Post-loan non-financial index credit assessment method for small and micro enterprises

Abstract

The invention relates to the field of artificial intelligence and financial science and technology, and discloses a non-financial index credit assessment method for small and micro enterprises after loan, and the method comprises the following steps: S1, obtaining the non-financial index data of a to-be-assessed small and micro enterprise, and constructing a heterogeneous relation graph based on the economic relation between the small and micro enterprises and the related parties thereof; s2, for each enterprise node in the heterogeneous relation graph, determining a risk state vector including an internal health degree and an overflow risk degree; and S3, the risk state vector is iteratively updated based on a preset time step length, and the updating comprises updating the internal health degree of the enterprise node based on the non-financial index data of the enterprise node. According to the method, the credit risk state of the enterprise is decomposed into two mutually independent dimensions, namely the internal health degree and the overflow risk degree, and the corresponding dynamic evolution equation is constructed, so that deep deconstruction and accurate attribution of enterprise risk causes are realized.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and financial technology, and specifically to a non-financial indicator credit assessment method for small and micro enterprises after lending. Background Art

[0002] As a vital component of the national economy, the healthy development of small and micro enterprises (SMEs) plays a vital role in maintaining market vitality and promoting employment. However, due to their inherent characteristics, such as small scale, inadequate financial systems, and weak risk tolerance, assessing the credit risk of SMEs has always been a major challenge in the financial sector. During the post-loan management phase, timely and accurate monitoring of changes in a company's credit status is crucial for preventing and mitigating credit risks.

[0003] Traditional post-loan credit risk assessment methods rely heavily on a company's financial statements. These methods analyze financial indicators such as a company's debt-to-asset ratio, profit margin, and cash flow to assess its debt repayment capacity. However, this approach has inherent limitations for small and micro enterprises. Firstly, they often lack standardized and complete financial statements, making data acquisition difficult and of low quality. Secondly, financial data often exhibits significant lags and is updated infrequently (e.g., quarterly or annually). This makes it difficult to reflect instantaneous changes in a company's operating conditions, resulting in delayed risk warnings and often passive detection of risks only after they have materialized.

[0004] To overcome over-reliance on financial data, the industry has gradually begun exploring the use of non-financial indicators for credit assessment. These non-financial indicators, such as a company's water, electricity, gas, and electricity usage, tax invoice information, legal records, and industrial and commercial changes, offer the advantages of rich data dimensions, frequent updates, and increased difficulty in forgery. They can reflect a company's true operating activities from multiple perspectives and in a more timely manner. However, existing assessment methods based on non-financial indicators still have significant shortcomings.

[0005] The current mainstream approach typically involves simply weighting and aggregating various non-financial indicators or inputting them into traditional machine learning models (such as logistic regression and support vector machines) to produce a comprehensive risk score. While this approach improves the timeliness of assessments to a certain extent, it still essentially analyzes companies as isolated entities. This ignores a crucial fact in the modern business environment: every company is a node in the business ecosystem it inhabits. Its credit risk depends not only on its own operations but also on the influence of its upstream and downstream partners, guarantors, investors, and other related entities. Existing technologies generally lack the ability to model and quantify this complex interdependent risk transmission mechanism, making it unable to effectively identify and warn of "contagious" risks caused by external related parties, resulting in systematic biases and blind spots in assessment results.

[0006] Furthermore, most existing assessment models, whether based on financial or non-financial data, often output a single, opaque risk score. This "black box" approach makes it difficult for post-loan managers to understand the specific drivers behind the risk score, making it difficult to clearly distinguish whether increased risk stems from a company's inherent operational deterioration or external environmental shocks. This lack of interpretability significantly limits the guiding value of assessment results in actual risk management, making it difficult to support accurate and efficient post-loan management decisions.

[0007] Therefore, there is an urgent need for a new post-loan credit assessment method for small and micro enterprises. This method should be able to effectively integrate high-frequency non-financial indicators and clearly model the risk transmission relationship between enterprises, thereby achieving a dynamic, explainable and traceable comprehensive assessment of corporate credit risk. Summary of the Invention

[0008] In response to the shortcomings of the existing technology, the present invention provides a non-financial indicator credit assessment method for small and micro enterprises after loan, which solves the technical problem that the existing technology cannot effectively model the dynamic evolution process and correlation transmission mechanism of risks when using non-financial indicators to conduct post-loan credit assessment of small and micro enterprises, resulting in opaque assessment results and lack of traceability.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: a non-financial indicator credit assessment method for small and micro enterprises after lending, comprising the following steps: S1. Obtain non-financial indicator data of the small and micro enterprises to be evaluated, and construct a heterogeneous relationship diagram based on the economic relationships between the small and micro enterprises and their related parties; S2. For each enterprise node in the heterogeneous relationship graph, determine a risk state vector including an intrinsic health degree and a spillover risk degree; S3. Iteratively updating the risk state vector based on a preset time step, wherein the updating includes: Based on the non-financial indicator data of the enterprise node itself, update its intrinsic health; Based on the risk status of the neighboring nodes of the enterprise node, update the spillover risk degree thereof; The credit risk assessment result of the enterprise node is determined according to the updated risk state vector.

[0010] Preferably, the step of updating the intrinsic health is specifically to calculate the intrinsic health at the next moment according to the following formula: ; in, is the intrinsic health of the enterprise node at the next moment, is the intrinsic health of the enterprise node at the current moment, is the preset intrinsic health decay rate of the enterprise node, is the preset input sensitivity vector of the enterprise node, It is the non-financial indicator data of the enterprise node at the current moment.

[0011] Preferably, the step of updating the spillover risk includes: Based on the risk status of the neighboring nodes of the enterprise node, the risk flux propagated from each neighboring node to the enterprise node is calculated, and the spillover risk of the enterprise node is updated in combination with a preset spillover risk persistence rate.

[0012] Preferably, the step of calculating the risk flux is specifically performed according to the following formula: ; in, is the risk flux, is the neighbor node set of the enterprise node, is any neighbor node in the neighbor node set, For neighbor nodes The dynamic attention weight pointing to the enterprise node, Neighbor nodes The risk state vector, is the preset risk output function.

[0013] Preferably, the dynamic attention weight is calculated by the following formula: ; in, is the dynamic attention weight, For neighbor nodes The attention score pointing to the enterprise node, is the neighbor node set of the enterprise node, is any neighbor node in the neighbor node set, is an exponential function.

[0014] Preferably, it also includes: Before iteratively updating the risk state vector, at least one model parameter involved in the method is estimated through a hierarchical Bayesian model, and the model parameter includes the intrinsic health decay rate, the spillover risk persistence rate or the parameter for calculating the attention score.

[0015] Preferably, the step of determining the credit risk assessment result includes: Based on the changes in the intrinsic health and spillover risk in the risk state vector, the credit risk causes of the enterprise node are attributed.

[0016] Preferably, when the cause of the credit risk is attributed to a change in spillover risk, the method further comprises: According to the dynamic attention weight, the main source node of the credit risk is determined.

[0017] Preferably, the step of determining the credit risk assessment result is specifically: The updated intrinsic health and spillover risk are combined into a quantitative risk score through a preset risk synthesis function.

[0018] The non-financial indicator credit assessment system for small and micro enterprises after lending includes: A data modeling module is used to obtain non-financial indicator data of the small and micro enterprises to be evaluated, and to construct a heterogeneous relationship diagram based on the economic relationships between the small and micro enterprises and their related parties; A state determination module, configured to determine a risk state vector including an intrinsic health degree and a spillover risk degree for each enterprise node in the heterogeneous relationship graph; A state evolution module is configured to iteratively update the risk state vector based on a preset time step, wherein the state evolution module further comprises: An intrinsic health updating unit, configured to update the intrinsic health of the enterprise node based on its own non-financial indicator data; A spillover risk updating unit, configured to update the spillover risk of the enterprise node based on the risk status of the neighboring nodes of the enterprise node; The risk assessment module is used to determine the credit risk assessment result of the enterprise node according to the updated risk state vector.

[0019] The present invention provides a non-financial credit assessment method for small and micro enterprises after lending. It has the following beneficial effects: 1. The present invention decomposes the credit risk status of an enterprise into two independent dimensions: intrinsic health and spillover risk, and constructs a corresponding dynamic evolution equation, thereby achieving a deep deconstruction and precise attribution of the causes of enterprise risk. When the credit risk of an enterprise changes, this method can clearly distinguish whether the change is an endogenous risk caused by the enterprise's own business activities or an input risk transmitted from other entities in its associated network. This highly interpretable assessment result overcomes the drawbacks of the "black box" of traditional assessment models, allowing post-loan managers to not only know the level of risk, but also gain insight into the root source of risk, so that more targeted risk control measures can be taken.

[0020] 2. This invention introduces a risk propagation kernel based on heterogeneous relationship graphs and a dynamic attention mechanism to accurately simulate the risk transmission process in complex business networks. This method not only considers the interconnected relationships between enterprises but, more importantly, dynamically calculates the intensity and direction of risk transmission based on different relationship types (such as guarantees and supply chains) and the real-time status of nodes. When externally input risks are identified, this method automatically and accurately locates the key source enterprises in the risk transmission path by backtracking attention weights. This traceability of risk transmission paths provides unprecedented and precise guidance for chain-based investigation and proactive intervention of post-loan risks, enabling a shift from passive response to proactive management.

[0021] 3. The present invention constructs a unified dynamic state space model to organically integrate the real-time changes in non-financial indicators with the topological structure of enterprises in the associated network, thereby achieving continuous and dynamic tracking of the credit risks of small and micro enterprises. Compared with traditional methods that rely on low-frequency financial data or static indicators, the present invention can capture subtle changes in operating conditions reflected by high-frequency non-financial data and amplify potential risk signals through network transmission mechanisms. This dynamic assessment paradigm makes early warning of risks possible, allowing them to be discovered in the early stages of risk events before substantial losses have occurred, thereby gaining a valuable time window for risk mitigation and disposal, and significantly improving the effectiveness and foresight of post-loan management. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] Example: Please see the attached Figure 1 -Attached Figure 2 The embodiment of the present invention provides a non-financial indicator credit assessment method for small and micro enterprises after lending, including the following steps: S1. Obtain non-financial indicator data of the small and micro enterprises to be evaluated, and construct a heterogeneous relationship diagram based on the economic relationships between the small and micro enterprises and their related parties; In this embodiment, step S1, namely, "obtaining non-financial indicator data for the small and micro enterprises to be assessed and constructing a heterogeneous relationship graph based on the economic relationships between the small and micro enterprises and their related parties," is intended to establish a comprehensive, structured data foundation that reflects the real business environment for subsequent dynamic risk assessment. This step is specifically implemented as follows.

[0025] First, this method securely connects with multiple external data source systems through pre-defined data interfaces to collect and integrate the various non-financial indicator data required for evaluation. In a preferred embodiment, the data interface can be a RESTful API, a direct database connection, or a standardized file transfer protocol. Data source systems may include, but are not limited to, commercial information query platforms, public judicial information platforms, business systems of partner financial institutions, tax management systems, and data systems of public utility management departments.

[0026] The data collected covers multiple dimensions that can reflect the business status and potential risks of the enterprise. Specifically, non-financial indicator data may include: Basic company information: such as the company's registered capital, shareholder structure, historical change records, branch information, and whether it is included in the list of abnormal operations.

[0027] Judicial litigation information: such as court hearing announcements, judgment documents, records of dishonest debtors, consumption restriction orders, equity freezes and other compulsory enforcement information where the company is the defendant.

[0028] Business behavior data: such as transaction details of the company's public accounts at financial institutions, the invoice amounts and frequency reported to the tax authorities, the tax credit rating assessment results, and the consumption and payment records of water, electricity, gas and other energy at the company's business premises.

[0029] Related party information: It also includes the collection of personal public information of core related parties such as the legal representative, actual controller, and major natural person shareholders of the enterprise, such as their personal litigation information or records of default.

[0030] After completing the collection and cleaning of multi-source heterogeneous data, one of the core points of this invention is to transform discrete data points into a structured model that can holistically describe the business ecosystem. To this end, this method abstracts the entire business network, including the small and micro enterprises to be evaluated, into a heterogeneous dynamic graph. .

[0031] The construction process of a heterogeneous dynamic graph is as follows: About Node Sets Determination of: Each node in the graph Each node corresponds to an entity with independent capacity to act in commercial activities. To comprehensively characterize the potential sources and transmission paths of risk, the node set encompasses not only the small and micro enterprise being assessed but also, preferably, its core affiliates that significantly influence its operations. These affiliates may include, but are not limited to, guarantee companies providing loan guarantees to the enterprise, upstream and downstream companies that occupy a core position in the supply chain, other companies with joint investments or cross-shareholdings with the enterprise, and the legal representative or actual controller of the enterprise. Including these entities in the node set aims to construct a comprehensive, closed-loop risk analysis network.

[0032] On Heterogeneous Edge Sets Determination of: Edges in the graph To represent a node and A key technical feature of the present invention is that the edge set is heterogeneous and dynamic.

[0033] Dynamic, by subscript This means that the connections in the graph evolve over time. For example, the signing of a new guarantee contract generates a new "guarantee" edge, while the completion of an old loan may result in the removal or status change of the corresponding edge. This method dynamically updates the graph topology based on the latest data input, ensuring that it is instantly reflective of the real-world business environment.

[0034] Heterogeneity is another key difference between this method and traditional homogeneous graph models. are assigned a specific relationship type The necessity of this is that different types of economic relationships have inherently different risk transmission mechanisms, intensities, and directions. For example, the risks transmitted by a "guarantee" relationship are usually direct and severe, while the risks transmitted by a "general supply chain" relationship may be relatively mild. In a preferred embodiment, the relationship type The collection of relationships can include: "guarantee relationships," "upstream and downstream supply chain relationships," "joint investment relationships," "capital flow relationships," "executive management relationships," etc. By typifying relationships, it provides a foundation for the use of targeted risk propagation calculation models (such as attention parameters specific to relationship types) in subsequent steps.

[0035] Finally, in order to drive the subsequent state evolution calculation, this step also requires each enterprise node Extract and organize the dynamic operating indicators generated in each time step (for example, a natural day). These indicators are organized into a External input vector This vector is designed to capture the latest "pulse" of the enterprise's own operating status and can be composed of: ; The vector It will serve as the direct input for updating the intrinsic health of the node in the subsequent step S3 and is the signal source for the model to perceive changes in the enterprise's own operations.

[0036] At this point, step S1 is completed. The output is a heterogeneous dynamic graph and the external input vector of each node , together constitute an information-rich and clearly structured data object, which provides sufficient and necessary input for accurate and interpretable dynamic risk evolution analysis in subsequent steps.

[0037] S2. For each enterprise node in the heterogeneous relationship graph, determine a risk state vector that includes intrinsic health and spillover risk; In this embodiment, step S2, namely, "Determining a risk state vector encompassing intrinsic health and spillover risk for each enterprise node in the heterogeneous relationship graph," is designed to establish a mathematical model capable of deeply deconstructing and quantifying enterprise credit risk. This step bridges static graph analysis with dynamic evolutionary analysis. By introducing state variables with clear economic implications, it provides a theoretical foundation for subsequent precise risk attribution and path tracing.

[0038] In the present invention, the traditional, single-dimensional risk score is replaced by a more informative multi-dimensional state vector. Specifically, for each enterprise node in the heterogeneous relationship graph constructed in step S1, , this method is that at any time Associate a two-dimensional risk state vector The formal expression of this vector is as follows: ; The vector decomposition used here isn't an arbitrary mathematical division; it's based on a deep understanding of the sources of risk for small and micro businesses. Splitting risk into "intrinsic" and "external" dimensions is essential because a business's survival and development are influenced by both its own operational capabilities and the external macroeconomic environment. These two influences, however, have distinct natures and response strategies. The state vector design of this invention aims to quantify and isolate these two influences, thereby enabling clear identification of the causes of risk.

[0039] The first component of the vector, , defined as Intrinsic Health.

[0040] This scalar value is used to quantify the number of enterprise nodes. Its own inherent operational stability and credit foundation. It is designed to mainly respond to the enterprise's own operating behavior data, that is, the external input vector generated in step S1 A higher The value of , intuitively corresponds to the good operation of the enterprise itself, such as sufficient cash flow, stable orders, no new negative judicial information, etc. On the contrary, a continuously declining A value of 0 clearly indicates the endogenous operating difficulties of the enterprise. By establishing this component, this method can separate the risks caused by changes in the fundamentals of the enterprise from complex risk signals.

[0041] The second component of the vector, , which is defined as the spillover risk (PropagatedRisk).

[0042] This scalar value is used to quantify the number of enterprise nodes. The input risk is passively received and accumulated through heterogeneous relationship edges from the associated network in which it is located. It is designed not to directly respond to the business data of the enterprise itself. , but responds to the status of its neighboring nodes. The purpose of this design is to capture the risk of "a fire in the city gate will affect all fish in the pond". For example, the internal health of an enterprise itself It may remain at a high level, but if the operating conditions of its core guarantors or important customers (i.e., neighbor nodes in the graph) deteriorate sharply, this risk will be transmitted through the network, reflecting the spillover risk of the enterprise. The establishment of this component enables this method to identify and quantify those external environmental risks that are not caused by the enterprise's own fault but are equally fatal.

[0043] At the beginning of the entire evaluation process , it is necessary to determine the initial state vector for all nodes in the system In a preferred embodiment, the initialization process can be performed in the following manner: Initial internal health : It can be obtained through a pre-trained mapping function based on the company's historical operating data for a period of time before the assessment begins, or directly converted using the company's existing static credit rating score.

[0044] Initial spillover risk : In the absence of prior information, it can be uniformly set to 0, which means that at the moment the assessment begins, the system assumes that all companies have not accumulated any external risks.

[0045] Through the above method, this step provides a clear, quantitative and business-meaningful calculation starting point for the subsequent dynamic evolution calculation (step S3). The introduction of this invention enables the present invention to go beyond the traditional “black box” model and realize “what is the risk” (through and value judgment), “Why is there a risk” (through monitoring and attribution of changes in This step makes every model output highly interpretable, greatly enhancing the business value and decision-making support capabilities of the evaluation results.

[0046] S3. Iteratively update the risk state vector based on the preset time step. The update includes: Update the internal health of the enterprise node based on its own non-financial indicator data; Based on the risk status of the neighboring nodes of the enterprise node, update its spillover risk; According to the updated risk status vector, the credit risk assessment result of the enterprise node is determined.

[0047] In this embodiment, step S3, namely, "iteratively updating the risk state vector based on a preset time step and determining the credit risk assessment result for the enterprise node based on the updated risk state vector," constitutes the core dynamic calculation engine of the present invention's method. It receives the real-time data input provided by step S1 and the initial state set by step S2. Through an iterative calculation process, it accurately simulates and quantifies the evolution of risk over time and its transmission within the enterprise network, ultimately outputting assessment results that are highly interpretable and valuable for decision support.

[0048] The steps are in a discrete time series =0,1,2,.... From the current moment Until the next moment In each iteration, the system will synchronously or asynchronously perform the following series of coupled update calculations on all enterprise nodes in the heterogeneous relationship graph.

[0049] Update on Inner Health: This updating process aims to characterize the changes in the endogenous risks of enterprises due to their own business activities. Specifically, for any enterprise node , its intrinsic health at the next moment The calculation of the And the external input vector newly obtained by step S1 at the current moment This design reflects the independence of inherent risks.

[0050] In a preferred embodiment, the update process is described by the following linear state space equation: In this equation, the physical meaning of each symbol is clear: Item 1 Represents "state inertia". Among them, Is a scalar value between [0,1], known as the intrinsic health decay rate. It reflects the stability or memory of the enterprise's health status. A value close to 1 This means that the health of the company is highly sustainable and is not likely to change drastically due to short-term disturbances.

[0051] Item 2 Represents the "external shock". Among them, is a vector that is The input sensitivity vector of the same dimension, each component of which represents the sensitivity of intrinsic health to changes in different non-financial indicators (such as transaction flow, water and electricity usage). This vector inner product operation converts multi-dimensional business behavior data into a single scalar impact on intrinsic health. Parameters and It is personalized for each enterprise and can be obtained through offline training (such as the hierarchical Bayesian model mentioned above) to reflect the differences in different enterprises' risk resistance and dependence on operating indicators.

[0052] Update on spillover risk: This update process is one of the core innovations of this invention, designed to accurately simulate and quantify the spread and accumulation of risk within a network of enterprise connections. The update of spillover risk is designed to be driven entirely by the state of its neighboring nodes, thus clearly separating it from endogenous risk.

[0053] The update process is also described by a state space equation: ; Among them, the first It also represents state inertia, It is called the spillover risk persistence rate, which reflects the speed at which the accumulated external risks dissipate. The key innovation is called total risk flux, which represents the current moment , flows from all neighbor nodes to the enterprise node the total amount of risk.

[0054] Total risk flux The calculation of is achieved through a weighted aggregation process based on the attention mechanism: ; In this formula, is a node In the set of neighbor nodes in the heterogeneous relationship graph, the sum operation traverses all possible pairs Directly affected parties. Is a risk output function for the neighbor nodes The complete state vector of In a preferred embodiment, it is possible to set The business logic is that the root cause of an enterprise's risk spreading to the outside world lies in the deterioration of its own operating conditions (i.e., its internal health).

[0055] The core lies in dynamic attention weight The weight is not statically preset, but dynamically generated according to the network status, reflecting the current ,Neighbor For Node The importance or risk transmission strength of is calculated by normalizing the attention scores of all neighbors through a Softmax function: ; Among them, the attention score The calculation of fully reflects the processing ability of the present invention on relational heterogeneity. It is a node to be evaluated Status, neighbor nodes The states and the types of edges between them is a function of the input. In a preferred embodiment, its specific form is a small feedforward neural network: ; In this formula, and Specific to the relationship type The learnable weight matrix and attention vector of . This design is necessary because it allows the model to learn completely different risk propagation patterns for different types of relationships (such as "guarantee" and "supply chain"). For example, the model may learn that risk transmission is more direct and intense (corresponding to a larger weight) in the "guarantee" relationship, while it is relatively mild in the "supply chain" relationship. Symbol Represents vector concatenation, and LeakyReLULeakyReLU is a nonlinear activation function.

[0056] Regarding the determination of credit risk assessment results: After completing an iteration, get all nodes at the next moment The latest state vector Finally, this method converts it into evaluation conclusions that are directly valuable to business personnel.

[0057] Comprehensive risk assessment: First, through a preset risk comprehensive function , the inner health and spillover risk The two components are combined into a single, quantitative overall risk score .

[0058] Risk cause attribution: The explainability advantage of the present invention is reflected here. The system continuously monitors and The changes in their respective time series can clearly attribute risks. If a company’s risk score rises, and the system detects that If the monitoring is caused by a significant decline, an "endogenous business risk warning" can be output; if the monitoring is If the risk is significantly increased, an "external correlation risk warning" can be output.

[0059] Risk path tracing: When the "external related risk warning" is triggered, the present invention can provide further decision support. The system can automatically trace back and analyze the causes The contribution of each neighboring node is increased, that is, the dynamic attention weight of the previous moment is checked .in, One or more neighbor nodes with the largest value , that is, it was identified and reported as the main source of infection of this risk event, thereby providing precise target guidance for the due diligence and risk disposal of the post-loan management team.

[0060] Reference Attachment Figure 2 Another embodiment of the present invention discloses a non-financial indicator credit assessment system for small and micro enterprises after lending, including: The data modeling module is used to obtain non-financial indicator data of small and micro enterprises to be evaluated and to construct a heterogeneous relationship diagram based on the economic relationships between small and micro enterprises and their related parties; A state determination module is used to determine a risk state vector containing intrinsic health and spillover risk for each enterprise node in the heterogeneous relationship graph; A state evolution module is used to iteratively update the risk state vector based on a preset time step, wherein the state evolution module further includes: The intrinsic health update unit is used to update the intrinsic health of the enterprise node based on its own non-financial indicator data; A spillover risk updating unit, configured to update the spillover risk of an enterprise node based on the risk status of its neighboring nodes; The risk assessment module is used to determine the credit risk assessment result of the enterprise node based on the updated risk state vector.

[0061] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A non-financial credit assessment method for small and micro enterprises after lending, characterized by: The following steps are involved: S1. Obtain non-financial indicator data of the small and micro enterprises to be evaluated, and construct a heterogeneous relationship diagram based on the economic relationships between the small and micro enterprises and their related parties; S2. For each enterprise node in the heterogeneous relationship graph, determine a risk state vector including an intrinsic health degree and a spillover risk degree; S3. Iteratively updating the risk state vector based on a preset time step, wherein the updating includes: Based on the non-financial indicator data of the enterprise node itself, update its intrinsic health; Based on the risk status of the neighboring nodes of the enterprise node, update the spillover risk degree thereof; The credit risk assessment result of the enterprise node is determined according to the updated risk state vector.

2. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 1 is characterized in that: The step of updating the intrinsic health is specifically to calculate the intrinsic health at the next moment according to the following formula: ； in, is the intrinsic health of the enterprise node at the next moment, is the intrinsic health of the enterprise node at the current moment, is the preset intrinsic health decay rate of the enterprise node, is the preset input sensitivity vector of the enterprise node, It is the non-financial indicator data of the enterprise node at the current moment.

3. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 1 is characterized in that: The steps of updating the spillover risk include: Based on the risk status of the neighboring nodes of the enterprise node, the risk flux propagated from each neighboring node to the enterprise node is calculated, and the spillover risk of the enterprise node is updated in combination with a preset spillover risk persistence rate.

4. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 3 is characterized in that: The step of calculating the risk flux is specifically to calculate according to the following formula: ； in, is the risk flux, is the neighbor node set of the enterprise node, is any neighbor node in the neighbor node set, For neighbor nodes The dynamic attention weight pointing to the enterprise node, Neighbor nodes The risk state vector, is the preset risk output function.

5. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 4 is characterized in that: The dynamic attention weight is calculated by the following formula: ； in, is the dynamic attention weight, For neighbor nodes The attention score pointing to the enterprise node, is the neighbor node set of the enterprise node, is any neighbor node in the neighbor node set, is an exponential function.

6. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 1 is characterized in that: Also includes: Before iteratively updating the risk state vector, at least one model parameter involved in the method is estimated through a hierarchical Bayesian model, and the model parameter includes the intrinsic health decay rate, the spillover risk persistence rate or the parameter for calculating the attention score.

7. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 1 is characterized in that: The step of determining the credit risk assessment result comprises: Based on the changes in the intrinsic health and spillover risk in the risk state vector, the credit risk causes of the enterprise node are attributed.

8. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 7 is characterized in that: When the credit risk cause is attributed to a change in spillover risk, the method further includes: According to the dynamic attention weight, the main source node of the credit risk is determined.

9. The non-financial indicator credit assessment method for small and micro enterprises after loan according to claim 1 is characterized in that: The steps of determining the credit risk assessment results are specifically as follows: The updated intrinsic health and spillover risk are combined into a quantitative risk score through a preset risk synthesis function.

10. A non-financial indicator credit assessment system for small and micro enterprises after lending, according to any one of claims 1 to 9, characterized in that: include: A data modeling module is used to obtain non-financial indicator data of the small and micro enterprises to be evaluated, and to construct a heterogeneous relationship diagram based on the economic relationships between the small and micro enterprises and their related parties; A state determination module, configured to determine a risk state vector including an intrinsic health degree and a spillover risk degree for each enterprise node in the heterogeneous relationship graph; A state evolution module is configured to iteratively update the risk state vector based on a preset time step, wherein the state evolution module further comprises: An intrinsic health updating unit, configured to update the intrinsic health of the enterprise node based on its own non-financial indicator data; A spillover risk updating unit, configured to update the spillover risk of the enterprise node based on the risk status of the neighboring nodes of the enterprise node; The risk assessment module is used to determine the credit risk assessment result of the enterprise node according to the updated risk state vector.

Images