Enterprise default prediction method and system based on meta-path denoising and capsule network modeling

Abstract

The present invention discloses a method and system for predicting corporate defaults based on meta-path denoising and capsule network modeling, belonging to the field of machine learning. The method comprises: collecting internal indicator information and external correlation information of an enterprise, and constructing a heterogeneous information network and meta-network of the enterprise based on the external correlation information of the enterprise; constructing an enterprise default assessment model that can balance mobility and noise interference, and effectively separate enterprise default risk and compliance risk-free information, thereby improving the accuracy of default identification; in the constructed enterprise default assessment model, the various internal business information collected from the enterprise, the constructed heterogeneous information network and meta-network of the enterprise are trained to obtain trained model weights; the internal business and external correlation information of the enterprise in real business is collected and input into the constructed enterprise default assessment model, and the trained model weights are used to evaluate the possibility of enterprise default in real business. The present invention can achieve more accurate enterprise default risk identification.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and in particular to a method and system for predicting enterprise defaults based on meta-path denoising and capsule network modeling. Background Art

[0002] Small and medium-sized enterprises (SMEs) play a crucial role in driving national economic growth. Therefore, analyzing corporate default risk is not only crucial for maintaining economic stability but also provides key insights for investment decisions and helps companies optimize their management and strategic planning. To this end, some corporate default prediction models incorporate internal corporate indicators and externally correlated default risk to assess each company's default risk.

[0003] Many traditional default assessment models rely on statistical analysis and traditional machine learning methods to predict corporate default risk, particularly by mining internal default risk based on indicator data. For example, a method for calculating the probability of default (PD) of small and micro-enterprise loans (CN111754341B) uses historical data from small and micro-enterprise loans, a migration matrix, and a Markov chain to calculate the default probability. The results are then forward-lookingly modified using adjustment parameters provided by business personnel, ultimately yielding accurate default rates under different scenarios. A method for establishing an enterprise default risk model based on XGBoost (CN109508864B) organizes enterprise data, slices the data using a sliding window, and uses the XGBoost algorithm to build a predictive model. The optimized model is then used to predict enterprise default risk. However, small and medium-sized enterprises (SMEs) generally face irregular or incomplete financial reporting, and the practical application of these methods faces numerous challenges. Furthermore, these methods ignore the propagation of external default risk between enterprises. In recent years, as the advantages of graph models in representing complex relationships have gradually emerged, some research has begun to explore their application to default prediction, particularly in simulating the propagation of external default risk. Methods such as HAT, ComRisk, and “A method, device, medium and electronic device for predicting enterprise defaults” (CN112990946B) attempt to model external risks through heterogeneous graph models, but these methods generally face the problem of insufficient interpretability.

[0004] To address the above challenges, researchers have proposed a meta-path-based default prediction model, which improves model interpretability by generating external correlation paths for enterprises to learn the explicitly propagated default risk. However, current default assessment models still face challenges such as limited portability and susceptibility to noise interference. Specifically, this type of method can generally be divided into two categories: the first category is models that rely on manually preset meta-paths. Due to excessive reliance on raw data, this limits the portability of the module; the second category is models that utilize all meta-paths that do not exceed a fixed length. This method introduces a large amount of irrelevant noise and reduces the accuracy of information aggregation. In addition, existing default assessment models usually compress the enterprise's default risk information and compliance risk-free information together. This approach easily confuses information, resulting in poor enterprise assessment results.

[0005] In summary, existing meta-path-based default assessment methods have limitations in terms of portability and noise reduction, making it difficult to achieve an effective balance. Furthermore, existing default assessment methods compress a company's risk and risk-free information into a single representation, leading to information confusion and poor assessment results. These two issues urgently need improvement. Summary of the Invention

[0006] To overcome the limitations of the existing technology, the present invention proposes a corporate default prediction method and system based on meta-path denoising and capsule network modeling. This method can better balance mobility and noise interference, and achieve more accurate corporate default risk identification by separating corporate default risk information from compliance risk-free information.

[0007] To achieve the above objectives, the technical solution of the present invention includes the following contents.

[0008] A method for predicting corporate defaults based on meta-path denoising and capsule network modeling, the method comprising:

[0009] Collect the target enterprise's internal operating information and external related information, and construct the enterprise's heterogeneous information network and meta-network based on the enterprise's external related information; wherein the nodes in the heterogeneous information network include: enterprise nodes, industry nodes, regional nodes and stakeholder nodes, and the meta-network is generated based on the abstraction of the heterogeneous information network;

[0010] Input the internal indicator information of the target enterprise into the internal default risk encoder to obtain the internal default risk embedding of the target enterprise;

[0011] Input the target enterprise's heterogeneous information network and meta-network into the meta-path-based explicit external default risk encoder to obtain the target enterprise's explicit external default risk embedding;

[0012] Input the target enterprise's heterogeneous information network into the implicit external default risk encoder based on the heterogeneous graph to obtain the implicit external default risk embedding of the target enterprise;

[0013] The internal default risk embedding, explicit external default risk embedding and implicit external default risk embedding are integrated with capsule network to obtain the default prediction results of the target enterprise.

[0014] Furthermore, the step of inputting the target enterprise's internal indicator information into an internal default risk encoder to obtain the target enterprise's internal default risk embedding includes:

[0015] Extracting numerical and / or discrete features of internal indicator information, standardizing and normalizing the numerical features, and converting the discrete features into embedding vectors; wherein the numerical features include registered capital, paid-in capital, registration time, and litigation duration; and the discrete features include litigation cause, court level, and litigation outcome;

[0016] The embedding vector of the discrete feature is concatenated with the processed numerical feature and mapped through a linear layer to obtain the internal default risk embedding of the target enterprise.

[0017] Furthermore, the step of inputting the heterogeneous information network and meta-network of the enterprise into the meta-path-based explicit external default risk encoder to obtain the explicit external default risk embedding of the target enterprise includes:

[0018] Based on the meta-network, it walks in the heterogeneous information network and generates path instances of all metadata types for the target enterprise that do not exceed a fixed length.

[0019] Calculate the effective value S of each path type P P ;

[0020] According to the effective value S P Sort the meta-path types and select the top K ranked meta-path types;

[0021] Among the top K ranked meta-path types, weighted aggregation is performed on the path instances of meta-path type P to obtain the semantic embedding Z under meta-path type P. P ;

[0022] Use attention mechanism to fuse semantic embedding Z P , and obtain the explicit external default risk embedding Z of the target.

[0023] Furthermore, the effective value of the meta-path Among them, T P represents the total number of all path instances of meta-path type P, represents the number of path instances in which both end nodes of meta-path type P are defaulting enterprises, Indicates the number of path instances in which both end nodes of meta-path type P are contract-abiding enterprises. represents the number of path instances in the meta-path type P where the target node is the contract-abiding enterprise and the source node is the defaulting enterprise, Indicates the number of path instances in the meta-path type P whose target node is the defaulting enterprise and whose source node is the abiding enterprise.

[0024] Furthermore, the step of inputting the target enterprise's heterogeneous information network into a heterogeneous graph-based implicit external default risk encoder to obtain the target enterprise's implicit external default risk embedding includes:

[0025] Initialize node embeddings of all nodes in a heterogeneous information network

[0026] Using the lth layer of the graph neural network, we calculate the embedding of aggregated neighbor information of each node under different external association relations π And based on node embedding and the embedding of aggregated neighbor information under all external association relations π Get the risk model embedding node embedding of the node Wherein, the graph neural network has a total of L layers;

[0027] Splicing node embedding Get the default risk pattern embedding matrix P of all nodes (L) ;

[0028] Embedding matrix P from the default risk pattern of all nodes (L) Extract the risk pattern embedding matrix of all enterprise nodes

[0029] Embedding matrix for risk patterns Perform global attention calculation to obtain the enterprise embedding matrix of each enterprise node

[0030] Based on the target enterprise's enterprise embeddedness matrix and risk pattern embedding matrix Obtain the implicit external default risk embedding of the target enterprise

[0031] Furthermore, by integrating internal default risk embedding, explicit external default risk embedding, and implicit external default risk embedding, and combining with capsule networks, the default prediction results of the target enterprise are obtained, including:

[0032] The internal default risk embedding, the explicit external default risk embedding and the implicit external default risk embedding are integrated to obtain the integrated embedding;

[0033] Based on fusion embedding, the dynamic routing mechanism of capsule network is used to generate embedding o for the target enterprise under state s. s ; Wherein, state s = {r, nr}, r represents the default risk state, and nr represents the compliance risk-free state;

[0034] Calculate the default risk score of the target enterprise r ‖ and compliance risk-free score‖o nr ‖; where ‖·‖ represents the modulo length operation; based on the default risk score ‖o r ‖ and compliance risk-free score‖o r ‖, calculate the default probability of the target enterprise

[0035] The probability of default score Compare with a set threshold to obtain the default prediction result of the target enterprise.

[0036] Furthermore, based on the fusion embedding, the dynamic routing mechanism of the capsule network is used to generate the embedding o for the target enterprise in state s. s ,include:

[0037] Extract the feature embedding u of the kth feature from the fused embedding k ;

[0038] Based on the feature embedding, obtain the feature embedding t of the kth feature extracted by the state s capsule from the fusion embedding s,k ;

[0039] Based on the feature embedding t s,k , using dynamic routing technology to fuse different feature embeddings to obtain the embedding t of the target enterprise in state s s ;

[0040] Based on the embedding s , generate embedding o under the target enterprise state s s .

[0041] A corporate default prediction system based on meta-path denoising and capsule network modeling, the system comprising:

[0042] A data collection module is used to collect the internal operating information and external related information of the target enterprise, and to construct the enterprise's heterogeneous information network and meta-network based on the enterprise's external related information; wherein the nodes in the heterogeneous information network include: enterprise nodes, industry nodes, regional nodes and stakeholder nodes, and the meta-network is generated based on the abstraction of the heterogeneous information network;

[0043] The first encoding module is used to input the internal indicator information of the target enterprise into the internal default risk encoder to obtain the internal default risk embedding of the target enterprise;

[0044] The second encoding module is used to input the heterogeneous information network and meta-network of the target enterprise into the explicit external default risk encoder based on the meta-path to obtain the explicit external default risk embedding of the target enterprise;

[0045] The third encoding module is used to input the target enterprise's heterogeneous information network into the implicit external default risk encoder based on the heterogeneous graph to obtain the implicit external default risk embedding of the target enterprise;

[0046] The default prediction module is used to integrate internal default risk embedding, explicit external default risk embedding, and implicit external default risk embedding, and combine it with the capsule network to obtain the default prediction results of the target enterprise.

[0047] An electronic device comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements any of the above-mentioned enterprise default prediction methods based on meta-path denoising and capsule network modeling.

[0048] A computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement any of the above-mentioned enterprise default prediction methods based on meta-path denoising and capsule network modeling.

[0049] Compared with the prior art, the present invention has at least the following beneficial effects.

[0050] 1) The meta-path-based explicit correlation risk encoder of the present invention focuses on generating and filtering invalid paths, thereby retaining key correlation information in default prediction, effectively balancing the model's mobility and noise interference, and at the same time hierarchically aggregating external risk information onto nodes, enhancing the interpretability of risk propagation.

[0051] 2) The implicit association risk modeling component implemented by the heterogeneous graph of the present invention is dedicated to capturing global implicit association information to compensate for the information loss caused by the limited length of the meta-path.

[0052] 3) Based on internal default risk embedding, explicit external default risk embedding and implicit external default risk embedding, the present invention uses capsule network-based dynamic routing technology to better integrate and utilize different risk information to achieve accurate default prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Flowchart of the corporate default prediction method based on meta-path denoising and capsule network modeling.

[0054] Figure 2 This is a flow chart of the model of the present invention during the training phase.

[0055] Figure 3 This is a flow chart of the model of the present invention in the prediction stage.

[0056] Figure 4 Graph showing the performance of the present invention and existing models on common public datasets. DETAILED DESCRIPTION

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0058] The enterprise default prediction method based on meta-path denoising and capsule network modeling of the present invention is as follows: Figure 1 As shown, the following steps are included.

[0059] Step 1: Collect the internal indicator information and external correlation information of the enterprise, and build the enterprise's heterogeneous information network and meta-network based on the enterprise's external correlation information.

[0060] This invention aims to improve the accuracy and interpretability of default predictions through the deep integration and precise modeling of multi-dimensional information. First, the method comprehensively collects and processes internal corporate information, including both numerical and discrete features. Specifically, numerical features include registered capital, paid-in capital, registration date, and litigation duration; discrete features encompass information such as the cause of litigation, court level, and litigation outcomes.

[0061] In addition, this method also collects a variety of external correlation information between the enterprise and other enterprises, industries, regions, and stakeholders, thereby comprehensively reflecting the external risk factors of the enterprise. Based on these different types of correlation information, the enterprise's heterogeneous information network G = {V, E} is constructed, where V is the node set of the enterprise and related entities, and E is the set of association relationships between nodes. Based on this heterogeneous information network, the enterprise's meta-network T is further abstracted. G ={A,R}; where T G It is the meta-template of G, A represents the abstract node type set, R represents the abstract edge type set, and satisfies the object mapping and the relational mapping ψ:E→R.

[0062] Step 2: Construct an enterprise default assessment model that can balance mobility and noise interference, effectively separate enterprise default risk and compliance risk-free information, and improve the accuracy of default identification.

[0063] This paper proposes a method for predicting corporate defaults based on meta-path denoising and capsule network modeling. It is essentially a binary classification model, which determines whether a company defaults. The model training flow chart is as follows: Figure 2 As shown in Figure 1, the model includes: inputting a company's internal indicators into an internal default risk encoder to obtain the company's internal default risk embedding; inputting the company's heterogeneous information network and meta-network into a meta-path-based explicit external default risk encoder to obtain the company's explicit external default risk embedding; inputting the company's heterogeneous information network into a heterogeneous graph-based implicit external default risk encoder to obtain the company's implicit external default risk embedding; fusing the company's internal default risk, explicit external default risk, and implicit external default risk, and using capsule network-based dynamic routing technology to obtain the company's default risk probability score and compliance-free risk probability score; and calculating the company's default probability based on these two probability scores. The detailed process of internal default risk, explicit external default risk, implicit external default risk, and risk fusion and assessment in this model is described in the following steps.

[0064] Step 2.1: Input the enterprise’s internal indicator information into the internal default risk encoder to obtain the enterprise’s internal default risk embedding.

[0065] For enterprise internal indicator data, it is divided into numerical and discrete types according to the feature type. For numerical features, standardization and normalization methods (such as Min-Max normalization and Z-Score normalization) are used to eliminate the dimension effect and prevent certain features from dominating the model. For discrete features, embedding technology is used to convert them into embedded vectors. Subsequently, the embedded vectors of discrete features are spliced ​​with numerical features and mapped through a linear layer to obtain the enterprise c i The internal default risk embedded in h i .

[0066] Step 2.2: Input the enterprise’s heterogeneous information network and meta-network into the meta-path-based explicit external default risk encoder to obtain the enterprise’s explicit external default risk embedding.

[0067] Given a heterogeneous information network G and a meta-network T G , a large number of meta-paths can be generated. However, most of them are irrelevant to the given prediction task. In order to balance the transferability and reduce the interference of irrelevant information, an automatic meta-path selection method is designed. Specifically, based on the consideration of computational cost and given prediction task, only meta-paths whose two end nodes are enterprises and whose length does not exceed a fixed length are considered. An effective meta-path should mainly provide default risk information or compliance non-risk information, focusing on one of the types. In addition, if the proportions of different information are too similar, the function of the meta-path may become unclear. Given a meta-path P, calculate its effectiveness in represents the total number of path instances of meta-path P, Indicates the number of instances where both end nodes of all path instances of meta-path type P are defaulting enterprises. Indicates the number of instances where both end nodes of all path instances of meta-path type P are contract-abiding enterprises. Indicates the number of instances where the target node is a contract-abiding enterprise and the source node is a contract-breaking enterprise among all path instances of meta-path type P, Represents the number of instances of meta-path type P where the target node is a defaulting enterprise and the source node is a compliant enterprise. All meta-path types are sorted according to their valid values, and the top K ranked meta-paths are selected to capture the explicit default risk embedding of enterprises.

[0068] To accurately model the external default risk characteristics of an enterprise, a hierarchical attention architecture is designed. This architecture consists of two levels: the instance level and the semantic level. The instance level aggregates information about instances along meta-paths and computes semantic embeddings for each node on different meta-paths. The semantic level further considers the semantic differences between different meta-paths and integrates their embeddings to generate external default risk embeddings along explicit paths.

[0069] First, instance-level fusion. For the target enterprise node c i Given a meta-path type P, there are many different path instances. These path instances have different importance to the target enterprise, depending on the source node c k In addition, a dynamic attention layer is designed to focus on the target node c i Perform weighted aggregation on the path instances of the meta-path type P to obtain the semantic embedding under the meta-path type P in Represents enterprise c i The current embedding, W P represents the embedding transformation matrix of meta-path type P, Represents enterprise c i The set of neighboring enterprises associated with the meta-path type P, Indicates that the company c i As the target node, enterprise c k The meta-path instance p of the source node ik Embedding, || represents the concatenation operation. Represents a meta-path instance p ik The fusion attention score of represents the learnable computational weight matrix under the meta-path type P, σ represents the activation function, and exp represents the exponential function.

[0070] Secondly, semantic level fusion. In order to integrate each enterprise node c i The semantic information of different meta-paths is integrated, and the attention mechanism is proposed to fuse the semantic representation Where T i Represents enterprise c i A collection of related meta paths. Represents the meta-path type P for enterprise c i The fusion weight of b Represents a learnable computational weight matrix.

[0071] Step 2.3: Input the enterprise’s heterogeneous information network into the implicit external default risk encoder based on the heterogeneous graph to obtain the implicit external default risk embedding of the enterprise.

[0072] Due to the limitation of meta-path length, some connections between enterprises may not be fully utilized. To address this issue, we propose a heterogeneous graph-based implicit association default risk encoder, which consists of a relationship-aware heterogeneous graph convolutional network and a global attention-based implicit association modeling module. The former aims to capture the default risk patterns of enterprises, while the latter explores implicit connections between enterprises with similar patterns from a global perspective to enhance the effectiveness of default prediction.

[0073] First, a relation-aware heterogeneous graph convolutional network is proposed. In the heterogeneous information network G, the embedding of the aggregated neighbor information of the computing enterprise under different relations is calculated. in Represents node v i The set of neighbor nodes under the relationship π, represents the kth neighbor node, Represents node v k The current embed. Represents node v k For node v i The contribution weight of π represents the matrix of learnable computational weights under the relation π, Represents node v i With node v k Embedding of edge relationships between entities. Calculating the risk model embedding of an enterprise Among them, θ represents the learnable weight parameter used to balance neighbor information and self information, σ represents the activation function, Represents node v i The current embedding, W G Represents a globally learnable parameter matrix used to embed different relationships are mapped into the same space. Represents the embedding under the relation π The fusion weight of and Both represent the embedded after mapping, All are learnable mapping parameters. Then, the results of all companies after L-layer aggregation are spliced ​​row by row to obtain the default risk pattern embedding matrix P of all companies. (L) , where the default risk model embedding matrix P (L) It means to stitch all the enterprise embeddings together in rows.

[0074] Secondly, the implicit association modeling module based on global attention is used to calculate the external implicit default risk embedding representation of each enterprise. Where FFN represents the feedforward network that enhances feature representation through multi-layer nonlinear mapping, and γ represents a learnable weight parameter for balancing local and global information. Indicates that from P (L) The risk model embedding matrix for all enterprises extracted from Represents enterprise c i implicit risk embedding. represents the enterprise embedding matrix obtained by global attention, where represents the attention learnable parameter matrix, represents the attention weight matrix, Both represent the attention learnable parameter matrix.

[0075] Step 2.4: Integrate the enterprise's internal default risk, explicit external default risk, and implicit external default risk, and obtain the enterprise's default risk probability score and compliance-free risk probability score based on the capsule network's dynamic routing technology. Based on the probability scores of default risk and compliance-free risk states, calculate the enterprise's default probability.

[0076] The present invention uses the risk aggregation layer to integrate the internal default risk of the enterprise into the h i , explicit external default risk embedded in Z i and implicit external default risk embedded Get fused embedding Where T r ={int,exp,imp} represents a set of risk types, including internal risk int, explicit external risk exp, and implicit external risk imp; Represents the fusion weight corresponding to the t-th type of risk. Represents the embedding of the t-th type of risk after mapping, including r i int =W int h i , r i exp =Wexp Z i , Where W int is the internal risk learnable embedding transformation matrix, W exp is the explicit external risk learnable embedding transformation matrix, W imp is a learnable embedding transformation matrix for implicit external risks.

[0077] Using the dynamic routing mechanism of capsule network to provide enterprise c i Generate embeddings of different states s∈S Among them, ‖·‖ obtains the modulus length of the vector, S = {r, nr} refers to the set of enterprise states, r represents the default risk state, and nr represents the compliance risk-free state; Represents enterprise c i Embedded in state s under the action of dynamic routing mechanism; Represents the fusion weight of the state s capsule for the k-th feature calculated by the dynamic routing mechanism; Represents the state s capsule from the fusion embedding r i The kth feature embedding extracted from s,k Represents the learnable embedding transformation matrix of the state s capsule to the k-th feature embedding; Represents the fusion embedding r i The kth feature embedding extracted from k represents the learnable embedding transformation matrix of the kth feature;

[0078] Computing Enterprise i Default risk probability score and the risk-free probability score of compliance

[0079] The calculation of the enterprise's probability of default score exp represents the exponential function.

[0080] Step 3: In the constructed enterprise default assessment model, various internal business information of the enterprise collected, the constructed enterprise heterogeneous information network and meta-network are trained to obtain the trained model weights.

[0081] Using the internal information and external related information of the training enterprise collected, the heterogeneous information network and meta-network of the enterprise are constructed and input into the constructed enterprise default assessment model to obtain the enterprise c i Probability of default

[0082] In each training cycle, the training loss function C train represents the set of enterprises used for training, yi Represents enterprise c i The true label of the enterprise, that is, whether the enterprise is in default, Represents enterprise c i The predicted default risk probability is quantified. By optimizing the objective function with Adam, the learning rate can be dynamically adjusted during model training. To improve the robustness of the trained model, techniques such as weight decay and regularization are applied.

[0083] Step 4: Collect internal business operations and external relationship information from real businesses, input it into the constructed enterprise default assessment model, and use the trained model weights to assess the possibility of enterprise default in real businesses.

[0084] The present invention uses the data flow in real business as Figure 3 As shown;

[0085] Collect internal enterprise information and external related information from real business operations to build the enterprise's heterogeneous information network and meta-network;

[0086] Load the trained default model weights;

[0087] Input the collected data and get the sample c of each test enterprise i Probability of default

[0088] when When it exceeds 0.5, it means that the model predicts that the enterprise will default; otherwise, it means that the model predicts that the enterprise will abide by the contract.

[0089] Figure 4 This is a comparison chart of the experimental results of the method proposed in this invention and different types of existing methods, where MDCN is the model of this invention. The dataset selected by this invention is a real-world dataset for corporate default prediction.

[0090] In summary, the present invention includes two novel modules, namely the external risk modeling module and the risk assessment module. Specifically, in order to better model external risks, the present invention designs an explicit correlation risk encoder based on meta-paths and an implicit correlation risk modeling component implemented through heterogeneous graphs. The former focuses on generating and filtering invalid paths, thereby retaining key correlation information in default prediction, effectively balancing the model's mobility and noise interference, while hierarchically aggregating external risk information to nodes, enhancing the interpretability of risk propagation. The latter is committed to capturing global implicit correlation information to make up for the information loss caused by the limited length of meta-paths. In addition, in order to better integrate and utilize different risk information to achieve accurate default prediction, the present invention designs a risk assessment method based on capsule networks.

[0091] Finally, the method of the present invention is only a preferred embodiment and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting corporate defaults based on meta-path denoising and capsule network modeling, characterized in that: The method comprises: Collect the target enterprise's internal operating information and external related information, and construct the enterprise's heterogeneous information network and meta-network based on the enterprise's external related information; wherein the nodes in the heterogeneous information network include: enterprise nodes, industry nodes, regional nodes and stakeholder nodes, and the meta-network is generated based on the abstraction of the heterogeneous information network; Input the internal indicator information of the target enterprise into the internal default risk encoder to obtain the internal default risk embedding of the target enterprise; Inputting the target enterprise's heterogeneous information network and meta-network into a meta-path-based explicit external default risk encoder to obtain the target enterprise's explicit external default risk embedding; wherein inputting the target enterprise's heterogeneous information network and meta-network into a meta-path-based explicit external default risk encoder to obtain the target enterprise's explicit external default risk embedding includes: Based on the meta-network, it walks in the heterogeneous information network and generates path instances of all metadata types for the target enterprise that do not exceed a fixed length. Calculate the effective value of each path type P The total number of all path instances of meta-path type P represents the number of path instances in which both end nodes of meta-path type P are defaulting enterprises, Indicates the number of path instances in which both end nodes of meta-path type P are contract-abiding enterprises. represents the number of path instances in the meta-path type P where the target node is the contract-abiding enterprise and the source node is the defaulting enterprise, Indicates the number of path instances in meta-path type P where the target node is the defaulting enterprise and the source node is the abiding enterprise; According to the effective value S P Sort the meta-path types and select the top K ranked meta-path types; Among the top K ranked meta-path types, weighted aggregation is performed on the path instances of meta-path type P to obtain the semantic embedding Z under meta-path type P. P ; Use attention mechanism to fuse semantic embedding Z P , get the explicit external default risk embedding Z of the target; Inputting the target enterprise's heterogeneous information network into an implicit external default risk encoder based on a heterogeneous graph to obtain the target enterprise's implicit external default risk embedding; wherein, inputting the target enterprise's heterogeneous information network into an implicit external default risk encoder based on a heterogeneous graph to obtain the target enterprise's implicit external default risk embedding includes: Initialize node embeddings of all nodes in a heterogeneous information network Using the lth layer of the graph neural network, we calculate the embedding of aggregated neighbor information of each node under different external association relations π And based on node embedding and the embedding of aggregated neighbor information under all external association relations π Get the risk model embedding node embedding of the node Wherein, the graph neural network has a total of L layers; Splicing node embedding Get the default risk pattern embedding matrix P of all nodes (L) ; Embedding matrix P from the default risk pattern of all nodes (L) Extract the risk pattern embedding matrix of all enterprise nodes Embedding matrix for risk patterns Perform global attention calculation to obtain the enterprise embedding matrix of each enterprise node Based on the target enterprise's enterprise embeddedness matrix and risk pattern embedding matrix Obtain the implicit external default risk embedding of the target enterprise The internal default risk embedding, explicit external default risk embedding and implicit external default risk embedding are integrated with capsule network to obtain the default prediction results of the target enterprise.

2. The method according to claim 1, characterized in that Inputting the internal indicator information of the target enterprise into the internal default risk encoder to obtain the internal default risk embedding of the target enterprise includes: Extracting numerical and / or discrete features of internal indicator information, standardizing and normalizing the numerical features, and converting the discrete features into embedding vectors; wherein the numerical features include registered capital, paid-in capital, registration time, and litigation duration; and the discrete features include litigation cause, court level, and litigation outcome; The embedding vector of the discrete feature is concatenated with the processed numerical feature and mapped through a linear layer to obtain the internal default risk embedding of the target enterprise.

3. The method according to claim 1, characterized in that By integrating internal default risk embedding, explicit external default risk embedding, and implicit external default risk embedding, and combining them with capsule networks, we can obtain the default prediction results for the target enterprise, including: The internal default risk embedding, the explicit external default risk embedding and the implicit external default risk embedding are integrated to obtain the integrated embedding; Based on fusion embedding, the dynamic routing mechanism of capsule network is used to generate embedding o for the target enterprise under state s. s ; Wherein, state s = {r, nr}, r represents the default risk state, and nr represents the compliance risk-free state; Calculate the default risk score of the target enterprise r ‖ and compliance risk-free score‖o nr ‖; where ‖·‖ represents the modulo length operation; Based on the default risk score r ‖ and compliance risk-free score‖o r ‖, calculate the default probability of the target enterprise ; The default probability Compare with a set threshold to obtain the default prediction result of the target enterprise.

4. The method according to claim 3, characterized in that Based on fusion embedding, the dynamic routing mechanism of capsule network is used to generate embedding o for the target enterprise under state s. s ,include: Extract the feature embedding u of the kth feature from the fused embedding k ; Based on the feature embedding, obtain the feature embedding t of the kth feature extracted by the state s capsule from the fusion embedding s,k ; Based on the feature embedding t s,k , using dynamic routing technology to fuse different feature embeddings to obtain the embedding t of the target enterprise in state s s ; Based on the embedding s , generate embedding o under the target enterprise state s s .

5. A corporate default prediction system based on meta-path denoising and capsule network modeling, characterized by: The system comprises: A data collection module is used to collect the internal operating information and external related information of the target enterprise, and to construct the enterprise's heterogeneous information network and meta-network based on the enterprise's external related information; wherein the nodes in the heterogeneous information network include: enterprise nodes, industry nodes, regional nodes and stakeholder nodes, and the meta-network is generated based on the abstraction of the heterogeneous information network; The first encoding module is used to input the internal indicator information of the target enterprise into the internal default risk encoder to obtain the internal default risk embedding of the target enterprise; The second encoding module is configured to input the target enterprise's heterogeneous information network and meta-network into a meta-path-based explicit external default risk encoder to obtain an explicit external default risk embedding of the target enterprise; wherein the inputting the target enterprise's heterogeneous information network and meta-network into the meta-path-based explicit external default risk encoder to obtain an explicit external default risk embedding of the target enterprise includes: Based on the meta-network, it walks in the heterogeneous information network and generates path instances of all metadata types for the target enterprise that do not exceed a fixed length. Calculate the effective value of each path type P The total number of all path instances of meta-path type P represents the number of path instances in which both end nodes of meta-path type P are defaulting enterprises, Indicates the number of path instances in which both end nodes of meta-path type P are contract-abiding enterprises. represents the number of path instances in the meta-path type P where the target node is the contract-abiding enterprise and the source node is the defaulting enterprise, Indicates the number of path instances in meta-path type P where the target node is the defaulting enterprise and the source node is the abiding enterprise; According to the effective value S P Sort the meta-path types and select the top K ranked meta-path types; Among the top K ranked meta-path types, weighted aggregation is performed on the path instances of meta-path type P to obtain the semantic embedding Z under meta-path type P. P ; Use attention mechanism to fuse semantic embedding Z P , get the explicit external default risk embedding Z of the target; The third encoding module is configured to input the target enterprise's heterogeneous information network into an implicit external default risk encoder based on a heterogeneous graph to obtain the target enterprise's implicit external default risk embedding; wherein, inputting the target enterprise's heterogeneous information network into the implicit external default risk encoder based on a heterogeneous graph to obtain the target enterprise's implicit external default risk embedding includes: Initialize node embeddings of all nodes in a heterogeneous information network Using the lth layer of the graph neural network, we calculate the embedding of aggregated neighbor information of each node under different external association relations π And based on node embedding and the embedding of aggregated neighbor information under all external association relations π , get the risk pattern embedding node embedding of the node Wherein, the graph neural network has a total of L layers; Splicing node embedding Get the default risk pattern embedding matrix P of all nodes (L) ; Embedding matrix P from the default risk pattern of all nodes (L) Extract the risk pattern embedding matrix of all enterprise nodes Embedding matrix for risk patterns Perform global attention calculation to obtain the enterprise embedding matrix of each enterprise node Based on the target enterprise's enterprise embeddedness matrix and risk pattern embedding matrix Obtain the implicit external default risk embedding of the target enterprise The default prediction module is used to integrate internal default risk embedding, explicit external default risk embedding, and implicit external default risk embedding, and combine it with the capsule network to obtain the default prediction results of the target enterprise.

6. An electronic device, characterized in that: The electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the enterprise default prediction method based on meta-path denoising and capsule network modeling as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the enterprise default prediction method based on meta-path denoising and capsule network modeling according to any one of claims 1 to 4.

Images