Enterprise credit evaluation method and device, equipment, storage medium and program product
By using a deep learning model that integrates a weighted bidirectional feature pyramid network, the enterprise credit assessment process is automated, solving the problem of low efficiency in manual assessment and improving assessment efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA UNITED NETWORK COMM GRP CO LTD
- Filing Date
- 2025-12-31
- Publication Date
- 2026-04-28
AI Technical Summary
Current technologies for corporate credit assessment mainly rely on manual analysis, resulting in low assessment efficiency.
A pre-trained deep learning model with a fusion weighted bidirectional feature pyramid network is used to evaluate corporate credit. An evaluation matrix is generated through target evaluation information, feature extraction and credit scoring are performed, and the evaluation result is finally determined.
It has achieved full automation of the process from data input to credit decision-making, improving the efficiency and accuracy of corporate credit assessment.
Smart Images
Figure CN121935532A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, device, storage medium, and program product for assessing corporate credit. Background Technology
[0002] Corporate credit assessment is a fundamental risk measurement and trust-building mechanism in the modern economic system. Its core task is to provide objective evidence for key economic activities such as credit approval, supply chain finance, and investment decisions by quantitatively analyzing the creditworthiness of market entities.
[0003] Currently, corporate credit assessments mainly rely on assessment experts to manually analyze and comprehensively judge corporate information, resulting in low assessment efficiency. Summary of the Invention
[0004] This application provides a method, apparatus, device, storage medium, and program product for assessing corporate credit, in order to solve the problem that the existing technology for assessing corporate credit mainly relies on assessment experts to manually analyze and comprehensively judge corporate information, resulting in low assessment efficiency.
[0005] In a first aspect, embodiments of this application provide a method for assessing corporate credit, comprising:
[0006] In response to receiving a corporate credit assessment request, obtain the target assessment information of the target company;
[0007] Based on the target evaluation information of the target enterprise, a target evaluation matrix for the target enterprise is determined; the target evaluation matrix is a multi-dimensional matrix.
[0008] The target credit score of the target enterprise is determined by employing a target credit assessment model and based on the target assessment matrix; the target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network; the target credit assessment model is pre-trained.
[0009] The target evaluation result of the target enterprise is determined by adopting a preset evaluation strategy and based on the target credit score.
[0010] In one possible implementation, determining the target evaluation matrix of the target enterprise based on the target enterprise's target evaluation information includes:
[0011] The initial evaluation matrix is determined using a preset quantification algorithm and based on the target evaluation information of the target enterprise;
[0012] The initial evaluation matrix is multiplied by a preset correction factor to obtain the enterprise's target evaluation matrix.
[0013] In one possible implementation, the step of employing a target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix includes:
[0014] A deformable convolutional network is used to extract features from the target evaluation matrix to obtain a first feature dataset.
[0015] A fused weighted bidirectional feature pyramid network is used, and a second feature dataset is determined based on the first feature dataset;
[0016] The target credit score of the target enterprise is determined based on the second feature dataset.
[0017] In one possible implementation, before employing the target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix, the method further includes:
[0018] Obtain a target sample set; the target sample set includes target evaluation sample data from multiple sample enterprises; the target evaluation sample data from the sample enterprises includes target image data and target label data;
[0019] The target sample set is divided into a training set, a validation set, and a test set according to a preset ratio;
[0020] The training set and the validation set are input into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model.
[0021] The target credit assessment model is determined based on the test set and the trained initial credit assessment model.
[0022] In one possible implementation, obtaining the target sample set includes:
[0023] Obtain target evaluation sample information from multiple sample companies;
[0024] Based on the target evaluation sample information, the corresponding target evaluation matrix is determined;
[0025] The target image data is determined based on the target evaluation matrix.
[0026] Extract a predetermined number of original label data from each of the target evaluation sample information;
[0027] A preset numerical strategy is used to convert a preset number of original label data corresponding to each target evaluation sample information into target label data;
[0028] The target image data and the corresponding target label data are determined as the target evaluation sample data for each sample enterprise;
[0029] The target evaluation sample data of each of the sample enterprises is determined as the target sample set.
[0030] In one possible implementation, the step of inputting the training set and the validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model includes:
[0031] Repeat the following operations until the initial credit assessment model completes the preset number of training rounds, and output the initial credit assessment model corresponding to the minimum validation loss in each training round as the trained initial credit assessment model. The following operations include:
[0032] Initialize the target loss weights;
[0033] The training set is input into the initial credit assessment model for training to obtain the predicted credit scores of each of the sample enterprises.
[0034] The training loss is determined by using a preset loss function and based on the target loss weight, the target label data of each sample enterprise, and the corresponding predicted credit score.
[0035] The backpropagation algorithm is used to update the parameters of the initial credit assessment model and the target loss weights based on the training loss.
[0036] The validation loss is determined based on the preset loss function and the validation set;
[0037] Save the initial credit assessment model for the corresponding round that minimizes the current verification loss.
[0038] Secondly, embodiments of this application provide a device for assessing corporate credit, comprising:
[0039] The acquisition module is used to obtain the target assessment information of the target company in response to receiving a corporate credit assessment request;
[0040] The determination module is used to determine the target evaluation matrix of the target enterprise based on the target evaluation information of the target enterprise; the target evaluation matrix is a multi-dimensional matrix;
[0041] The determination module is further configured to determine the target credit score of the target enterprise based on the target credit assessment matrix using a target credit assessment model; the target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network; the target credit assessment model is pre-trained;
[0042] The determination module is also used to determine the target evaluation result of the target enterprise based on the target credit score using a preset evaluation strategy.
[0043] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;
[0044] The memory stores computer-executed instructions;
[0045] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.
[0046] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.
[0047] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.
[0048] The enterprise credit assessment method, apparatus, device, storage medium, and program product provided in this application embodiment, in response to receiving an enterprise credit assessment request, obtains the target assessment information of the target enterprise, determines the target assessment matrix of the target enterprise based on the target assessment information, then inputs the target assessment matrix into the target credit assessment model, outputs the target credit score of the target enterprise, further adopts a preset assessment strategy and determines the target assessment result of the target enterprise based on the target credit score. The target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network, realizing full-process automation from data input to credit decision-making, and improving the efficiency and accuracy of enterprise credit assessment. Attached Figure Description
[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0050] Figure 1 This is a diagram illustrating an application scenario for the enterprise credit assessment method applicable to the embodiments of this application.
[0051] Figure 2 A flowchart illustrating a method for assessing corporate credit provided in an embodiment of this application;
[0052] Figure 3 A flowchart of a method for assessing corporate credit provided in another embodiment of this application;
[0053] Figure 4 A schematic diagram of the structure of the enterprise credit assessment device provided in this application;
[0054] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.
[0055] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0056] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0057] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0058] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0059] Corporate credit assessment is a fundamental risk measurement and trust-building mechanism in the modern economic system. Its core task is to provide objective evidence for key economic activities such as credit approval, supply chain finance, and investment decisions by quantitatively analyzing the creditworthiness of market entities. With the rapid growth in the number of businesses, the increasing diversity of business models, and the continuous expansion of data dimensions, traditional assessment methods have gradually revealed limitations in processing capacity and response speed when faced with massive, heterogeneous, and dynamically changing information. Currently, corporate credit assessment mainly relies on manual analysis and comprehensive judgment of corporate information by assessment experts, resulting in low assessment efficiency.
[0060] To address the aforementioned technical problems, this application proposes the following technical concept: A pre-trained target credit assessment model is used to conduct enterprise credit assessments on target enterprises. This target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network. Specifically, upon receiving an enterprise credit assessment request, the server responds by acquiring the target enterprise's target assessment information and determining the target enterprise's target assessment matrix based on this information. This target assessment matrix is a multi-dimensional matrix. The target credit assessment model is then used to determine the target enterprise's target credit score based on the target assessment matrix. Finally, based on the target credit score and a preset assessment strategy, the target assessment result for the target enterprise is determined. Compared to manual credit assessment by experts, this application automates the entire process from data input to credit decision-making, improving the efficiency of enterprise credit assessment. Furthermore, the target credit assessment model can automatically learn and combine the most effective features for credit judgment from the target enterprise's target assessment matrix, thereby enhancing the objectivity and accuracy of the assessment.
[0061] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0062] Figure 1 This is an application scenario diagram of the enterprise credit assessment method applicable to the embodiments of this application, such as... Figure 1 As shown. The system corresponding to the enterprise credit assessment method in this embodiment may include: a client device 20 and a server device 10. A user triggers an enterprise credit assessment request through the client device 20, which then sends the request to the server device 10. Upon receiving the enterprise credit assessment request, the server device 10 obtains the target assessment information of the target enterprise. The server device 10 determines the target assessment matrix of the target enterprise based on the target assessment information, wherein the target assessment matrix is a multi-dimensional matrix. The server device 10 uses a target credit assessment model and determines the target credit score of the target enterprise based on the target assessment matrix, wherein the target credit assessment model is a pre-trained deep learning model that integrates a weighted bidirectional feature pyramid network. Further, the server device 10 uses a preset assessment strategy and determines the target assessment result of the target enterprise based on the target credit score. The server device 10 sends the target assessment result of the target enterprise to the client device 20 for display.
[0063] Figure 2 A flowchart of a corporate credit assessment method provided in one embodiment of this application is shown below. Figure 2As shown, the execution subject of this embodiment is a corporate credit assessment device. This corporate credit assessment device can be implemented through a computer program, or through a medium storing the relevant computer program, such as a USB flash drive and / or optical disc; alternatively, it can be implemented through a physical device that integrates or installs the relevant computer program, such as an electronic device. The corporate credit assessment method provided in this embodiment includes the following steps:
[0064] S201: In response to receiving a corporate credit assessment request, obtain the target assessment information of the target company.
[0065] The target company is the company to be assessed for creditworthiness. The assessment information consists of multi-dimensional information about the target company.
[0066] Optionally, the target enterprise's target evaluation information may include business registration information, tax information, balance sheet information, profit and loss statement, etc., which are not limited in this embodiment.
[0067] It is understood that, in this embodiment, the data sources for the target enterprise's target evaluation information include, but are not limited to, obtaining it in text form and obtaining it by calling the enterprise's website interface.
[0068] Specifically, in this embodiment, target evaluation information of the target enterprise is obtained from a preset database.
[0069] S202: Determine the target company's target evaluation matrix based on the target company's target evaluation information.
[0070] The target evaluation matrix is a multidimensional matrix.
[0071] As an optional implementation, the target evaluation matrix of the target enterprise can be determined based on the target enterprise's target evaluation information using the following steps: determining an initial evaluation matrix using a preset quantification algorithm and based on the target enterprise's target evaluation information; and multiplying the initial evaluation matrix with a preset correction factor to obtain the enterprise's target evaluation matrix.
[0072] The initial evaluation matrix is a matrix obtained by quantifying the evaluation information of the target company.
[0073] Specifically, in this embodiment, the target enterprise's target evaluation information is an evaluation information matrix, which can be... Where N represents the total number of dimensions describing the target evaluation information, This represents the individual information corresponding to each dimension of the target evaluation information. Then, for each line in the target evaluation information matrix... Quantization is performed according to a preset quantization strategy to obtain the corresponding quantized value. This allows us to obtain the quantified evaluation information matrix. Then, we solve... The maximum number of elements M in the matrix is then used to supplement the matrix dimension of the quantized evaluation information matrix when the number of columns is insufficient. Data, through supplementation The zeros adjust the matrix dimension of the quantified evaluation information matrix to... Thus, a preliminary evaluation matrix is obtained.
[0074] Specifically, finding the maximum value of array M is expressed as:
[0075]
[0076] The primary evaluation matrix has N×M dimensions.
[0077] In this embodiment, n is the index of a specific dimension in the target evaluation information.
[0078] For example, the evaluation information matrix corresponding to the target company's target evaluation information can include the company's business registration information. Company litigation information The company has been established for more than 24 months. Registration address information Is the company listed? Corporate tax information Value-added tax paid in the past year Corporate income tax paid in the past year There are tax filing records in the past 3 months. Company Industry Sector Enterprise type Company growth Tax nature of enterprises wait.
[0079] Optionally, the preset quantification strategy includes a strategy for quantifying each piece of information in each dimension of the assessment information matrix corresponding to the target enterprise's target assessment information. The quantification strategy varies depending on the different pieces of information in each dimension and can be set independently according to requirements. This embodiment does not impose any limitations on this strategy.
[0080] For example, The corresponding quantitative rule is: if the company has litigation information, then For example, if the company has no litigation information, then... Let's say the target company has litigation information. Let's say it's a.
[0081] Optionally, a and b are hyperparameters, and the values of a and b are greater than or equal to 0 and less than or equal to 255. In this embodiment, the specific values of a and b are not limited and can be set by the developers.
[0082] Understandable The values in the range all satisfy the condition that they are greater than or equal to 0 and less than or equal to 255.
[0083] Furthermore, the initial evaluation matrix is multiplied by a preset correction factor to obtain the enterprise's target evaluation matrix.
[0084] Optionally, the preset correction factor can be set independently as needed, and this embodiment does not impose any limitations.
[0085] For example, the specific calculation process of the target evaluation matrix is as follows:
[0086]
[0087] In the formula, This represents the Hadamard product operation; This represents the preset correction factor. This represents the initial evaluation matrix. This represents the target evaluation matrix.
[0088] S203: The target credit rating of the target enterprise is determined by using the target credit rating model and the target rating matrix.
[0089] Specifically, in this embodiment, the target evaluation matrix is converted into an image format and input into the target credit evaluation model, and the target credit score of the target enterprise is output.
[0090] Understandably, the target evaluation matrix is converted into an image format, and the matrix is normalized to a value between 0 and 255.
[0091] The target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network. The target credit assessment model is pre-trained.
[0092] As an optional implementation, the following steps can be used to employ a target credit assessment model and determine the target credit score of the target enterprise based on the target assessment matrix, including: using a deformable convolutional network to extract features from the target assessment matrix to obtain a first feature dataset; using a fused weighted bidirectional feature pyramid network and determining a second feature dataset based on the first feature dataset; and determining the target credit score of the target enterprise based on the second feature dataset.
[0093] The target credit assessment model uses the YOLOv5 (You Only Look Once version 5, a single-stage target detection algorithm) target detection model as its network framework and integrates a deep learning model that includes deformable convolutional layers and a weighted bidirectional feature pyramid network.
[0094] Specifically, in this embodiment, the target evaluation matrix is converted into an image format and input into the target credit evaluation model. A deformable convolutional network module in the target credit evaluation model extracts features from the target evaluation matrix to obtain a first feature dataset. The first feature dataset is then input into a weighted bidirectional feature pyramid network for feature extraction and fusion to obtain a second feature dataset. The second feature dataset is then input into the prediction head of the target credit evaluation model to output the target credit score for the target enterprise.
[0095] The specific operation of the deformable convolutional network module in the target credit assessment model: The deformable convolutional network module contains at least one deformable convolutional layer. Unlike standard convolutional layers that sample at fixed grid positions, deformable convolutional layers introduce a learnable offset for each sampling point. During forward propagation, the network dynamically calculates and applies these offsets based on the content of the input matrix, allowing the sampling points of the convolutional kernel to "actively" focus on regions in the matrix that are more informative for credit judgment, rather than being confined to fixed local neighborhoods. After a series of convolutions, non-linear activations, and pooling operations, a set of multi-channel first feature datasets is output.
[0096] Specifically, the deformable convolutional layer module includes:
[0097]
[0098] in, , As weight, For the input feature map, To output the feature map, This is the offset. , , which represents the number of elements in the grid;
[0099] Due to offset It is usually of fractional order, so it is calculated using bilinear interpolation:
[0100]
[0101] in, Represents any fraction position and , Enumeration feature mapping All locations in the integration space, G It is a bilinear interpolation kernel, and its formula is:
[0102]
[0103] in,
[0104] The operation of the Weighted Bidirectional Feature Pyramid Network (GBN) involves receiving multiple feature maps from different depths (i.e., different scales) of a deformable convolutional network as input. Its structure includes bidirectional (top-down and bottom-up) cross-scale connectivity pathways. At the feature fusion node, instead of simply adding or concatenating the input feature maps, a learnable weight parameter is assigned to each input feature map. During fusion, the input features are weighted and summed according to these weights, and the network is then trained stably using a fast normalization method. This structure allows for sufficient interaction and complementarity between deep semantic features and shallow detail features. After weighted bidirectional fusion, an enhanced second feature dataset is obtained.
[0105] Specifically, the extracted feature information is fused using a weighted bidirectional feature pyramid network (BiFPN), with the following formula:
[0106]
[0107] in, To output the feature map, For the first One input feature map, For the first Learnable weights corresponding to each input feature map This is the sum of the squared weights of all input feature maps, used for normalization.
[0108] Among them, Deformable Convolutional Networks (DCNs) are an improved convolutional neural network architecture in which the sampling positions of the convolutional kernels can be dynamically and adaptively shifted spatially according to the current input content. In this embodiment, it is used to adaptively extract features from the target evaluation matrix, and is particularly suitable for capturing the dependencies between features that are non-local and irregularly arranged in the matrix.
[0109] The weighted bidirectional feature pyramid network is a highly efficient multi-scale feature fusion network architecture. "Bidirectional" refers to the simultaneous inclusion of top-down (transmitting high-level semantic information) and bottom-up (transmitting low-level detailed information) feature propagation paths. "Weighted" refers to introducing learnable scalar weights to characterize the importance of each feature map when fusing features from different scales or inputs, thereby achieving better feature fusion results. In this embodiment, it is used to integrate features at different levels of abstraction in enterprise information.
[0110] The prediction head typically consists of one or more fully connected layers. It maps the fused high-level features to the final credit scoring space.
[0111] S204: The target assessment results of the target enterprise are determined by adopting a pre-set assessment strategy and based on the target credit score.
[0112] In this embodiment, the target credit score is denoted as .
[0113] Specifically, the pre-set evaluation strategy includes:
[0114] like ,and ,and ,and ,and The loan amount is A1, the loan term is B1, and the enterprise credit information is recorded. ;
[0115] like ,and ,and ,and ,and The loan amount is A2, the loan term is B2, and the enterprise credit information is recorded. ;
[0116] like ,and ,and ,and ,and The loan amount is A3, the loan term is B3, and the enterprise credit information is recorded. ;
[0117] like ,and ,and ,and ,and The loan amount is A4, the loan term is B4, and the enterprise credit information is recorded. ;
[0118] like ,and ,and ,and ,and The loan amount is A5, the loan term is B5, and the enterprise credit information is recorded. .
[0119] Optionally, v, w, x, and y, as well as A1, A2, A3, A4, A5, and B1, B2, B3, B4, B5 are also hyperparameters set manually, and are not limited in this embodiment.
[0120] Specifically, a pre-set evaluation strategy is adopted and the target evaluation result of the target enterprise is determined based on the target credit score.
[0121] For example, the target assessment result could be that the target company has a loan amount of A5 and a loan period of B5.
[0122] The enterprise credit assessment method provided in this application, in response to receiving an enterprise credit assessment request, obtains the target assessment information of the target enterprise, determines the target assessment matrix of the target enterprise based on the target assessment information, then inputs the target assessment matrix into the target credit assessment model, outputs the target credit score of the target enterprise, further adopts a preset assessment strategy and determines the target assessment result of the target enterprise based on the target credit score. The target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network, realizing full automation from data input to credit decision-making, and improving the efficiency and accuracy of enterprise credit assessment.
[0123] Figure 3 A flowchart of a corporate credit assessment method provided for another embodiment of this application is shown below. Figure 3 As shown in the illustration, as an optional implementation, based on any of the above embodiments, before employing a target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix, the method further includes the following steps:
[0124] S301: Obtain the target sample set.
[0125] The target sample set includes target evaluation sample data from multiple sample companies. This target evaluation sample data from the sample companies includes target image data and target label data.
[0126] As an optional implementation, the target sample set can be obtained using the following steps: acquiring target evaluation sample information of multiple sample companies; determining corresponding target evaluation matrices based on each target evaluation sample information; determining target image data based on each target evaluation matrix; extracting a preset number of original label data from each target evaluation sample information; converting the preset number of original label data corresponding to each target evaluation sample information into target label data using a preset numericalization strategy; determining each target image data and the corresponding target label data as target evaluation sample data of each sample company; and determining the target evaluation sample data of each sample company as the target sample set.
[0127] Among them, the target evaluation sample information refers to the sample information used to train the initial credit evaluation model.
[0128] It is understandable that each target evaluation sample represents the evaluation information of a company.
[0129] Specifically, in this embodiment, the server device obtains multiple target evaluation sample information from a preset database, and determines the corresponding target evaluation matrix based on each target evaluation sample information. This step is implemented in the same way as step S202, that is, the corresponding S202 step is executed for each target evaluation sample information. Each target evaluation matrix normalizes its element values to the pixel value range of a standard image through linear scaling, and saves them as an image format file, thereby obtaining the target image data.
[0130] The preset quantity is 5.
[0131] For example, suppose the set of target image data is E.
[0132] in, , Representing the Target image data for each enterprise.
[0133] In this embodiment, the original label data are: .
[0134] in, Start the first Individual businesses are flagged; credit approval is flagged as successful. ;in, Indicates the first Whether the corporate loan was successful. If a company successfully obtains a loan, the label is defined as 1; if the company fails to obtain a loan, the label is defined as 0.
[0135] Among them, loan amount marking ;in, Indicates the first Loan amount obtained by individual enterprises , The unit is "ten thousand yuan".
[0136] Among them, credit time stamp Mark. Among them, Indicates the first Loan time for individual enterprises , The unit is "month".
[0137] Among them, the number of successful loan applications is marked. Mark. Among them, Indicates the first Number of times a company successfully obtains credit and repays it on time , The unit is "times";
[0138] Among them, other markers Mark. Among them, Indicates the first Other markers for individual companies , .
[0139] Among them, the Generating original label data for each enterprise;
[0140]
[0141] Furthermore, by analogy, the corresponding original label data is extracted from the target evaluation sample information of each sample enterprise.
[0142] Specifically,
[0143]
[0144] Among them, the sample enterprises refer to the enterprises used for model training.
[0145] Furthermore, in this embodiment, the preset numericalization strategy refers to numerating each original label data to generate each target label data.
[0146] Specifically, the preset numericalization strategy is as follows:
[0147]
[0148] in, , , , , , , , , , , , , and , and , and , and .
[0149] In this embodiment, the set of target label data is as follows: .
[0150] Specifically, .
[0151] Specifically, in this embodiment, the collection of each target label data Convert the data to TXT format, and then use the target image data and corresponding target label data to determine the target evaluation sample data for each sample company. This target evaluation sample data for each sample company is then used to define the target sample set.
[0152] S302: Divide the target sample set into training set, validation set and test set according to a preset ratio.
[0153] Optionally, the preset ratio can be 7:2:1, or it can be set independently according to needs. This embodiment does not impose any limitations.
[0154] S303: Input the training set and validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model.
[0155] As an optional implementation, the following steps can be used to input the training set and validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model. These steps include: repeatedly performing the following operations until the initial credit assessment model completes the preset number of training rounds, and outputting the initial credit assessment model corresponding to the minimum validation loss in each training round as the trained initial credit assessment model. The following operations include: initializing the target loss weights; inputting the training set into the initial credit assessment model for training to obtain the predicted credit scores for each sample enterprise; determining the training loss using a preset loss function and based on the target loss weights, the target label data for each sample enterprise, and the corresponding predicted credit scores; updating the parameters of the initial credit assessment model and the target loss weights using a backpropagation algorithm based on the training loss; determining the validation loss based on the preset loss function and the validation set; and saving the initial credit assessment model corresponding to the round with the minimum validation loss.
[0156] The target label data contains 5 labels, therefore the target loss weight includes 5 sub-weights, which sum to 1.
[0157] Optionally, the preset number of rounds can be set independently according to needs, and this embodiment does not impose any limitations.
[0158] Specifically, in this embodiment, the target loss weights are initialized, the training set is then input into the initial credit assessment model for training, and the predicted credit scores for each sample enterprise are output.
[0159] Furthermore, in this embodiment, the preset loss function is:
[0160]
[0161] in, .
[0162] Specifically,
[0163]
[0164]
[0165]
[0166]
[0167]
[0168] Specifically, Represents the loss function; , , , and For hyperparameters; Predicted credit scores for sample companies The target label data corresponding to the sample companies; A loss function representing whether a company's credit application was successful; A loss function representing the loss on the loan amount obtained by the company; The loss function representing the duration of the loan obtained by the company; This represents the number of times a company successfully obtains credit and repays it on time; A loss function indicating whether other markers for the enterprise were successful.
[0169] In this embodiment, after the server-side device has calculated the corresponding loss for each sample enterprise in each round, it calculates the average loss for each round, thereby obtaining the training loss for each training round.
[0170] In this embodiment, the server-side device uses the backpropagation algorithm to calculate the gradient of the loss with respect to all trainable parameters in the initial credit assessment model, starting from the calculated training loss and working backward layer by layer. These parameters include not only the convolutional kernel weights and offsets of DCNv, the fusion weights of BiFPN, the fully connected weights of the prediction head, but also the target loss weights.
[0171] In this embodiment, after the server device completes one training round (i.e., traverses the entire training set), it uses the current model to perform forward propagation and prediction on the independent validation set, and calculates the loss value using the same preset loss function as the training loss (using the currently updated target loss weights), thus obtaining the validation loss. It also records the model state when the current validation loss is at its historical lowest.
[0172] In this embodiment, the training continues until a preset number of training rounds are reached. Throughout the training process, the validation loss calculated after each round is continuously monitored. After training is complete, the model parameters corresponding to the global minimum validation loss value are selected from the model states saved from all rounds, and this initial credit assessment model is output as the trained initial credit assessment model.
[0173] Specifically, by defining the target loss weights as trainable parameters and jointly optimizing them with the main model parameters, the initial credit assessment model can dynamically perceive the changes in the difficulty and importance of each credit assessment sub-task during the training process and automatically adjust the learning focus. This ensures that the initial credit assessment model can automatically find the task weight configuration that is most conducive to the overall credit assessment performance, thereby significantly improving the comprehensive prediction accuracy of the target credit assessment model.
[0174] S304: Determine the target credit assessment model based on the test set and the initial credit assessment model after training.
[0175] Specifically, in this embodiment, after training, the initial credit assessment model is tested using a test set. Preset evaluation metrics are used to quantitatively analyze the model's detection results, and these metrics are compared with corresponding preset evaluation thresholds. When a preset evaluation metric meets the corresponding preset evaluation threshold, the initial credit assessment model is determined as the target credit assessment model. If it does not meet the threshold, the initial credit assessment model needs to be retrained.
[0176] Optionally, the preset evaluation threshold may vary depending on the preset evaluation index, and this embodiment does not impose any limitation.
[0177] The preset evaluation metrics may include precision, recall, and average precision (AP), etc., but are not limited in this embodiment.
[0178] The specific formula for accuracy is as follows:
[0179]
[0180] The specific formula for recall rate is as follows:
[0181]
[0182] Where TP represents the number of samples correctly predicted as positive, FP represents the number of negative samples incorrectly predicted as positive, and FN represents the number of positive samples incorrectly predicted as negative.
[0183] Specifically, by explicitly dividing the target sample set into training, validation, and test sets, and using an independent test set for final evaluation as a threshold for determining the target credit assessment model, overfitting of the model is prevented and the accuracy of the target credit assessment model is improved.
[0184] Figure 4 A schematic diagram of the structure of a corporate credit assessment device provided in an embodiment of this application is shown below. Figure 4 As shown, the enterprise credit assessment device provided in this embodiment is located in an electronic device. The enterprise credit assessment device provided in this embodiment includes: an acquisition module 41 and a determination module 42.
[0185] Specifically, the acquisition module 41 is used to acquire the target assessment information of the target enterprise in response to receiving the enterprise credit assessment request; the determination module 42 is used to determine the target assessment matrix of the target enterprise based on the target assessment information of the target enterprise; the target assessment matrix is a multi-dimensional matrix; the determination module 42 is also used to determine the target credit score of the target enterprise by adopting the target credit assessment model and based on the target assessment matrix; the target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network; the target credit assessment model is pre-trained; the determination module 42 is also used to determine the target assessment result of the target enterprise by adopting a preset assessment strategy and based on the target credit score.
[0186] Optionally, when determining the target evaluation matrix of the target enterprise based on the target enterprise's target evaluation information, the determining module 42 is specifically used to: determine the initial evaluation matrix by using a preset quantification algorithm and based on the target enterprise's target evaluation information; and multiply the initial evaluation matrix with a preset correction factor to obtain the enterprise's target evaluation matrix.
[0187] Optionally, the determining module 42, when using the target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix, specifically performs the following: using a deformable convolutional network to extract features from the target assessment matrix to obtain a first feature dataset; using a fused weighted bidirectional feature pyramid network and determining a second feature dataset based on the first feature dataset; and determining the target credit score of the target enterprise based on the second feature dataset.
[0188] Optionally, the enterprise credit assessment device may also include a segmentation module and a training module.
[0189] Accordingly, module 41 is used to acquire a target sample set before using the target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix; the target sample set includes target assessment sample data of multiple sample enterprises; the target assessment sample data of the sample enterprises includes target image data and target label data. The partitioning module is used to divide the target sample set into a training set, a validation set, and a test set according to a preset ratio. The training module is used to input the training set and validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model. The determination module 42 is used to determine the target credit assessment model based on the test set and the trained initial credit assessment model.
[0190] Optionally, the acquisition module 41, when acquiring the target sample set, is used to acquire target evaluation sample information of multiple sample enterprises; determine the corresponding target evaluation matrix based on each target evaluation sample information; determine each target image data based on each target evaluation matrix; extract a corresponding preset number of original label data from each target evaluation sample information; convert the preset number of original label data corresponding to each target evaluation sample information into target label data using a preset numericalization strategy; determine each target image data and the corresponding target label data as the target evaluation sample data of each sample enterprise; and determine the target evaluation sample data of each sample enterprise as the target sample set.
[0191] Optionally, the training module, when inputting the training set and validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model, is used to: loop through the following operations until the initial credit assessment model completes the preset number of training rounds, and output the initial credit assessment model corresponding to the minimum validation loss in each training round as the trained initial credit assessment model. These operations include: initializing the target loss weights; inputting the training set into the initial credit assessment model for training to obtain the predicted credit scores of each sample enterprise; determining the training loss using a preset loss function and based on the target loss weights, the target label data of each sample enterprise, and the corresponding predicted credit scores; updating the parameters of the initial credit assessment model and the target loss weights using a backpropagation algorithm and based on the training loss; determining the validation loss based on the preset loss function and the validation set; and saving the initial credit assessment model corresponding to the round with the minimum validation loss.
[0192] It should be noted that the technical effects of the enterprise credit assessment device provided in this embodiment have been described in the above method embodiments, and will not be repeated here.
[0193] The enterprise credit assessment device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.
[0194] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, as shown below. Figure 5 As shown, the electronic device 50 provided in this embodiment includes: a processor 51 and a memory 52 communicatively connected to the processor 51.
[0195] The memory 52 stores computer-executable instructions; the processor 51 executes the computer-executable instructions stored in the memory 52 to implement the method provided in any of the above embodiments.
[0196] The program may include program code, which includes computer-executable instructions. Memory 52 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device.
[0197] In this embodiment, the memory 52 and the processor 51 are connected via a bus. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single straight line, but this does not mean that there is only one bus or one type of bus.
[0198] This application also provides a computer-readable storage medium, including computer-executable instructions stored in the computer-readable storage medium, which, when executed by a processor, are used to implement the enterprise credit assessment method provided in any of the above embodiments.
[0199] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the enterprise credit assessment method provided in any of the above embodiments.
[0200] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.
[0201] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.
[0202] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.
[0203] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.
[0204] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.
[0205] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.
[0206] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.
[0207] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.
[0208] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method for assessing corporate credit, characterized in that, include: In response to receiving a corporate credit assessment request, obtain the target assessment information of the target company; Based on the target evaluation information of the target enterprise, a target evaluation matrix for the target enterprise is determined; the target evaluation matrix is a multi-dimensional matrix. The target credit score of the target enterprise is determined by employing a target credit assessment model and based on the target assessment matrix; the target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network; the target credit assessment model is pre-trained. The target evaluation result of the target enterprise is determined by adopting a preset evaluation strategy and based on the target credit score.
2. The method according to claim 1, characterized in that, The process of determining the target evaluation matrix of the target enterprise based on the target enterprise's target evaluation information includes: An initial evaluation matrix is determined using a preset quantification algorithm and based on the target evaluation information of the target enterprise; The initial evaluation matrix is multiplied by a preset correction factor to obtain the enterprise's target evaluation matrix.
3. The method according to claim 1, characterized in that, The step of employing a target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix includes: A deformable convolutional network is used to extract features from the target evaluation matrix to obtain a first feature dataset. A fused weighted bidirectional feature pyramid network is used, and a second feature dataset is determined based on the first feature dataset; The target credit score of the target enterprise is determined based on the second feature dataset.
4. The method according to claim 3, characterized in that, Before employing the target credit assessment model and determining the target credit score of the target enterprise based on the target assessment matrix, the method further includes: Obtain a target sample set; the target sample set includes target evaluation sample data from multiple sample enterprises; the target evaluation sample data from the sample enterprises includes target image data and target label data; The target sample set is divided into a training set, a validation set, and a test set according to a preset ratio; The training set and the validation set are input into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model. The target credit assessment model is determined based on the test set and the trained initial credit assessment model.
5. The method according to claim 4, characterized in that, The acquisition of the target sample set includes: Obtain target evaluation sample information from multiple sample companies; Based on the target evaluation sample information, the corresponding target evaluation matrix is determined; The image data of each target is determined based on the target evaluation matrix. Extract a predetermined number of original label data from each of the target evaluation sample information; A preset numerical strategy is used to convert a preset number of original label data corresponding to each target evaluation sample information into target label data; The target image data and the corresponding target label data are determined as the target evaluation sample data for each sample enterprise; The target evaluation sample data of each of the sample enterprises is determined as the target sample set.
6. The method according to claim 4, characterized in that, The step of inputting the training set and the validation set into the initial credit assessment model for a preset number of training rounds to obtain the trained initial credit assessment model includes: Repeat the following operations until the initial credit assessment model completes the preset number of training rounds, and output the initial credit assessment model corresponding to the minimum validation loss in each training round as the trained initial credit assessment model. The following operations include: Initialize the target loss weights; The training set is input into the initial credit assessment model for training to obtain the predicted credit scores of each of the sample enterprises. The training loss is determined by using a preset loss function and based on the target loss weight, the target label data of each sample enterprise, and the corresponding predicted credit score. The backpropagation algorithm is used to update the parameters of the initial credit assessment model and the target loss weights based on the training loss. The validation loss is determined based on the preset loss function and the validation set; Save the initial credit assessment model for the corresponding round that minimizes the current verification loss.
7. A device for assessing corporate credit, characterized in that, include: The acquisition module is used to obtain the target assessment information of the target company in response to receiving a corporate credit assessment request; The determination module is used to determine the target evaluation matrix of the target enterprise based on the target evaluation information of the target enterprise; the target evaluation matrix is a multi-dimensional matrix; The determination module is further configured to determine the target credit score of the target enterprise based on the target credit assessment matrix using a target credit assessment model; the target credit assessment model is a deep learning model that integrates a weighted bidirectional feature pyramid network; the target credit assessment model is pre-trained; The determination module is also used to determine the target evaluation result of the target enterprise based on the target credit score using a preset evaluation strategy.
8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the enterprise credit assessment method as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the enterprise credit assessment method as described in any one of claims 1 to 6.
10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the enterprise credit assessment method as described in any one of claims 1 to 6.