capital to flow efficiently and safely to high-quality computing power companies, demonstrating significant practical value.

[0053] The beneficial effects of the embodiments of this application are as follows: Firstly, when the total loss meets the preset conditions, the training of the neural network model is stopped, the trained neural network model is saved, the evaluation data of the computing power enterprise to be evaluated is obtained, and based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model, and the credit risk assessment result of the computing power enterprise to be evaluated is generated through the trained neural network model. Since the computing power enterprise to be evaluated does not require manual evaluation, the generation time of the credit risk assessment result of the computing power enterprise to be evaluated is reduced, which is conducive to improving the efficiency of obtaining the credit risk assessment result of the computing power enterprise to be evaluated. Secondly, the smaller the total loss value of the neural network model, the stronger the comprehensive performance of the neural network model in terms of prediction ability and feature discrimination ability. When the total loss meets the preset conditions, it means that the trained neural network model has converged. Stop training the neural network model and save the trained neural network model. Therefore, the credit risk assessment results of the computing power enterprise to be evaluated output by the trained neural network model are more consistent and robust, which can effectively improve the credibility of the credit risk assessment results of the computing power enterprise to be evaluated.

[0054] Please see Figure 3 , Figure 3 The implementation flowchart of S205 provided in the embodiments of this application is described in detail below: S301, Obtain the evaluation data of the computing power enterprise to be evaluated. The evaluation data includes the financial data, policy data, technical data, and capital data of the computing power enterprise to be evaluated. Among them, the hyperedges of the first hypergraph of the computing power enterprises to be evaluated are used to represent the pairwise relationships of financial transactions and fund lending among the computing power enterprises to be evaluated; the hyperedges of the second hypergraph of the computing power enterprises to be evaluated are used to represent the group relationships of policy adaptation, policy coordination, and policy application among the enterprises; the hyperedges of the third hypergraph of the computing power enterprises to be evaluated are used to represent the relationships of technological cooperation, technology sharing, and technology R&D alliances; and the hyperedges of the fourth hypergraph of the computing power enterprises to be evaluated are used to represent the relationships of equity investment and financing, capital linkage, and listing coordination.

[0055] S302, input the financial data, policy data, technical data, and capital data of the computing power enterprise to be evaluated into the hypergraph construction module, and generate the first hypergraph, the second hypergraph, the third hypergraph, and the fourth hypergraph of the computing power enterprise to be evaluated, respectively. S303, input the first hypergraph, the second hypergraph, the third hypergraph, and the fourth hypergraph of the computing power enterprise to be evaluated into the hypergraph neural network, and generate the feature vector of the computing power enterprise to be evaluated through the hypergraph neural network. S304. Input the feature vector of the computing power enterprise to be evaluated into the trained neural network model. Through the trained neural network model, generate the credit risk assessment result of the computing power enterprise to be evaluated. The computing power enterprise to be evaluated is the computing power enterprise that needs to be evaluated at present.

[0056] In this embodiment, the feature vector of the computing power enterprise to be evaluated is generated through a hypergraph neural network, which fully leverages the powerful modeling capabilities of hypergraph neural networks for high-dimensional, heterogeneous, and interconnected complex data. Hypergraph neural networks can effectively mine deep relationships between multi-dimensional data such as enterprise financial data, bond information, industry correlations, and risk transmission, thereby generating feature vectors of the computing power enterprise to be evaluated that are more expressive and complete. This avoids the information loss and insufficient representation problems of traditional feature extraction methods. Inputting the more expressive and complete feature vectors of the computing power enterprise to be evaluated into the trained neural network model can significantly improve the prediction accuracy of the neural network model for credit risk, thus improving the accuracy of the credit risk assessment results for the computing power enterprise to be evaluated.

[0057] For the credit risk assessment method described in the above examples, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic block diagram of a credit risk assessment device provided in an embodiment of this application. Figure 4 The credit risk assessment device 400 shown can be applied to, for example... Figure 1 The application scenario diagram shows electronic devices. The following section uses electronic devices as an example to illustrate this. Figure 4 The credit risk assessment device 400 shown will be described in detail. The credit risk assessment device 400 may include a first acquisition module 401, a second acquisition module 402, a first generation module 403, a storage module 404, and a second generation module 405.

[0058] The first acquisition module 401 is used to acquire multiple default samples and multiple non-default samples, and to form a training set by combining multiple different default samples and multiple different non-default samples. The second acquisition module 402 is used to train the neural network model based on the training set, obtain the contrastive loss of the neural network model on the training set using a predefined contrastive loss function, and obtain the regression loss of the neural network model on the training set using a predefined regression loss function. The first generation module 403 is used to generate the total loss of the neural network model based on the contrastive loss of the neural network model on the training set, the regression loss of the neural network model on the training set, and the total loss function; The save module 404 is used to stop training the neural network model and save the trained neural network model when the total loss meets the preset conditions. The second generation module 405 is used to obtain the evaluation data of the computing power enterprise to be evaluated. Based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model. Through the trained neural network model, the credit risk assessment result of the computing power enterprise to be evaluated is generated.

[0059] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0060] The beneficial effects of the embodiments of this application are as follows: Firstly, when the total loss meets the preset conditions, the training of the neural network model is stopped, the trained neural network model is saved, the evaluation data of the computing power enterprise to be evaluated is obtained, and based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model, and the credit risk assessment result of the computing power enterprise to be evaluated is generated through the trained neural network model. Since the computing power enterprise to be evaluated does not require manual evaluation, the generation time of the credit risk assessment result of the computing power enterprise to be evaluated is reduced, which is conducive to improving the efficiency of obtaining the credit risk assessment result of the computing power enterprise to be evaluated. Secondly, the smaller the total loss value of the neural network model, the stronger the comprehensive performance of the neural network model in terms of prediction ability and feature discrimination ability. When the total loss meets the preset conditions, it means that the trained neural network model has converged. Stop training the neural network model and save the trained neural network model. Therefore, the credit risk assessment results of the computing power enterprise to be evaluated output by the trained neural network model are more consistent and robust, which can effectively improve the credibility of the credit risk assessment results of the computing power enterprise to be evaluated.

[0061] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0062] like Figure 5 As shown, Figure 5The electronic device includes: at least one processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor 20, wherein the processor 20 executes the computer program 22 to implement the steps in any of the above method embodiments.

[0063] The electronic device may include, but is not limited to, processor 20 and memory 21. Those skilled in the art will understand that... Figure 5 This is merely an example of an electronic device and does not constitute a limitation on electronic devices. It may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0064] The processor 20 is used to run a computer program 22 stored in the memory 21, and performs the following steps when executing the computer program 22: Obtain multiple default samples and multiple non-default samples, and combine the multiple different default samples and multiple different non-default samples into a training set; The neural network model is trained based on the training set. A predefined contrastive loss function is used to obtain the contrastive loss of the neural network model on the training set, and a predefined regression loss function is used to obtain the regression loss of the neural network model on the training set. The total loss of the neural network model is generated based on the contrastive loss, regression loss, and total loss function of the neural network model on the training set. When the total loss meets the preset conditions, stop training the neural network model and save the trained neural network model. The evaluation data of the computing power companies to be evaluated is obtained. Based on the evaluation data, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power companies to be evaluated is determined. The feature vector of the computing power companies to be evaluated is input into the trained neural network model. The credit risk assessment result of the computing power companies to be evaluated is generated through the trained neural network model.

[0065] The processor 20 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors, field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0066] In some embodiments, the memory 21 may be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device. In other embodiments, the memory 21 may also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the electronic device.

[0067] Furthermore, the memory 21 may include both internal storage units and external storage devices of the electronic device. The memory 21 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 21 can also be used to temporarily store data that has been output or will be output.

[0068] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0069] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0070] The computer-readable storage medium may also be an external storage device of a credit risk assessment device or electronic device, such as a plug-in hard drive, a smart media card (SMC), a secure digital card (SD), a flash card, or a non-transitory computer-readable storage medium.

[0071] Since the computer program stored in the computer-readable storage medium can execute any of the credit risk assessment methods for computing power-based enterprises provided in the embodiments of this application, the computer-readable storage medium can achieve the beneficial effects that any of the credit risk assessment methods for computing power-based enterprises provided in the embodiments of this application can achieve, as detailed in the preceding embodiments, and will not be repeated here.

[0072] This application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the credit risk assessment method described above.

[0073] When a computer program is loaded into an electronic device, it can perform the following steps: Obtain multiple default samples and multiple non-default samples, and combine the multiple different default samples and multiple different non-default samples into a training set; The neural network model is trained based on the training set. A predefined contrastive loss function is used to obtain the contrastive loss of the neural network model on the training set, and a predefined regression loss function is used to obtain the regression loss of the neural network model on the training set. The total loss of the neural network model is generated based on the contrastive loss, regression loss, and total loss function of the neural network model on the training set. When the total loss meets the preset conditions, stop training the neural network model and save the trained neural network model. The evaluation data of the computing power companies to be evaluated is obtained. Based on the evaluation data, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power companies to be evaluated is determined. The feature vector of the computing power companies to be evaluated is input into the trained neural network model. The credit risk assessment result of the computing power companies to be evaluated is generated through the trained neural network model.

[0074] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0075] Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium includes: an entity or device for carrying computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium.

[0076] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0077] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A credit risk assessment method based on computing power enterprises, characterized in that, The credit risk assessment method, applied to electronic devices, includes: Obtain multiple default samples and multiple non-default samples, and combine the multiple different default samples and multiple different non-default samples into a training set; The neural network model is trained based on the training set. A predefined contrastive loss function is used to obtain the contrastive loss of the neural network model on the training set, and a predefined regression loss function is used to obtain the regression loss of the neural network model on the training set. The total loss of the neural network model is generated based on the contrastive loss, regression loss, and total loss function of the neural network model on the training set. When the total loss meets the preset conditions, stop training the neural network model and save the trained neural network model. The evaluation data of the computing power companies to be evaluated is obtained. Based on the evaluation data, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power companies to be evaluated is determined. The feature vector of the computing power companies to be evaluated is input into the trained neural network model. The credit risk assessment result of the computing power companies to be evaluated is generated through the trained neural network model.

2. The credit risk assessment method according to claim 1, characterized in that, The process of obtaining multiple default samples and multiple non-default samples, and combining these multiple different default samples and multiple different non-default samples into a training set, includes: The evaluation data and default labels of defaulting companies are used to form a default sample. The evaluation data of defaulting companies includes financial data, policy data, technical data, and capital data of defaulting companies. Defaulting companies are computing power companies that have defaulted on credit bonds. The non-default sample is composed of the evaluation data and non-default labels of non-default enterprises. The evaluation data of non-default enterprises includes financial data, policy data, technical data, and capital data of non-default enterprises. Non-default enterprises are computing power enterprises that have not defaulted on credit bonds. The training set consists of multiple different default samples and multiple different non-default samples.

3. The credit risk assessment method according to claim 1, characterized in that, The step of stopping the training of the neural network model and saving the trained neural network model when the total loss meets a preset condition includes: With the goal of minimizing the total loss of the neural network model, an optimizer is used to iteratively update the model parameters of the neural network model through backpropagation; When the total loss of the neural network model is less than the preset loss, stop training the neural network model and save the trained neural network model.

4. The credit risk assessment method according to claim 1, characterized in that, The process involves acquiring evaluation data of the computing power enterprise to be evaluated, determining its feature vector based on the evaluation data, the hypergraph construction module, and the hypergraph neural network, inputting the feature vector into the trained neural network model, and generating a credit risk assessment result for the computing power enterprise through the trained neural network model. This includes: Obtain evaluation data from the computing power companies to be evaluated. The evaluation data includes the financial data, policy data, technical data, and capital data of the computing power companies to be evaluated. Input the financial data, policy data, technical data, and capital data of the computing power enterprise to be evaluated into the hypergraph construction module, and generate the first, second, third, and fourth hypergraphs of the computing power enterprise to be evaluated, respectively. The first, second, third, and fourth hypergraphs of the computing power enterprise to be evaluated are input into the hypergraph neural network, and the feature vectors of the computing power enterprise to be evaluated are generated through the hypergraph neural network. The feature vectors of the computing power companies to be evaluated are input into the trained neural network model. The trained neural network model generates the credit risk assessment results of the computing power companies to be evaluated. The computing power companies to be evaluated are the computing power companies that need to be evaluated at present.

5. The credit risk assessment method according to claim 1, characterized in that, The comparison loss function is defined as follows: ； This represents the contrastive loss of the neural network model on the training set; The larger the contrastive loss of a neural network model, the weaker its ability to distinguish between defaulted and non-defaulted samples; the smaller the contrastive loss of a neural network model, the stronger its ability to distinguish between defaulted and non-defaulted samples. This represents the original features of the i-th default sample under the encoder; This represents the enhanced feature of the i-th default sample under the encoder; This represents the original feature of the j-th non-default sample under the encoder; Represents the cosine similarity function; Temperature parameters used to control the smoothness of the distribution; This represents the size of the current training batch.

6. The credit risk assessment method according to claim 1, characterized in that, The regression loss function is defined as follows: ； This represents the regression loss of the neural network model on the training set; the larger the regression loss of the neural network model, the lower its predictive ability for default samples; the smaller the regression loss of the neural network model, the stronger its predictive ability for default samples. This represents the true label of the i-th default sample; This represents the predicted probability value that the i-th default sample is a default category; This represents the proportion of defaulted samples; This represents the proportion of samples that did not default. To adjust the parameters.

7. The credit risk assessment method according to claim 1, characterized in that, The total loss function is defined as follows: ； in, This represents the total loss value of the neural network model on the training set. The larger the total loss value of the neural network model, the worse its overall performance in terms of prediction ability and feature discrimination ability. The smaller the total loss value of the neural network model, the stronger its overall performance in terms of prediction ability and feature discrimination ability. A set of parameters representing a neural network model; : Represents the regression loss of the neural network model; : Represents the contrastive loss of the neural network model; It is the first weighting coefficient; It is the second weighting coefficient; It is the square of the L2 norm of the model parameters.

8. A credit risk assessment device based on computing power enterprises, characterized in that, Applied to electronic devices, including: The first acquisition module is used to acquire multiple default samples and multiple non-default samples, and to form a training set by combining multiple different default samples and multiple different non-default samples. The second acquisition module is used to train the neural network model based on the training set, using a predefined contrastive loss function to obtain the contrastive loss of the neural network model on the training set, and using a predefined regression loss function to obtain the regression loss of the neural network model on the training set. The first generation module is used to generate the total loss of the neural network model based on the contrastive loss, regression loss and total loss function of the neural network model on the training set. The save module is used to stop training the neural network model and save the trained neural network model when the total loss meets the preset conditions. The second generation module is used to obtain the evaluation data of the computing power enterprise to be evaluated. Based on the evaluation data of the computing power enterprise to be evaluated, the hypergraph construction module and the hypergraph neural network, the feature vector of the computing power enterprise to be evaluated is determined. The feature vector of the computing power enterprise to be evaluated is input into the trained neural network model. Through the trained neural network model, the credit risk assessment result of the computing power enterprise to be evaluated is generated.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the credit risk assessment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the credit risk assessment method as described in any one of claims 1 to 7.

Images