Enterprise development prediction method and device based on credit evaluation

Abstract

The invention discloses an enterprise development prediction method and device based on credit assessment, and relates to the field of enterprise credit assessment, and the method comprises the steps: collecting historical violent thunder enterprise information, respectively constructing credit knowledge maps before and after violent thunder, extracting a map structure feature vector, and constructing a state vector library; converting the violent thunder trigger event into a structural disturbance vector, and simulating the influence of the structural disturbance vector on the atlas; collecting a credit graph of a target enterprise, extracting a structural feature vector of the target enterprise, and performing similarity comparison on the structural feature vector and a pre-violent-thunder state vector library of a historical violent-thunder enterprise; if a high-similarity map exists, applying a corresponding disturbance vector to a target enterprise map vector to generate a disturbed structural feature vector; and comparing the vector with a state vector library after sudden thunder, and if the similarity is higher than a set threshold value, marking the enterprise as a potential credit risk enterprise and outputting early warning information. According to the invention, early identification and development trend prediction of potential credit risks of enterprises can be realized.

Description

Technical Field

[0001] This invention relates to the field of corporate credit assessment, and more specifically, to a method and apparatus for predicting corporate development based on credit assessment. Background Technology

[0002] In the field of corporate credit risk management and development forecasting, with the increasing complexity of the market environment and the growing uncertainty of economic activities, traditional financial indicator analysis methods are no longer sufficient to accurately identify potentially risky enterprises. This is especially true for "defaulting" enterprises—those that suddenly expose significant problems in the credit system, triggering a chain reaction of risks—as their risks cannot always be captured promptly through a single financial indicator or periodic reports.

[0003] In recent years, knowledge graphs have demonstrated strong expressive power in modeling corporate credit structures, relational networks, and dynamic evolution, and have been gradually introduced into corporate credit assessment. However, several technical challenges remain: First, it is difficult to systematically depict the credit evolution patterns of enterprises before and after defaults, lacking modeling methods for dynamic changes in credit structures; second, in the face of diverse and complex credit anomalies, there is a lack of effective simulation and extrapolation mechanisms for the impact structure of events; and third, predictions of current corporate credit status often remain at the level of static comparison, lacking the ability to simulate and predict "potential evolution paths." These problems result in a lack of specificity, timeliness, and explanatory power in predicting future corporate development, especially when facing groups of enterprises that may trigger systemic risks. Therefore, there is an urgent need for a predictive method that can integrate historical evolution data, anomaly event modeling, and structural similarity reasoning to improve the foresight and accuracy of risk identification. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and apparatus for predicting enterprise development based on credit assessment, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for predicting enterprise development based on credit assessment includes the following steps:

[0007] S1. Collect information on multiple historically known defaulting companies, construct credit knowledge graphs for each of the historical defaulting companies before and after the default, and extract the structural feature vectors of the credit knowledge graphs to form a pre-default state vector library and a post-default state vector library.

[0008] S2. Extract the default triggering event for each of the historical defaulting companies, and convert the default triggering event into a structural perturbation vector to simulate the impact of the default triggering event on the credit knowledge graph;

[0009] S3. Collect target enterprise information, construct the current credit knowledge graph of the target enterprise, and extract the structural feature vector of the current credit knowledge graph;

[0010] S4. Compare the structural feature vector of the target enterprise with the structural feature vector in the pre-default state vector library. If there is a credit knowledge graph of a similar defaulted enterprise with a similarity higher than the first preset threshold, then continue to step S5; otherwise, terminate the prediction process.

[0011] S5. Apply the structural perturbation vector corresponding to the similar defaulting companies to the structural feature vector of the target company to generate the perturbed structural feature vector of the target company.

[0012] S6. Compare the perturbed target enterprise structural feature vector with the structural feature vector of the similar defaulted enterprise in the defaulted state vector library. If the similarity is higher than the second preset threshold, mark the target enterprise as a potential credit risk enterprise and output the corresponding similar defaulted enterprise identification information and default trigger event information as a warning.

[0013] In some embodiments, step S1 specifically includes:

[0014] Based on the publicly available financial data, upstream and downstream transaction relationships, guarantee chain information, and senior management change records of each of the historical defaulting companies before and after the default, a heterogeneous credit knowledge graph is constructed with the historical defaulting companies as the central nodes and credit association relationships as the edges. The structural feature vectors of the heterogeneous credit knowledge graph are extracted using a graph embedding method based on graph neural networks, forming the pre-default state vector library and the post-default state vector library, respectively.

[0015] In some embodiments, step S2 specifically includes:

[0016] Identify key credit anomaly events for each of the historical defaulting companies within the time window prior to the default. Based on the typical evolutionary patterns of how each type of key credit anomaly event affects the structure of the credit knowledge graph, preset corresponding perturbation rule templates. Apply the perturbation rule templates to the credit knowledge graph before the default to generate a perturbed credit knowledge graph, and encode the structural feature differences of the credit knowledge graph before and after the perturbation as the structural perturbation vector.

[0017] In some embodiments, the key credit anomaly events include any one or more of the following: default by a core customer, breakage of a guarantee chain, exposure of financial fraud, change of controlling entity, or downgrade of credit rating.

[0018] In some embodiments, the disturbance rule template includes structural change rules for different types of credit anomaly events, and the change rules include any one or more of the following operations:

[0019] Reduce or remove the outgoing edge weights of key nodes to simulate default or credit breakdown scenarios.

[0020] Disconnect the bridge edges between nodes to simulate a break in the collateral chain or a disruption in the transaction path.

[0021] The centrality index value of a specific node is reduced to reflect its weakened control or loss of credit influence.

[0022] In some embodiments, step S4 employs a metric method based on cosine similarity or Mahalanobis distance to calculate the similarity score between the structural feature vector of the target enterprise and each structural feature vector in the pre-default state vector library.

[0023] In some embodiments, in step S5, the structural perturbation vector corresponding to the similar defaulting companies is subjected to a vector weighting operation with the structural feature vector of the target company to generate the perturbed structural feature vector of the target company.

[0024] In some embodiments, in step S6, the identification information of the defaulting company that matches the target company and the corresponding default triggering event information are extracted, a risk report containing the name of the defaulting company, the default time, and the type of triggering event is generated, and the risk report is displayed through a visual interface.

[0025] This invention also discloses a business development prediction device based on credit assessment, comprising the following modules:

[0026] The historical graph construction module is used to collect information on multiple historically known defaulting companies, construct credit knowledge graphs for each of the historical defaulting companies before and after the default, and extract the structural feature vectors of the credit knowledge graphs to form a pre-default state vector library and a post-default state vector library.

[0027] The trigger event modeling module is used to extract the trigger events for each of the historical defaulting companies and convert the trigger events into structural perturbation vectors to simulate the impact of the trigger events on the credit knowledge graph.

[0028] The target knowledge graph construction module is used to collect target enterprise information, construct the current credit knowledge graph of the target enterprise, and extract the structural feature vector of the current credit knowledge graph.

[0029] The initial similarity comparison module is used to compare the structural feature vector of the target enterprise with the structural feature vector in the pre-default state vector library. If there is a credit knowledge graph of a similar defaulted enterprise with a similarity higher than the first preset threshold, then step S5 is continued; otherwise, the prediction process is terminated.

[0030] The structural disturbance simulation module is used to apply the structural disturbance vector corresponding to the similar defaulting companies to the structural feature vector of the target company, and generate the perturbed structural feature vector of the target company.

[0031] The risk identification module is used to compare the structural feature vector of the disturbed target enterprise with the structural feature vector of the similar defaulted enterprise in the post-default state vector library. If the similarity is higher than the second preset threshold, the target enterprise is marked as a potential credit risk enterprise, and the corresponding similar defaulted enterprise identification information and default trigger event information are output as a warning.

[0032] In some embodiments, the step of converting a lightning-triggered event into a structural perturbation vector includes:

[0033] Identify key credit anomaly events for each of the historical defaulting companies within the time window prior to the default, and based on the typical evolutionary patterns of the impact of each type of key credit anomaly event on the credit knowledge graph structure, preset corresponding perturbation rule templates; apply the perturbation rule templates to the credit knowledge graph before the default to generate a perturbed credit knowledge graph, and encode the structural feature differences of the credit knowledge graph before and after the perturbation as the structural perturbation vector.

[0034] The advantage of this invention over existing technologies lies in its ability to fully explore the credit evolution of historically defaulted companies before and after their defaults, constructing a credit knowledge graph and structural feature vectors to form a structural vector library of pre-default and post-default states. Based on this, similarity comparison technology is used to match and identify the current credit structure of target companies, effectively discovering companies whose structural characteristics are highly similar to those of historically defaulted companies, thus achieving accurate early warning of potential credit risks. Compared to methods relying on financial indicators or a single credit scoring system, this invention captures potential "default-like characteristics" at the structural level, significantly enhancing the penetration and foresight of risk identification. Attached Figure Description

[0035] Figure 1 This is the overall flowchart of the method of the present invention;

[0036] Figure 2 This is a flowchart of the construction and disturbance generation process for the historical enterprise map of defaults in this invention;

[0037] Figure 3 This is a flowchart of the target enterprise similarity comparison process of this invention;

[0038] Figure 4 This is a flowchart of the disturbance application and risk identification process of the present invention. Detailed Implementation

[0039] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0040] like Figure 1 As shown, this invention proposes a method and apparatus for predicting enterprise development based on credit assessment. By constructing a credit knowledge graph, extracting structural feature vectors, simulating the impact of credit anomalies, and performing similarity comparisons, it can accurately identify the potential credit risks of target enterprises and provide detailed early warning information.

[0041] In a specific embodiment, to predict corporate credit risk, it is first necessary to conduct in-depth analysis of the historical data of known defaulting companies, construct a credit knowledge graph, and extract its structural features. The purpose of this step is to establish a reference model using historical data, providing a basis for subsequent risk assessment of target companies.

[0042] Specifically, such as Figure 2 As shown, publicly available information from multiple known defaulting companies was collected, including pre- and post-default financial data, upstream and downstream transaction relationships, guarantee chain information, and senior management change records. Financial data includes the companies' balance sheets, cash flow statements, and profit and loss statements; upstream and downstream transaction relationships cover transaction records between suppliers, customers, and partners; guarantee chain information records guarantee relationships between companies, such as loan guarantees or debt guarantees; and senior management change records include information on changes in management personnel and equity ownership. This data forms the basis of the credit knowledge graph.

[0043] Based on the above data, a heterogeneous credit knowledge graph is constructed with enterprises as the central nodes. The nodes in the graph include entities such as enterprises, suppliers, customers, guarantors, and executives, while edges represent credit relationships between different entities, such as transaction relationships, guarantee relationships, or management relationships. The weight of each edge is determined according to the strength of the relationship, such as the size of the transaction amount or the proportion of the guarantee amount; in some embodiments, the weights can be normalized.

[0044] To quantify the structural characteristics of the knowledge graph, this invention employs a graph neural network (GNN) method for graph embedding to extract structural feature vectors. Specifically, in some embodiments, a graph neural network model based on GraphSAGE can be used to generate a low-dimensional vector representation of each node by aggregating information about the node and its neighborhood. Finally, the credit knowledge graphs before and after the default are transformed into structural feature vectors, forming a pre-default state vector library and a post-default state vector library, respectively.

[0045] For ease of understanding, suppose that the credit knowledge graph of a defaulting company A before its default includes node A (the company itself), node B (its main supplier), node C (its core customer), and node D (its guarantor). Edges include transaction relationships between A and B (e.g., with a weight of 0.8) and guarantee relationships between A and D (e.g., with a weight of 0.6). Using a graph neural network, the extracted structural feature vectors represent the topological characteristics of this graph. After the default, if the transaction relationship between company A and node C is interrupted, the graph structure changes, generating new structural feature vectors, which are stored in the pre-default and post-default state vector databases, respectively.

[0046] When analyzing historical companies that have defaulted, it is crucial to further identify the triggering events and transform them into quantifiable structural perturbation vectors. Triggering events refer to key credit anomalies that lead to a company's credit collapse. Some examples include core customer defaults, broken guarantee chains, exposure of financial fraud, changes in controlling entities, or credit rating downgrades. These events often cause significant changes to the credit knowledge graph structure.

[0047] In practice, for each historically defaulting company, key credit anomalies within a specific time window prior to the default, such as 6 or 12 months, are analyzed. For example, if company A's core customer C defaults, the transaction relationship edge (A,C) may be deleted or its weight significantly reduced. To simulate these changes, perturbation rule templates are preset for different types of credit anomalies. These templates define typical evolution patterns of the graph structure, such as:

[0048] For defaults by core customers, the outgoing edge weight of key nodes will be reduced or removed to reflect the breakdown of credit relationships.

[0049] In the event of a broken guarantee chain, the bridging edges between nodes are disconnected to simulate an interruption of the transaction path or guarantee path.

[0050] When financial fraud is exposed or credit ratings are downgraded, the centrality index value of a specific node is reduced, reflecting a weakening of its control or credit influence in the graph.

[0051] Based on these rules, a perturbation rule template is designed and applied to the credit knowledge graph before the default, generating a perturbed graph. Subsequently, the difference in structural feature vectors between the graphs before and after the perturbation is calculated, generating a structural perturbation vector. The following formula represents the generation of the structural perturbation vector:

[0052] ΔV=V post -V pre ;

[0053] Where ΔV is the structural perturbation vector, V pre V represents the structural feature vector of the credit knowledge graph before the collapse. postThe above formula captures the direct impact of credit anomaly events on the graph structure, where the perturbed structural feature vector is represented.

[0054] Taking company A as an example, suppose its default event is the default of its core customer C. The perturbation rule template might stipulate that the weight of edge (A,C) be reduced from 0.8 to 0. After applying this rule, the structural feature vector of the graph is recalculated, resulting in V. post ΔV is then calculated using the formula described above. This vector records the changes in the graph structure caused by the default event and is used for subsequent simulations.

[0055] For the target company to be predicted, its currently available publicly available information is collected, including financial data, upstream and downstream transaction relationships, guarantee chain information, and senior management change records. Using the same construction method as for historical defaulted companies, a current credit knowledge graph of the target company is generated. Similarly, a graph neural network is used to extract the structural feature vector of the graph, denoted as V. target .

[0056] To ensure consistency in feature extraction, the construction and feature extraction process of the target company's graph are consistent with that of historical defaulted companies, including node types, edge weight definitions, and parameter settings for the graph neural network. For example, if the graph of target company B includes relationships with suppliers, customers, and guarantors, a structural feature vector of the same dimension as that of the defaulted companies is generated using the GraphSAGE model, representing its current credit network structure.

[0057] like Figure 3 As shown, the structural feature vector V of the target enterprise is... target The similarity is compared with all structural feature vectors in the pre-default state vector database to identify whether there are any historical defaulting companies with similar credit status to the target company. The similarity metric uses cosine similarity or Mahalanobis distance. The formula for calculating cosine similarity is:

[0058]

[0059] in, Let be the structural feature vector of the i-th historical defaulted company in the pre-default state vector library. The dot product represents the vector dot product, and || represents the vector magnitude. If the similarity score of a historical defaulted company is higher than the first preset threshold, such as 0.85, then the credit status of the target company is considered to be highly similar to that of the defaulted company, and the subsequent steps continue; otherwise, the prediction process is terminated, indicating that the target company currently has no significant credit risk.

[0060] For example, suppose the target company B's V target Compared with historically defaulted company A If the cosine similarity is 0.90, which is higher than the threshold of 0.85, then it is considered that the credit status of company B is similar to that of company A before its collapse, and proceed to the next step of analysis.

[0061] For companies with similar financial problems identified through the initial similarity comparison, their corresponding structural perturbation vector ΔV is obtained. i The perturbation vector is applied to the target firm's structural feature vector to generate the perturbed target firm's structural feature vector. The weighted perturbation formula is defined as follows:

[0062]

[0063] Here, α is a weighting coefficient, which is usually between 0.5 and 1, and is used to control the influence of the disturbance intensity. This represents the structural feature vector of the target company after perturbation. This formula simulates the potential changes to the credit knowledge graph of the target company should a similar default-triggered event occur.

[0064] Taking Company B as an example, if it is similar to Company A, and the triggering event for Company A's financial crisis is the default of a core customer, then the corresponding ΔV is obtained. A Calculate using the above formula. This indicates the credit status of Company B after a simulated default event.

[0065] like Figure 4 As shown, the perturbed target enterprise structural feature vector The structural feature vectors of similar defaulting companies in the post-default state vector database are compared for similarity, using the same measurement method as the first similarity comparison step described above. If the similarity is higher than a second preset threshold, such as 0.80, the target company is marked as a potential credit risk company.

[0066] For companies flagged as potentially risky, the system extracts matching information about defaulting companies, including company name, default date, and corresponding triggering events, such as event type (e.g., core customer default). A risk report containing this information is generated and displayed through a visual interface. The risk report lists the defaulting company name, default date, triggering event type, and similarity score in tabular form, while also providing a visual view of the credit knowledge graph, highlighting changes in key nodes and edges.

[0067] For example, if Company B is marked as a potentially risky company, the risk report might show: similar to Company A, which has a history of defaults, with a similarity of 0.90; the default occurred in June 2023; and the triggering event was a default by a core customer. The interface displays Company B's credit knowledge graph, marking potentially broken transaction relationship edges to help users intuitively understand the sources of risk.

[0068] To implement the above method, the present invention designs an apparatus comprising the following modules:

[0069] The historical graph construction module is responsible for collecting information on companies that have defaulted in the past, constructing credit knowledge graphs before and after the defaults, and extracting structural feature vectors through graph neural networks to form a state vector library. This module relies on databases and graph computing frameworks such as Neo4j and PyTorch Geometric.

[0070] The trigger event modeling module is used to identify default trigger events and generate structural perturbation vectors based on perturbation rule templates. The module has a built-in rule library covering perturbation rules for events such as core customer defaults and broken guarantee chains.

[0071] The target knowledge graph construction module is used to collect target enterprise information, construct the current credit knowledge graph and extract structural feature vectors, and share the graph neural network model with the historical knowledge graph construction module.

[0072] The initial similarity comparison module is used to perform vector similarity comparison, filter out companies with a history of defaults that are similar to the target company, and calculate using cosine similarity or Mahalanobis distance.

[0073] The structural disturbance simulation module is used to apply the structural disturbance vector to the target enterprise feature vector to generate the disturbed vector, which is achieved based on weighted operations.

[0074] The risk identification module performs secondary similarity comparisons to identify enterprises with potential credit risks and generates and displays risk reports. The module integrates visualization tools, such as D3.js, for displaying graphs and reports.

[0075] Taking target company B as an example, its credit knowledge graph shows strong associations with supplier S1, customer C1, and guarantor G1. Initial similarity comparison reveals a similarity of 0.90 between its graph and that of historically defaulted company A. Company A's default was triggered by a core customer default, leading to a decrease in the weight of the transaction edge. After applying perturbation rules, a perturbed vector for company B is generated and compared with the vector of company A after its default; the similarity is 0.85, higher than the threshold of 0.80. Ultimately, the system labels company B as a potential credit risk company, outputting a report indicating that the risk may be triggered by customer C1's default, and displaying the changes in the transaction edge (B, C1) in the graph.

[0076] Through the above-described embodiments, the present invention can effectively predict corporate credit risk, and by combining historical data and graph neural network technology, it provides accurate early warning information, providing strong support for corporate risk management.

[0077] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting enterprise development based on credit assessment, characterized in that, Includes the following steps: S1. Collect information on multiple historically known defaulting companies, construct credit knowledge graphs for each of the historical defaulting companies before and after the default, and extract the structural feature vectors of the credit knowledge graphs to form a pre-default state vector library and a post-default state vector library. S2. Extract the default triggering event for each of the historical defaulting companies, and convert the default triggering event into a structural perturbation vector to simulate the impact of the default triggering event on the credit knowledge graph; S3. Collect target enterprise information, construct the current credit knowledge graph of the target enterprise, and extract the structural feature vector of the current credit knowledge graph; S4. Compare the structural feature vector of the target enterprise with the structural feature vector in the pre-default state vector library. If there is a credit knowledge graph of a similar defaulted enterprise with a similarity higher than the first preset threshold, then continue to step S5; otherwise, terminate the prediction process. S5. Apply the structural perturbation vector corresponding to the similar defaulting companies to the structural feature vector of the target company to generate the perturbed structural feature vector of the target company. S6. Compare the perturbed target enterprise structural feature vector with the structural feature vector of the similar defaulted enterprise in the defaulted state vector library. If the similarity is higher than the second preset threshold, mark the target enterprise as a potential credit risk enterprise and output the corresponding similar defaulted enterprise identification information and default trigger event information as a warning.

2. The enterprise development forecasting method based on credit assessment according to claim 1, characterized in that, Step S1 specifically includes: Based on the publicly available financial data, upstream and downstream transaction relationships, guarantee chain information, and senior management change records of each of the historical defaulting companies before and after the default, a heterogeneous credit knowledge graph is constructed with the historical defaulting companies as the central nodes and credit association relationships as the edges. The structural feature vectors of the heterogeneous credit knowledge graph are extracted using a graph embedding method based on graph neural networks, forming the pre-default state vector library and the post-default state vector library, respectively.

3. The enterprise development forecasting method based on credit assessment according to claim 1, characterized in that, Step S2 specifically includes: Identify key credit anomaly events for each of the historical defaulting companies within the time window prior to the default. Based on the typical evolutionary patterns of how each type of key credit anomaly event affects the structure of the credit knowledge graph, preset corresponding perturbation rule templates. Apply the perturbation rule templates to the credit knowledge graph before the default to generate a perturbed credit knowledge graph, and encode the structural feature differences of the credit knowledge graph before and after the perturbation as the structural perturbation vector.

4. The enterprise development forecasting method based on credit assessment according to claim 3, characterized in that, The key credit anomalies include any one or more of the following: default by core customers, breakage of guarantee chains, exposure of financial fraud, change of controlling entity, or downgrade of credit rating.

5. The enterprise development forecasting method based on credit assessment according to claim 3, characterized in that, The disturbance rule template includes structural change rules for different types of credit anomaly events, and the change rules include any one or more of the following operations: Reduce or remove the outgoing edge weights of key nodes to simulate default or credit breakdown scenarios. Disconnect the bridge edges between nodes to simulate a break in the collateral chain or a disruption in the transaction path. The centrality index value of a specific node is reduced to reflect its weakened control or loss of credit influence.

6. The enterprise development forecasting method based on credit assessment according to claim 1, characterized in that, In step S4, a measurement method based on cosine similarity or Mahalanobis distance is used to calculate the similarity score between the structural feature vector of the target enterprise and each structural feature vector in the pre-implementation state vector library.

7. The enterprise development forecasting method based on credit assessment according to claim 1, characterized in that, In step S5, the structural perturbation vector corresponding to the similar defaulting companies is subjected to a vector weighting operation with the structural feature vector of the target company to generate the perturbed structural feature vector of the target company.

8. The enterprise development forecasting method based on credit assessment according to claim 1, characterized in that, In step S6, extract the identification information of the defaulting companies that match the target company and the corresponding default triggering event information, generate a risk report containing the name of the defaulting company, the time of the default, and the type of triggering event, and display the risk report through a visual interface.

9. An apparatus for implementing the enterprise development forecasting method based on credit assessment as described in claim 1, characterized in that, Includes the following modules: The historical graph construction module is used to collect information on multiple historically known defaulting companies, construct credit knowledge graphs for each of the historical defaulting companies before and after the default, and extract the structural feature vectors of the credit knowledge graphs to form a pre-default state vector library and a post-default state vector library. The trigger event modeling module is used to extract the trigger events for each of the historical defaulting companies and convert the trigger events into structural perturbation vectors to simulate the impact of the trigger events on the credit knowledge graph. The target knowledge graph construction module is used to collect target enterprise information, construct the current credit knowledge graph of the target enterprise, and extract the structural feature vector of the current credit knowledge graph. The initial similarity comparison module is used to compare the structural feature vector of the target enterprise with the structural feature vector in the pre-default state vector library. If there is a credit knowledge graph of a similar defaulted enterprise with a similarity higher than the first preset threshold, then step S5 is continued; otherwise, the prediction process is terminated. The structural disturbance simulation module is used to apply the structural disturbance vector corresponding to the similar defaulting companies to the structural feature vector of the target company, and generate the perturbed structural feature vector of the target company. The risk identification module is used to compare the structural feature vector of the disturbed target enterprise with the structural feature vector of the similar defaulted enterprise in the post-default state vector library. If the similarity is higher than the second preset threshold, the target enterprise is marked as a potential credit risk enterprise, and the corresponding similar defaulted enterprise identification information and default trigger event information are output as a warning.

10. The apparatus according to claim 9, characterized in that, The steps to convert a lightning-triggered event into a structural perturbation vector include: Identify key credit anomaly events for each of the historical defaulting companies within the time window prior to the default, and based on the typical evolutionary patterns of the impact of each type of key credit anomaly event on the credit knowledge graph structure, preset corresponding perturbation rule templates; apply the perturbation rule templates to the credit knowledge graph before the default to generate a perturbed credit knowledge graph, and encode the structural feature differences of the credit knowledge graph before and after the perturbation as the structural perturbation vector.

Images