Evaluation method for dynamically calculating quality degree based on commercial network

Abstract

The invention discloses an evaluation method for dynamically calculating the quality degree based on a commercial network. The evaluation method comprises the following steps: acquiring multi-source heterogeneous commercial network data; wherein the commercial network data comprises enterprise internal data, public data and authorization data; performing data processing and integration on the multi-source heterogeneous commercial network data to obtain a commercial network knowledge graph; analyzing the commercial network knowledge graph by using a graph neural network to obtain an association mode between enterprises; performing risk quantification on the commercial network knowledge graph by using a risk quantification model to obtain a risk assessment result corresponding to the enterprise; and performing enterprise high-quality scoring by using the commercial network dynamic high-quality evaluation model in combination with the risk assessment result, the association mode and the commercial network knowledge graph. In this way, dependence on subject credit and static financial data can be reduced.

Description

Technical Field

[0001] This application relates to the technical field of business networks, and particularly to an evaluation method for dynamically calculating the quality degree based on a business network. Background Art

[0002] In modern manufacturing and the global economic system, small and medium-sized enterprises (SMEs) play a crucial role. Especially in the industrial chain and supply chain, they are an important source of innovation vitality and economic resilience. However, SMEs, especially those under the business-to-business (B2B) model, have long faced challenges in financing difficulties and high financing costs. When traditional financial institutions conduct credit risk assessments, they overly rely on static data such as the credit of the enterprise entity and financial statements, resulting in the following drawbacks:

[0003] Information asymmetry and data gap: SMEs, especially manufacturing enterprises in the upstream of the industrial chain, have relatively low information transparency, and their financial data may be incomplete or difficult to reflect their true operating conditions. Moreover, overseas financial institutions lack effective local data for assessment when facing Chinese manufacturing enterprises going global.

[0004] Limitations of static assessment: Traditional credit assessment models are often static and unable to capture market changes, industry trends, and the dynamic evolution of the enterprise business ecosystem in a timely manner. This leads to a lag in risk assessment results and is difficult to effectively support rapidly changing credit decisions.

[0005] Lack of network perspective: Traditional assessment methods usually analyze the enterprise entity in isolation, ignoring the status, value, and dynamic performance of the enterprise in the supply chain and business ecosystem network. This makes it impossible for risk assessment to comprehensively consider the impact of upstream and downstream related enterprises, and may underestimate systemic risks and the true quality degree of the enterprise.

[0006] Dependence on entity credit: Overemphasis on entity credit and financial data makes it difficult for those SMEs with good operations but shortfalls in entity credit or financial data to obtain necessary financial support, hindering their development and growth. Summary of the Invention

[0007] The evaluation method for dynamically calculating the quality degree based on a business network provided by this application can reduce the dependence on entity credit and static financial data.

[0008] In a first aspect, the present application provides an evaluation method for dynamically calculating the quality of a business network. The evaluation method includes: obtaining multi-source heterogeneous business network data; where the business network data includes: enterprise internal data, public data, and authorized data; performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; using a graph neural network to analyze the business network knowledge graph to obtain the association patterns between enterprises; using a risk quantification model to quantify the risks of the business network knowledge graph to obtain the risk assessment results corresponding to the enterprises; and using a business network dynamic quality evaluation model to combine the risk assessment results, association patterns, and business network knowledge graph to perform quality scoring on the enterprises.

[0009] Among them, performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph includes: cleaning the multi-source heterogeneous business network data, and extracting information related to enterprises from the cleaned data as entities, and events and relationships between enterprises as edges to construct a business network knowledge graph.

[0010] Among them, the business network dynamic quality evaluation model is trained using federated learning technology.

[0011] Among them, the business network dynamic quality evaluation model is constructed using an interpretable machine learning model and an ensemble learning method.

[0012] Among them, the business network knowledge graph is a dynamic knowledge graph, and the graph neural network is a heterogeneous graph neural network. Using the graph neural network to analyze the business network knowledge graph to obtain the association patterns between enterprises includes: using the heterogeneous graph neural network to learn independent embedding representations for each type of node and edge in the business network knowledge graph; and when performing information aggregation, designing different message passing functions according to the types of nodes and edges; and combining the attention mechanism to dynamically learn the importance of different neighbor nodes and different types of edges in the information aggregation process, so as to obtain the association patterns between enterprises.

[0013] Among them, using the risk quantification model to quantify the risks of the business network knowledge graph to obtain the risk assessment results corresponding to the enterprises includes: using causal inference technology to analyze the causal relationships between different risk factors in the business network; using the risk quantification model to quantify the causal relationships to obtain the risk assessment results corresponding to the enterprises.

[0014] Among them, the method further includes: deeply analyzing the key business data of enterprise customers to obtain the business conditions and risk levels of the enterprises.

[0015] Among them, after obtaining the enterprise quality score, based on the capital and repayment situation of multiple downstream customers of the main company, comprehensively considering the status, influence, and risk level of the enterprise in the business network, dynamically adjust the loan amount of the main company or stop the loan.

[0016] Among them, the enterprise quality score is calculated by using the dynamic business network quality evaluation model in combination with the risk assessment results, association patterns, and business network knowledge graph, including: calculating the node influence in the business network knowledge graph by using the dynamic business network quality evaluation model; calculating the weight value of each transaction by using the dynamic business network quality evaluation model; calculating the comprehensive influence ratio of each node by using the dynamic business network quality evaluation model; calculating the quality score of each node by using the dynamic business network quality evaluation model; calculating the business ecosystem quality of the enterprise to be evaluated by using the dynamic business network quality evaluation model; calculating the status score of the enterprise to be evaluated in the business ecosystem by using the dynamic business network quality evaluation model; calculating the comprehensive status and influence score of the enterprise to be evaluated by using the dynamic business network quality evaluation model.

[0017] Among them, calculating the quality score of each node by using the dynamic business network quality evaluation model includes: performing quality scoring on each node by using expert scoring and machine learning model scoring, and normalizing the scoring results to obtain the node quality score.

[0018] The beneficial effects of this application are: different from the prior art, the evaluation method for dynamically calculating the quality based on the business network provided by this application includes: obtaining multi-source heterogeneous business network data; among them, the business network data includes: enterprise internal data, public data, and authorized data; performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; analyzing the business network knowledge graph by using a graph neural network to obtain the association pattern between enterprises; performing risk quantification on the business network knowledge graph by using a risk quantification model to obtain the corresponding risk assessment results of the enterprises; calculating the enterprise quality score by using the dynamic business network quality evaluation model in combination with the risk assessment results, association patterns, and business network knowledge graph. Through the above methods, make full use of advanced technologies such as big data, knowledge graph, and machine learning, get rid of the excessive dependence on the main body's credit and static financial data, and comprehensively and accurately evaluate the real operating conditions, risk levels, and development potential of enterprises from a dynamic business network perspective, so as to improve the efficiency and accuracy of credit decision-making of financial institutions, and ultimately promote the financing of small and medium-sized enterprises and support the high-quality development of the real economy. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings. Among them:

[0020] Figure 1 is a schematic flowchart of an embodiment of the evaluation method for the quality of business network dynamic computing provided by the present application;

[0021] Figure 2 is Figure 1 a schematic flowchart of an embodiment of step 13 in

[0022] Figure 3 is Figure 1 a schematic flowchart of an embodiment of step 15 in

[0023] Figure 4 is a schematic structural diagram of an embodiment of the electronic device provided by the present application;

[0024] Figure 5 is a schematic structural diagram of an embodiment of the computer-readable storage medium provided by the present application. Specific Embodiments

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. It can be understood that the specific embodiments described herein are only used to explain the present application, rather than limiting the present application. Additionally, it should be noted that for the sake of convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all the structures. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0026] Referring to "embodiments" in this context means that the specific features, structures, or characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0027] In modern manufacturing and the global economic system, small and medium-sized enterprises (SMEs) play a crucial role. Especially in the industrial chain and supply chain, they are an important source of innovation vitality and economic resilience. However, SMEs, especially those under the business-to-business (B2B) model, have long faced challenges in financing, including difficulties in obtaining financing and high financing costs. When traditional financial institutions conduct credit risk assessments, they overly rely on static data such as the credit of the enterprise entity and financial statements, resulting in the following drawbacks:

[0028] Information asymmetry and data gap: SMEs, especially manufacturing enterprises in the upstream of the industrial chain, have relatively low information transparency, and their financial data may be incomplete or difficult to reflect their true operating conditions. Moreover, overseas financial institutions lack effective local data for evaluating Chinese manufacturing enterprises going global.

[0029] Limitations of static assessment: Traditional credit assessment models are often static and unable to capture market changes, industry trends, and the dynamic evolution of the enterprise's business ecosystem in a timely manner. This leads to a lag in risk assessment results and is difficult to effectively support rapid credit decisions.

[0030] Lack of network perspective: Traditional assessment methods usually analyze the enterprise entity in isolation, ignoring the enterprise's position, value, and dynamic performance in the supply chain and business ecosystem network. This makes it impossible for risk assessment to comprehensively consider the impact of upstream and downstream related enterprises, and may underestimate systemic risks and the true quality of the enterprise.

[0031] Dependence on entity credit: Overemphasis on entity credit and financial data makes it difficult for those SMEs with good operations but shortfalls in entity credit or financial data to obtain necessary financial support, hindering their growth and expansion.

[0032] When existing financial institutions conduct enterprise quality assessment, they mainly face the following technical problems, which are also the bottlenecks that traditional methods are difficult to effectively solve:

[0033] Poor decision-making quality: Unable to effectively identify the dynamic risks of the industry and ecological environment. Existing models overly rely on the enterprise's own financial and credit data, lacking in-depth understanding and dynamic analysis capabilities of the industry and business ecological environment where the enterprise is located. With insufficient sample data, the model is difficult to learn and generalize, resulting in low-quality risk assessment, easy misjudgment of high-quality enterprises, and missed investment opportunities.

[0034] Static assessment method: Unable to reflect the dynamic changes of the market and enterprise conditions in a timely manner. Traditional static assessment methods cannot capture the real-time impact of external factors such as market fluctuations, policy adjustments, and changes in the competition pattern on the enterprise's operating conditions and risk levels, resulting in a lag in credit strategy adjustments and inability to adapt to the rapidly changing market environment.

[0035] Lack of supply chain perspective: Failure to fully consider the transmission effect of risks of upstream and downstream enterprises. Existing evaluation methods usually only focus on the credit risk of the main enterprise, ignoring the potential risks of upstream and downstream enterprises in the supply chain and their mutual transmission

[0036] effect. In a complex business network, a risk event of a single enterprise may quickly spread throughout the network, leading to systemic risks.

[0037] Limitations and insufficient depth of single entity analysis: Difficult to deeply explore and analyze the complex relationships between enterprises. Traditional methods are difficult to conduct in-depth analysis and monitoring of the upstream and downstream of the supply chain and the deeper business ecosystem, and cannot effectively identify and quantify the mutual influence between enterprises and the overall role of the business ecosystem. Especially in the ToB business, the related party transactions and dependencies between enterprises are more complex, and the single entity analysis method is difficult to handle.

[0038] Fragmented data utilization and insufficient value mining: Lack of effective methods to deeply explore the value of ToB business data. Enterprises generate a large amount of dynamic data in the supply chain and business ecosystem, especially the operation data and transaction data related to the ToB business. These data contain rich information on the enterprise's operating conditions and risks, but existing methods are difficult to effectively integrate and deeply explore the value of these data, resulting in a waste of data resources.

[0039] Based on this, the present application proposes an evaluation method for calculating the quality degree dynamically based on a business network. The evaluation method includes: obtaining multi-source heterogeneous business network data; wherein, the business network data includes: enterprise internal data, public data, and authorized data; performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; using a graph neural network to analyze the business network knowledge graph to obtain the association patterns between enterprises; using a risk quantification model to quantify the risks of the business network knowledge graph to obtain the corresponding risk assessment results of the enterprises; using a business network dynamic quality degree evaluation model to combine the risk assessment results, association patterns, and business network knowledge graph to perform quality degree scoring on the enterprises. Through the above methods, making full use of advanced technologies such as big data, knowledge graphs, and machine learning, getting rid of the excessive dependence on the entity's credit and static financial data, starting from the perspective of a dynamic business network, comprehensively and accurately evaluating the true operating conditions, risk levels, and development potential of enterprises, thereby improving the efficiency and accuracy of credit decision-making of financial institutions, ultimately promoting the financing of small and medium-sized enterprises, and supporting the high-quality development of the real economy. For specific reference, see any of the following embodiments.

[0040] See Figure 1 , Figure 1 which is a schematic flowchart of an embodiment of the evaluation method for calculating the quality degree dynamically based on a business network provided by the present application. The evaluation method includes:

[0041] Step 11: Acquire multi-source heterogeneous business network data; wherein the business network data includes: internal enterprise data, public data, and authorized data.

[0042] In some embodiments, intelligent data lake technology is used to build an adaptive, self-learning data lake platform that can automatically identify, classify, and integrate business network data from different sources and formats, including: internal enterprise data, public data, and authorized data.

[0043] Internal enterprise data mainly includes: business data, transaction data, customer data, supply chain data, etc. generated by the enterprise ERP system, financial system, CRM system, etc.

[0044] Public data mainly include: industrial and commercial information, legal litigation information, industry reports, macroeconomic data, market news and public opinion, etc.

[0045] Authorized data mainly includes: corporate credit data, industry data, transaction behavior data, etc. provided by third-party data service providers.

[0046] The above-mentioned intelligent data lake is a platform for centralized storage, management and processing of various structured, semi-structured and unstructured data. It has data governance, data mining and data analysis capabilities, and can support enterprises to build flexible and scalable data applications.

[0047] In this step, multi-source heterogeneous data are integrated to construct a more comprehensive representation of the problem state, which is applied to business network data integration, aiming to build a data foundation that can reflect the overall picture of corporate operations and the complexity of the business ecosystem.

[0048] Step 12: Process and integrate multi-source heterogeneous business network data to obtain a business network knowledge graph.

[0049] In some embodiments, data cleaning is performed on multi-source heterogeneous business network data, and enterprise-related information is extracted from the cleaned data as entities, and events and relationships between enterprises as edges to construct a business network knowledge graph.

[0050] Specifically, the integrated multi-source heterogeneous data is cleaned, converted, standardized and other pre-processing operations are performed, such as deduplication, missing value processing, outlier detection, data format unification, etc. Enterprise-related information is extracted from the cleaned data as entities (nodes), and events and relationships between enterprises are used as edges (relationships) to construct a business network knowledge graph.

[0051] The above-mentioned knowledge graph is a structured form of knowledge representation, which organizes knowledge in the form of a graph. Nodes represent entities (concepts or objects), and edges represent the relationships between entities. It can effectively express and organize complex knowledge and support knowledge reasoning and intelligent applications.

[0052] This solution uses the idea of graph structure modeling, applies it to the business network, and constructs a business network knowledge graph, aiming to more effectively express the complex association relationships and business ecosystem structures among enterprises.

[0053] Step 13: Analyze the business network knowledge graph using a graph neural network to obtain the association patterns between enterprises.

[0054] In some embodiments, a business network knowledge graph that can be continuously updated is constructed. This graph can capture and reflect the changes in the relationships between enterprises, the dynamic adjustments of the supply chain, and the evolution of the business ecosystem in real time. The entities, relationships, and attributes in the graph are continuously updated using real-time data streams to ensure the timeliness and accuracy of the knowledge graph.

[0055] The above-mentioned dynamic business network knowledge graph is a knowledge graph that can be updated and evolved in real time. It can continuously absorb new knowledge over time and reflect the changes in existing knowledge, and can more effectively meet the knowledge management and application requirements in a dynamic environment.

[0056] In this step, a dynamic business network knowledge graph is constructed to reflect the changes in the relationships between enterprises, and a dynamic business network knowledge graph is constructed to capture the dynamic characteristics of the business ecosystem and improve the timeliness and effectiveness of the evaluation.

[0057] In some embodiments, referring to Figure 2 , Step 13 can be the following process:

[0058] Step 131: Use a heterogeneous graph neural network to learn independent embedding representations for each type of node and edge in the business network knowledge graph.

[0059] Among them, the business network knowledge graph is a dynamic knowledge graph, and the graph neural network is a heterogeneous graph neural network.

[0060] Use a graph neural network (especially a heterogeneous graph neural network, Heterogeneous Graph Neural Network, HGNN)) to perform deep learning and analysis on the constructed business network knowledge graph. HGNN can effectively handle the diversity of node and edge types in the business network, capture node features, edge features, and network structure information, mine the hidden association patterns between enterprises, and quantify the association strength, so as to more accurately evaluate the status and influence of enterprises in the business network.

[0061] The above-mentioned graph neural network is a neural network specifically for processing graph-structured data. It can learn the representations of nodes in the graph and the representation of the graph, effectively mine the deep patterns and information in the graph data, and is widely used in fields such as social network analysis, recommendation systems, and knowledge graph reasoning.

[0062] The above heterogeneous graph neural network is a graph neural network that can process graph data with different node and edge types, and is more suitable for graph data analysis in complex scenarios, such as various types of nodes and relationships among enterprises, products, industries, etc. in a business network.

[0063] Use HGNN to model the task scheduling problem and apply it to the analysis of the business network knowledge graph, aiming to utilize the powerful graph data processing ability of HGNN to mine the deep associations and complex patterns in the business network and improve the accuracy and depth of enterprise quality assessment.

[0064] Step 132: When performing information aggregation, design different message passing functions according to the types of nodes and edges.

[0065] Step 133: Combine the attention mechanism to dynamically learn the importance of different neighbor nodes and different types of edges in the information aggregation process, and then obtain the association patterns between enterprises.

[0066] In some embodiments, the main introduction of the heterogeneous graph neural network is as follows:

[0067] 1. Understand the core advantage of the heterogeneous graph neural network (HGNN): handling heterogeneity.

[0068] First of all, it is necessary to understand that the key feature that differentiates HGNN from traditional graph neural networks (GNNs) lies in its "heterogeneity handling ability". "Heterogeneous" means that the nodes and edges in the graph are not homogeneous, but have different types and attributes. Traditional GNNs (homogeneous graph neural networks): usually assume that all nodes and all edges in the graph are of the same type. For example, in a social network, all relationships are between users, or in a paper citation network, all citations are between papers. They perform well in processing such homogeneous graph data.

[0069] HGNN (heterogeneous graph neural network): is specifically designed to process graphs with diverse node and edge types, such as real-world complex networks like business networks, biological networks, knowledge graphs, etc.

[0070] Among them, the business network has natural heterogeneity, and the business network is essentially a heterogeneous graph:

[0071] Diversity of node types: The business network contains various different types of entities. For example: Enterprises (different types): core enterprises, suppliers (primary, secondary, etc.), customers, competitors, financial institutions, logistics enterprises, technology service providers, etc. Products / services: raw materials, components, finished products, financial products, technology services, etc. Industries / domains: manufacturing, finance, Internet industry, energy industry, etc. Geographical locations: different countries, cities, regions, etc.

[0072] Diversity of edge types: There are many different types of relationships between entities, such as: Transaction relationships: procurement, sales, investment, loans, guarantees, etc. Cooperation relationships: strategic cooperation, technical cooperation, R&D cooperation, market cooperation, etc. Competitive relationships: direct competition, indirect competition. Supply chain relationships: upstream and downstream supply, distribution relationships. Intellectual property relationships: patent licensing, technology transfer. Equity relationships: investment, controlling, and equity participation.

[0073] Limitations of traditional GNNs: If traditional GNNs are used directly to process such heterogeneous business networks, the following problems will be encountered: Information confusion: Traditional GNNs treat all nodes and edges as the same type and cannot distinguish the different semantic information contained in different types of nodes and edges, resulting in information confusion and reduced feature expression capabilities. Feature loss: The information aggregation method of traditional GNNs may not be able to effectively capture and utilize the specific attributes and features of different types of nodes and edges, resulting in information loss. Relationship misunderstanding: Traditional GNNs have difficulty distinguishing the semantic differences and importance of different types of relationships, and may misunderstand the real connections between enterprises, resulting in biased analysis results.

[0074] 2. How does HGNN handle the heterogeneity of business networks and capture features?

[0075] HGNN effectively handles the heterogeneity of business networks and captures node features, edge features, and network structure information through the following key mechanisms:

[0076] Type-aware Node and Edge Embeddings: Mechanism: HGNN learns independent embeddings for each type of node and edge. This means that different types of enterprises (such as core enterprises and suppliers) will have different embedding spaces, and different types of relationships (such as transaction relationships and cooperation relationships) will also have different embedding spaces.

[0077] Function: Through type-dependent embedding, HGNN can distinguish the semantic differences between different types of nodes and edges, and encode these difference information into the embedding representation. For example, the embedding of the core enterprise will focus on reflecting its central position and resource control ability in the network, while the embedding of the supplier may focus more on reflecting its professional ability and degree of dependence on the core enterprise.

[0078] Type-aware Message Passing: Mechanism: When performing information aggregation, HGNN will design different message passing functions according to the types of nodes and edges. This means that the information transmission methods between different types of nodes may be different, and the roles played by edges of different types of relationships in information transmission may also be different.

[0079] Function: Through type-related message passing, HGNN can more finely control the flow and aggregation of information, making the information aggregation process more effective and accurate. For example, when dealing with transaction relationships, it may focus more on aggregating information such as transaction amounts and frequencies; when dealing with cooperation relationships, it may focus more on aggregating information such as cooperation scopes and depths.

[0080] Attention Mechanism: HGNN usually combines an attention mechanism to dynamically learn the importance of different neighbor nodes and different types of edges in the information aggregation process. The attention mechanism can automatically calculate the attention weights of each neighbor node or each type of edge based on the features of the current node and neighbor nodes. The higher the weight, the more important the neighbor node or the type of edge is in information aggregation.

[0081] Function: Through the attention mechanism, HGNN can adaptively select important neighbor nodes and key relationships for information aggregation, thereby better capturing key information and important patterns in the network and quantifying the importance of different relationships and neighbor nodes.

[0082] 3. How does HGNN discover hidden association patterns in business networks?

[0083] The above mechanisms of HGNN enable it to deeply discover various hidden association patterns in business networks. For example: Identifying key suppliers and customers: HGNN can identify key suppliers and customers crucial for enterprise operations and evaluate their risk levels and stability by analyzing transaction relationships, supply chain relationships, cooperation relationships, etc. among enterprises.

[0084] Discovering potential competitors and partners: HGNN can discover potential competitors and strategic partners by analyzing competition relationships, cooperation relationships, industry associations, etc. among enterprises, providing support for enterprises to formulate competition strategies and cooperation strategies.

[0085] Predicting risk propagation paths and impacts: HGNN can predict the propagation paths and impact scopes of risks in the business network by analyzing the association relationships and risk conduction mechanisms among enterprises, helping financial institutions identify systemic risks and associated risks.

[0086] Quantifying the degree of dependence and complementarity between enterprises: HGNN can quantify the degree of dependence and complementarity between enterprises and evaluate the status and value of enterprises in the business ecosystem by analyzing transaction data, cooperation data, intellectual property data, etc. among enterprises.

[0087] Revealing the evolution trends of industries and ecosystems: HGNN can capture the trends of the evolution of business networks over time by analyzing dynamic knowledge graphs, such as industry reshuffle, supply chain restructuring, ecosystem upgrading, etc., providing forward-looking insights for enterprises and financial institutions.

[0088] 4. How does HGNN quantify the association strength and evaluate the status and influence of enterprises?

[0089] HGNN quantifies the association strength and evaluates the status and influence of enterprises in the business network in the following ways:

[0090] Attention weights as indicators of association strength: The attention weights learned by the attention mechanism can directly reflect the importance of different neighbor nodes and different types of edges in the information aggregation process. The higher the attention weight, the higher the association strength can be regarded as. For example, if an enterprise has a high attention weight for its key suppliers, it indicates that the enterprise has a high degree of dependence on the supplier and a high association strength.

[0091] Node embeddings represent the information of network status: The node embeddings learned by HGNN can effectively encode the structural position and neighbor information of nodes in the network. By analyzing the node embeddings, various network centrality metrics (such as degree centrality, betweenness centrality, closeness centrality, etc.) can be calculated, and these centrality metrics can quantify the status and influence of enterprises in the business network. For example, an enterprise with a high degree centrality may have more partners in the network and relatively greater influence.

[0092] Edge embeddings represent the relationship strength: The edge embeddings learned by HGNN can encode the type and attribute information of edges, and the similarity of edge embeddings can reflect the strength of the relationship. For example, for a transaction relationship with a large transaction amount, its edge embedding may be more similar to the edge embeddings of other high-amount transaction relationships, indicating a higher transaction relationship strength.

[0093] Downstream task-driven association strength learning: The training of HGNN is usually end-to-end, driven by downstream tasks such as enterprise quality assessment. The model will automatically learn the association patterns and association strengths beneficial to downstream tasks during the training process. For example, if certain types of association relationships play an important role in predicting enterprise quality, HGNN will automatically enhance the attention and learning of these association relationships, thereby quantifying the strength of these association relationships.

[0094] The core reason why HGNN can more accurately evaluate the status and influence of enterprises in the business network lies in its powerful heterogeneous graph data processing ability. It can: effectively handle the diversity of node and edge types in the business network.

[0095] Capture node features, edge features, and network structure information.

[0096] Discover the hidden complex association patterns among enterprises.

[0097] Quantify the strength and importance of different types of association relationships.

[0098] Step 14: Use the risk quantification model to quantify the risks of the business network knowledge graph and obtain the corresponding risk assessment results for the enterprises.

[0099] Use causal inference technology to analyze the causal relationships among different risk factors in the business network; use the risk quantification model to quantify the causal relationships and obtain the corresponding risk assessment results for the enterprises.

[0100] Risk Quantification and Modeling.

[0101] Causal Inference based Risk Quantification Model: Technical means: Use causal inference technology to analyze the causal relationships among different risk factors in the business network, such as how risk factors such as macroeconomic downturn, industry policy changes, supply chain disruptions, and customer defaults interact with each other and ultimately transmit to the enterprises to be evaluated. Construct a Causal Graph Model to clearly display the risk transmission path and impact degree. Based on the Causal Graph Model, quantify different risk factors and calculate the paths and impacts of risk propagation, so as to more comprehensively and systematically evaluate the risk level faced by enterprises.

[0102] Technical term explanation: Causal inference: A method of inferring causal relationships from data, aiming to understand the causal connections between events, such as whether event A causes event B to occur. Causal inference has important application values in fields such as risk analysis, policy evaluation, and decision-making support.

[0103] Causal Graph Model: A probabilistic graphical model that represents causal relationships using a graph structure, where nodes represent variables and directed edges represent causal relationships, which can clearly display the causal relationships and impact paths between variables.

[0104] Use causal inference for risk quantification and apply it to business network risk assessment, aiming to use causal inference technology to deeply analyze the interactions and transmission mechanisms among complex risk factors in the business network and construct a more accurate and interpretable risk quantification model.

[0105] Among them, this method also includes: deeply analyzing the key business data of enterprise customers to obtain the business conditions and risk levels of the enterprises.

[0106] Deep Analysis of ToB Business Operation Data: Technical Means: Conduct in-depth analysis on the key business operation data of enterprise customers (ToB), with a focus on the following data types: Operating Cash Flow: Comprehensively record the capital flows generated by various business activities of the enterprise, reflecting the daily business scale and activity level of the enterprise.

[0107] Bank Statement: Detail the capital receipts and payments of the enterprise's bank accounts, including receipts, payments, transfers, etc., reflecting the specific details of the enterprise's capital turnover.

[0108] Invoice Records: Record the invoice information issued and received by the enterprise, reflecting the true transaction situation and transaction scale of the enterprise.

[0109] Equipment Information Corresponding to Bank Statement: Analyze the capital flows related to equipment procurement, leasing, etc. in the enterprise's bank statement to understand the enterprise's production capacity and expansion situation.

[0110] Downstream Customer Cash Flow Corresponding to Equipment Information: Track the sales situation of the enterprise's products through equipment information, analyze the capital flows of downstream customers, and evaluate their payment collection ability and risks.

[0111] Deep Analysis Dimensions: Conduct the following in-depth analysis on the above data: Income and Expenditure Analysis: Analyze the income and expenditure structure of the enterprise and evaluate its profitability and cash flow situation.

[0112] End-of-Day Balance Analysis: Analyze the changes in the daily bank account balance of the enterprise and evaluate its capital liquidity and short-term debt repayment ability.

[0113] Ranking of Downstream Customer Cash Flows: Rank the cash flows of the enterprise's downstream customers to identify major customers and potential risk customers.

[0114] Comparison between Cash Flow and Invoice: Compare the enterprise's bank statement with invoice records to verify the authenticity of transactions and identify false transactions and potential risks.

[0115] Calculation of Suspected Payment Period: Calculate the suspected payment period of downstream customers based on the enterprise's bank statement and invoice records, and evaluate their payment collection speed and credit status.

[0116] Identification of Related Party Transactions: Identify related party transactions between the enterprise and its upstream and downstream customers through technologies such as graph neural networks, and evaluate the risks of related party transactions.

[0117] Risk Assessment of Downstream Customers: Input the above information into the risk control model to obtain the capital risk level of downstream customers, and evaluate the difficulties in payment collection and bad debt risks.

[0118] In-depth analysis of enterprise operation data to more comprehensively evaluate the enterprise status. And it is concretized into in-depth analysis of key operation data for ToB business in multiple dimensions and at multiple levels, aiming to more accurately evaluate the operation status and risk level of ToB enterprises and provide more reliable data support for quality evaluation.

[0119] Step 15: Use the dynamic quality evaluation model of the business network to combine the risk assessment results, association patterns, and business network knowledge graph to score the enterprise quality.

[0120] Among them, the dynamic quality evaluation model of the business network is trained using federated learning technology.

[0121] In the context where data privacy protection is becoming increasingly important, in order to securely and compliantly use multi-party data for model training, this solution introduces federated learning technology. Using federated learning technology, without sharing the original data, it combines the data of multiple financial institutions, enterprises, and data providers to jointly train the enterprise quality evaluation model, realizes the sharing of multi-party data value, and at the same time ensures data privacy and security.

[0122] Federated learning is a distributed machine learning framework that allows multiple participants to collaboratively train a machine learning model without sharing the original data. Each participant only trains the model locally and regularly exchanges model parameters or gradient information, and finally aggregates to obtain a global model.

[0123] This solution applies federated learning to solve the problems of data privacy and security in the financial field, enabling more effective use of multi-party data to improve the performance of the enterprise quality evaluation model on the premise of data security and compliance.

[0124] Among them, the dynamic quality evaluation model of the business network is constructed using an interpretable machine learning model and an ensemble learning method.

[0125] Among them, after obtaining the enterprise quality score, based on the funds and repayment situation of multiple downstream customers of the main company, comprehensively considering the status, influence, and risk level of the enterprise in the business network, dynamically adjust the loan amount of the main company, or stop the loan.

[0126] Loan Amount Adjustment for Main Entity: Technical means: Based on the funds and repayment situation of multiple downstream customers of the main company, comprehensively considering factors such as the status, influence, and risk level of the enterprise in the business network, dynamically adjust the loan amount of the main company, or stop the loan. Achieve intelligent credit decision-making based on the results of dynamic quality assessment of the business network.

[0127] Apply the idea of decision optimization, apply the result of quality evaluation to credit decision-making, realize the intelligent adjustment of loan amount based on the result of dynamic quality evaluation, and improve the efficiency and accuracy of financial services.

[0128] In some embodiments, referring to Figure 3 , step 15 may include the following processes:

[0129] Step 151: Calculate the node influence in the business network knowledge graph using the business network dynamic quality evaluation model.

[0130] The model formula is as follows:

[0131] Node influence calculation formula: Node influence = Hierarchical influence coefficient * Node quantity coefficient * (Decay factor ^ Hierarchical distance).

[0132] Explanation of formula parameters:

[0133] Hierarchical influence coefficient: A preset parameter, set according to the level of the node in the business network (such as core enterprise, first-level supplier, second-level supplier, etc.). The higher the level, the greater the influence coefficient.

[0134] Node quantity coefficient: A preset parameter, set according to the number of nodes directly associated with this node. The more associated nodes, the greater the quantity coefficient.

[0135] Decay factor: A preset parameter, used to control the speed at which influence decays with hierarchical distance. The larger the decay factor, the faster the decay speed.

[0136] Hierarchical distance: The hierarchical distance between nodes in the business network. The farther the distance, the greater the hierarchical distance.

[0137] This model is used to evaluate the influence size of each node in the business network. The greater the influence of a node, the greater its contribution to the overall quality of the network.

[0138] Step 152: Calculate the weight value of each transaction using the business network dynamic quality evaluation model.

[0139] Define Transaction Weight Formula: Model formula: W(transaction) = Coefficient of node category * Transaction amount * Time decay factor

[0140] Explanation of formula parameters: Coefficient of node category: A preset parameter, set according to the transaction type (such as invoice, fund flow, etc.) and the node categories of both parties to the transaction (such as core enterprise, supplier, customer, etc.). Different types of transactions and node categories contribute differently to the quality.

[0141] Transaction amount: The specific amount of the transaction. The larger the amount, the higher the weight.

[0142] Time decay factor: Calculated based on the time difference between the transaction occurrence time and the current time. The longer the time, the smaller the decay factor and the lower the weight.

[0143] Time decay factor formula: Time decay factor = exp(-λ * time difference), where λ is the decay rate, a preset parameter.

[0144] This model is used to calculate the weight value of each transaction, and the weight value reflects the contribution degree of the transaction to the enterprise quality.

[0145] Calculating Transaction Weight Value: Model formula: Use the above transaction weight formula to calculate the weight value of each transaction.

[0146] Formula parameter configuration: The node relationship coefficient is set according to the relationship type between the two parties of the transaction (such as supplier, customer, etc.). The decay rate λ in the time decay factor can be adjusted according to the actual business scenario.

[0147] Step 153: Use the dynamic business network quality evaluation model to calculate the comprehensive influence ratio of each node.

[0148] The formula is as follows:

[0149] Node comprehensive influence = Σ(W(transaction) * hierarchical influence coefficient * node quantity coefficient * (decay factor ^ hierarchical distance)) / Σ(W(transaction) * hierarchical influence coefficient * node quantity coefficient * (decay factor ^ hierarchical distance) of all nodes).

[0150] This model is used to calculate the comprehensive influence ratio of each node in the business network. The higher the ratio, the greater the influence of the node in the network.

[0151] Step 154: Use the dynamic business network quality evaluation model to calculate the quality score of each node.

[0152] Scoring dimension: Refer to various factors, such as enterprise scale, financial status, industry position, historical transaction records, etc., and combine with preset scoring indicators for scoring.

[0153] Adopt expert scoring and machine learning model scoring to conduct quality scoring for each node, and normalize the scoring results to obtain the node quality score.

[0154] Step 155: Use the dynamic business network quality evaluation model to calculate the business ecosystem quality of the enterprise to be evaluated.

[0155] The formula is as follows:

[0156] The quality level of the business ecosystem of the enterprise to be evaluated = (Σ (quality level of node influence scope) / (number of nodes * 100)) * 100.

[0157] The quality level of node influence scope = quality level score of the node * proportion of node comprehensive influence.

[0158] This model is used to calculate the overall quality level of the enterprise to be evaluated in the business ecosystem, comprehensively considering the enterprise's own quality level and its influence in the network, and more comprehensively reflecting the value and potential of the enterprise.

[0159] Step 156: Calculate the status score of the enterprise to be evaluated in the business ecosystem using the dynamic quality level evaluation model of the business network.

[0160] The formula is as follows:

[0161] The status score of the enterprise to be evaluated in the business ecosystem = α * connection breadth + β * bridging role + γ * reach efficiency + δ * resource control power.

[0162] Control coefficient = a × bridging role + (1 - a) × reach efficiency, a = 0.4.

[0163] Explanation of formula parameters:

[0164] Connection breadth: It refers to the number of partners directly connected by an enterprise in the business network. The higher the connection breadth, the more central the node is in the network.

[0165] Bridging role: It refers to the key degree of an enterprise as a communication bridge between two other enterprises in the business network. The higher the bridging role, the more important the node is in the network.

[0166] Reach efficiency: It refers to the ease of an enterprise establishing connections with other enterprises in the business network, or the average efficiency of "reaching" other enterprises in the network. The higher the reach efficiency, the easier it is for the node to reach other nodes in the network.

[0167] Resource control power: It measures the ability of a node to control the resources and information flow of other nodes in the network. The stronger the control power, the higher the status of the node in the network.

[0168] α, β, γ, δ: Preset weight parameters used to adjust the influence degree of different centrality indicators on the status score, which can be adjusted according to the actual business scenario.

[0169] Control coefficient: Quantify the control degree and influence of the enterprise to be evaluated on resources and information flow in the business ecosystem.

[0170] a: A preset parameter used to adjust the weights of the bridge effect and reach efficiency in the control force coefficient, which can be adjusted according to the actual business scenario.

[0171] This model is used to calculate the status score of the enterprise to be evaluated in the business ecosystem, and quantify the importance and influence of the enterprise in the network from the perspective of network centrality.

[0172] Step 157: Calculate the comprehensive status and influence score of the enterprise to be evaluated using the business network dynamic quality evaluation model.

[0173] The formula is as follows:

[0174] The comprehensive status and influence score of the enterprise to be evaluated = (Business ecosystem quality + Status score of the business ecosystem + Quality of the enterprise to be evaluated) / 3

[0175] This model is used to synthesize the business ecosystem quality, status score and the quality of the enterprise to be evaluated itself to obtain a comprehensive enterprise quality score, which can more comprehensively reflect the true value and development potential of the enterprise.

[0176] Weight coefficient:

[0177] The weight coefficients of the three indicators in the formula can be adjusted according to the actual business scenario and evaluation focus. For example, if more emphasis is placed on the influence of the business ecosystem, the weights of the business ecosystem quality and status score can be appropriately increased.

[0178] The core of this model lies in "dynamic" and "self-evaluation".

[0179] Dynamic nature: The model fully considers the dynamic characteristics of the business network, uses a dynamic knowledge graph to capture network changes in real time, uses a time decay factor to reflect the timeliness of transactions, and uses in-depth ToB enterprise operation data analysis to reflect the dynamic changes in the enterprise operation status.

[0180] Self-evaluation: Through the node influence model, transaction weight model and comprehensive quality evaluation model, self-evaluate the quality of each node and the overall network in the business network, continuously iterate and optimize the evaluation results, and improve the accuracy and reliability of the evaluation.

[0181] Refer to Figure 4 , Figure 4 FIG. is a schematic structural diagram of an embodiment of an electronic device provided in the present application. The electronic device 40 includes a memory 41 and a processor 42 coupled to the memory 41. The memory 41 is used to store a computer program, and when the computer program is executed by the processor 42, it is used to implement the following method:

[0182] Obtain multi-source heterogeneous business network data; among them, the business network data includes: enterprise internal data, public data, and authorized data; perform data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; use a graph neural network to analyze the business network knowledge graph to obtain the association patterns between enterprises; use a risk quantification model to quantify the risk of the business network knowledge graph to obtain the corresponding risk assessment results of enterprises; use a business network dynamic quality evaluation model to combine the risk assessment results, association patterns, and business network knowledge graph to perform enterprise quality scoring.

[0183] In some embodiments, when the computer program is executed by the processor 42, it is further configured to implement the method of any of the above embodiments.

[0184] Refer to Figure 5 , Figure 5 FIG. is a schematic structural diagram of an embodiment of a computer-readable storage medium provided by the present application. The computer-readable storage medium 50 is used to store a computer program 51, and when the computer program 51 is executed by a processor, it is used to implement the following method:

[0185] Obtain multi-source heterogeneous business network data; among them, the business network data includes: enterprise internal data, public data, and authorized data; perform data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; use a graph neural network to analyze the business network knowledge graph to obtain the association patterns between enterprises; use a risk quantification model to quantify the risk of the business network knowledge graph to obtain the corresponding risk assessment results of enterprises; use a business network dynamic quality evaluation model to combine the risk assessment results, association patterns, and business network knowledge graph to perform enterprise quality scoring.

[0186] In some embodiments, when the computer program 51 is executed by a processor, it is further configured to implement the method of any of the above embodiments.

[0187] In summary, there is a clear and interlocking internal logical relationship among the technical means of the technical solution of the present application, which can be summarized as a progressive architecture of "data foundation - network construction - intelligent analysis - decision-making application - system support - value evaluation". They do not exist in isolation, but rely on each other and work together to jointly build a complete, efficient, and intelligent enterprise quality dynamic evaluation system.

[0188] The following is a detailed explanation of the internal logical relationship between each technical means:

[0189] 1. Data processing and integration provide a high-quality data foundation for subsequent links

[0190] Logical starting point: Data processing and integration are the cornerstone and starting point of the entire technical solution. The method for evaluating the dynamic quality of a business network, like any data-driven intelligent system, relies on high-quality data support.

[0191] Dependency: The results of the data processing and integration stage, such as integrated multi-source heterogeneous data, cleansed structured data, and constructed business network knowledge graphs, serve as the data input for subsequent associated network construction, risk quantification and modeling, and machine learning applications. Without high-quality, comprehensive, and structured data, the subsequent intelligent analysis and value extraction will be like "castles in the air".

[0192] Internal logic: "Garbage in, garbage out" is a fundamental principle in the field of data science. The goal of the data processing and integration stage is to achieve "garbage out, gold in", that is, through means such as cleansing, integration, and modeling, to transform raw, messy, and heterogeneous data into high-quality, structured, and information-rich "gold" data, laying a solid foundation for subsequent intelligent analysis and value extraction.

[0193] 2. Network construction: Mining the business network structure and dynamic characteristics from data

[0194] Linking the preceding and the following: The associated network construction stage builds on the results of data processing and integration and provides the basis for network analysis in risk quantification and modeling. It utilizes the business network knowledge graph constructed in the data processing and integration stage to further mine network structure and dynamic characteristics.

[0195] Dependency: The associated network construction stage depends on the business network knowledge graph constructed in the data processing and integration stage. The construction of a dynamic knowledge graph requires high-quality and real-time updated data streams, and the training of graph neural networks also needs to be based on the structured data of the knowledge graph.

[0196] Internal logic: Data itself is scattered and isolated, making it difficult to directly reflect an enterprise's position, influence, and risks in the business ecosystem. The goal of the associated network construction stage is to "integrate the fragmented" and "endow with intelligence through graphs", connecting scattered data points into an organic business network, and using technologies such as knowledge graphs and graph neural networks to mine deeper business rules and enterprise characteristics from the network structure and dynamic evolution, providing richer network context information for subsequent intelligent analysis and risk assessment.

[0197] 3. Intelligent analysis: Using network information for risk assessment and value extraction.

[0198] Core Processes: Risk quantification and modeling and the application of machine learning are the core processes of the entire technical solution and the key to demonstrating intelligence. These two processes together build the "brain" of the dynamic evaluation model for enterprise quality.

[0199] Dependency: The risk quantification and modeling process depends on the business network structure and dynamic features provided by the associated network construction process. The causal inference model needs to utilize the network association relationships and feature representations mined by the graph neural network, and in-depth ToB data analysis also requires a more in-depth interpretation in the context of the business network.

[0200] In the machine learning application process, whether it is interpretable machine learning or ensemble learning, model training and optimization need to be based on the data prepared in the data processing and integration process and the network features extracted in the associated network construction process.

[0201] Internal Logic: The business network knowledge graph constructed in the associated network construction process, although providing rich network structure information, is still only "static knowledge" itself. The goals of the risk quantification and modeling and machine learning application processes are "knowledge monetization" and "intelligent decision-making". By using intelligent technologies such as causal inference, in-depth data analysis, and machine learning, the value contained in the business network knowledge graph is mined and transformed into quantifiable risk assessment results and executable credit decision support, truly realizing the intelligent application of data.

[0202] 4. Decision Application: Apply the evaluation results to credit decisions.

[0203] Value Embodiment: The link of adjusting the loan amount of the main company is the ultimate embodiment of the value of the technical solution. The ultimate goal of the dynamic evaluation model for enterprise quality is to serve the credit decisions of financial institutions, and the adjustment of the loan amount is the most direct and crucial manifestation of credit decisions.

[0204] Dependency: The link of adjusting the loan amount of the main company directly depends on the evaluation results of enterprise quality in the risk quantification and modeling process. The adjustment of the loan amount needs to be based on the evaluation results of enterprise quality, and the loan amount is dynamically adjusted or the loan is stopped according to the risk level and development potential of the enterprise.

[0205] Internal Logic: The ultimate goal of the technical solution is not just to stay at the technical level, but to "implement technology" and "create value", applying advanced intelligent technologies to actual financial business scenarios, solving the pain points of difficult and expensive financing for small and medium-sized enterprises, improving the efficiency and accuracy of credit decisions of financial institutions, and ultimately promoting the high-quality development of the real economy.

[0206] 5. System Support: Ensure the stable and efficient operation of the system

[0207] Infrastructure: The system architecture design phase is the infrastructure and guarantee of the entire technical solution. Cloud-native architecture and microservices architecture provide solid technical support for the stable operation, efficient expansion, and flexible maintenance of the enterprise quality dynamic evaluation system.

[0208] Service guarantee: The system architecture design phase serves all aspects of the entire technical solution. Whether it is data processing and integration, associated network construction, risk quantification and modeling, or machine learning applications, a stable, reliable, and scalable system platform is required to host and run them.

[0209] Internal logic: No matter how advanced the algorithms and models are, a stable and efficient system platform is needed to support their operation and application. The goal of the system architecture design phase is to "build a nest to attract phoenixes" and "escort", and build a cloud-native system platform with high performance, high availability, and easy maintenance, providing solid infrastructure guarantee for the implementation and application of the enterprise quality dynamic evaluation model.

[0210] Value assessment: The core assessment model runs through the whole process.

[0211] Core model: The business network dynamic quality evaluation model is the core and soul of the entire technical solution. It runs through all aspects of the technical solution, integrates the achievements of all aspects, and finally outputs the enterprise quality score.

[0212] Carrying the achievements of all aspects: The business network dynamic quality evaluation model integrates the achievements of data processing and integration, associated network construction, risk quantification and modeling, and machine learning applications. Based on the business network knowledge graph, it uses the network features extracted by graph neural networks, combines the risk factors quantified by causal reasoning, and the business data obtained from in-depth ToB data analysis, and finally calculates the enterprise quality score.

[0213] Internal logic: The goal of the business network dynamic quality evaluation model is to be the "master of all achievements" and "value output", integrating the achievements of all technical aspects, forming a complete and self-consistent evaluation system, and finally outputting a credible and usable enterprise quality score, providing the most direct and crucial reference basis for the credit decision-making of financial institutions.

[0214] Summary:

[0215] A complete logical closed-loop of data-driven, network-empowered, intelligent analysis, decision-making application, system support, and value evaluation is formed among the technical means. Each link is indispensable and interlocked, jointly supporting the effective operation and value realization of the innovative technical solution of "Evaluation Method for Quality Degree Based on Dynamic Calculation of Business Network". This strict logical relationship ensures the systematicness, integrity, and feasibility of the technical solution, enabling it to truly address the challenges faced by existing technologies in enterprise quality degree evaluation and bringing tangible value to financial institutions and small and medium-sized enterprises.

[0216] Collaboration and non-obviousness of technical means:

[0217] The various technical means of this technical solution are not simply superimposed, but are organically coordinated and deeply integrated, forming an overall solution with complementary advantages and powerful functions, reflecting the non-obviousness and innovative breakthroughs compared with existing technologies:

[0218] Data integration + knowledge modeling: The integration of multi-source heterogeneous data constructs a comprehensive data foundation. Data cleaning and knowledge modeling transform massive unstructured data into structured and computable knowledge graphs, laying a foundation for subsequent graph neural network analysis and risk quantification modeling. This data processing method transcends the dependence of traditional methods on single data sources and structured data, achieving the effective integration and value mining of multi-source heterogeneous data.

[0219] Dynamic knowledge graph + graph neural network: The dynamic knowledge graph captures the changes in the business network in real time, and the graph neural network uses the structured information of the knowledge graph for deep learning to mine complex associations and potential patterns in the network. The combination of the two realizes the effective modeling and analysis of the dynamic characteristics of the business network, overcomes the limitations of traditional static evaluation methods, and can more accurately evaluate the status and influence of enterprises in the dynamic business network.

[0220] Causal reasoning + in-depth ToB data analysis: Causal reasoning technology is used to analyze the causal relationships among risk factors in the business network, and in-depth ToB data analysis provides rich enterprise operation data as an important basis for risk quantification. The combination of the two realizes the systematic and in-depth analysis of business network risks, overcomes the shortcomings of traditional methods lacking a network perspective and insufficient depth, and can more comprehensively and accurately evaluate the risk level faced by enterprises.

[0221] Self-evaluation model + interpretable machine learning: The business network dynamic quality evaluation model draws on the idea of self-evaluation and realizes the dynamic and iterative evaluation of network quality. Interpretable machine learning and ensemble learning ensure the interpretability, accuracy, and generalization ability of the model. The combination of the three constructs an interpretable, high-precision, and self-optimizing enterprise quality evaluation system, overcoming the deficiencies of poor interpretability and weak generalization ability in traditional method models and enhancing the credibility and practicality of evaluation results.

[0222] Examples of synergy effects:

[0223] Synergy effect 1: The integration of multi-source heterogeneous data and the construction of a dynamic knowledge graph provide high-quality and dynamically updated data inputs for the graph neural network, enabling the graph neural network to learn more comprehensive and real-time business network features, thereby enhancing the accuracy of the graph neural network in analyzing enterprise association relationships and risk propagation paths.

[0224] Synergy effect 2: The enterprise association relationships and network structure information mined by the graph neural network can serve as important input features for the causal reasoning model, helping the causal reasoning model to more accurately identify risk conduction paths and quantify risk impacts, and enhancing the accuracy and reliability of the risk quantification model.

[0225] Synergy effect 3: The results of in-depth ToB data analysis can serve as an important basis for node quality scoring and business ecosystem quality evaluation, enriching the dimensions of quality evaluation and making the quality evaluation results more comprehensive and objective.

[0226] Synergy effect 4: The self-evaluation model can timely detect and correct evaluation biases through continuous iterative optimization of evaluation results, improving the accuracy and robustness of evaluation. Interpretable machine learning can help people understand the decision-making process of the self-evaluation model and enhance trust in evaluation results.

[0227] Non-obviousness:

[0228] Applying the technical idea of the "Self-Evaluation for Job-Shop Scheduling" paper to the dynamic evaluation of enterprise quality is not obvious and is an innovative cross-domain application.

[0229] Domain crossing: The "Self-Evaluation for Job-Shop Scheduling" paper mainly addresses optimization problems in the task scheduling field, while enterprise quality evaluation belongs to the fields of financial risk management and credit decision-making. Transferring the self-evaluation idea in the task scheduling field to the enterprise quality evaluation field requires a profound understanding of the commonalities and differences between the two fields and creative transformation and application.

[0230] Problem Abstraction: The problem of enterprise quality evaluation is far more complex than the Job-Shop Scheduling problem, involving multi-dimensional and multi-modal data, as well as complex business network association relationships. How to abstract the enterprise quality evaluation problem into a self-evaluation decision-making process similar to the Job-Shop Scheduling problem requires in-depth thinking and innovative modeling.

[0231] Technology Integration: Integrating the self-evaluation idea of the Self-Evaluation for Job-Shop Scheduling paper with technologies such as graph neural networks and causal reasoning to build a complete and implementable technical solution requires overcoming many technical challenges and is not a simple technical stack.

[0232] 4. Technical Effects

[0233] By adopting the above technical means, this technical solution is expected to achieve the following improvement effects and practical application results:

[0234] More Comprehensive Risk View: Based on the traditional entity credit evaluation, by analyzing the upstream and downstream enterprises associated with the entity and a wider business ecosystem network, a more comprehensive risk view can be obtained, including not only the risks of the enterprise entity itself but also external environmental risks such as supply chain risks, industry risks, and market demand changes, thus more accurately identifying and assessing the real risks faced by the enterprise.

[0235] Improved Risk Management Model: Combining the data of upstream and downstream enterprises and the business ecosystem network to build a risk quantification model based on causal reasoning, making the risk management model more accurate and effective, capable of more effectively predicting and preventing enterprise risks and reducing credit risks.

[0236] Improved Decision Quality: Using the dynamic quality evaluation model of the business network, the quality of enterprises can be evaluated more comprehensively and dynamically, improving the ability of financial institutions to identify enterprise risks and values, optimizing credit decisions, providing a more fair and transparent financing environment for enterprises, and promoting the more efficient allocation of financial resources.

[0237] Adapt to Market Changes: Through the dynamic knowledge graph and real-time data stream, effectively monitor the risk transmission and business ecosystem changes during the financing process to ensure that financial institutions can respond to market changes in a timely manner, adjust credit strategies, and reduce the risks brought by market fluctuations.

[0238] Promote Financing Opportunities: Break down the financing barriers for small and medium-sized manufacturing enterprises, reduce the excessive reliance on entity credit and collateral guarantees, provide financing opportunities for more high-quality small and medium-sized enterprises, and promote the healthy development of the real economy.

[0239] Dynamic Monitoring Capability: By monitoring and dynamically evaluating the quality and risk levels of enterprises in real time, provide timely and comprehensive risk warning information and decision-making support for financial institutions, and achieve intelligent and refined management of credit business.

[0240] Improved Risk Assessment Accuracy: By considering the mutual influence among enterprises and the overall role of the business ecosystem, as well as in-depth data analysis of ToB business, significantly improve the accuracy and recall rate of financial institutions in identifying enterprise risks, reduce the misjudgment rate, and enhance the overall quality of risk assessment.

[0241] Adaptability and Flexibility: Have strong adaptability and flexibility, and can flexibly adjust model parameters and evaluation indicators and customize configurations according to different industries, enterprises of different scales, and different business scenarios to meet the personalized needs of different financial institutions and enterprises.

[0242] Supply Chain Perspective: Through supply chain perspective analysis, it is possible to evaluate the status and influence of an enterprise in its business ecosystem, better understand the operational risks of the enterprise, and provide more accurate risk assessment and credit support for supply chain finance business.

[0243] In several implementation manners provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation manners described above are only illustrative. For example, the division of the modules or units is only a logical function division, and there can be other division manners in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0244] If the integrated units in the above-mentioned other embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0245] The above is only the embodiment of this application, and it does not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of this application by the same token.

Claims

1. An evaluation method for dynamically calculating the quality level based on a business network, characterized in that The evaluation method includes: Obtaining multi-source heterogeneous business network data; wherein, the business network data includes: enterprise internal data, public data, and authorized data; Performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph; Analyzing the business network knowledge graph using a graph neural network to obtain the association patterns between enterprises; Performing risk quantification on the business network knowledge graph using a risk quantification model to obtain the risk assessment results corresponding to the enterprises; Using a business network dynamic quality evaluation model to combine the risk assessment results, the association patterns, and the business network knowledge graph to perform enterprise quality scoring.

2. The evaluation method according to claim 1, wherein The performing data processing and integration on the multi-source heterogeneous business network data to obtain a business network knowledge graph includes: Performing data cleaning on the multi-source heterogeneous business network data, and extracting enterprise-related information from the cleaned data as entities, and events and relationships between enterprises as edges to construct the business network knowledge graph.

3. The evaluation method according to claim 1, wherein The business network dynamic quality evaluation model is trained using federated learning technology.

4. The evaluation method according to claim 3, wherein The business network dynamic quality evaluation model is constructed using an interpretable machine learning model and an ensemble learning method.

5. The evaluation method according to claim 1, characterized in that The business network knowledge graph is a dynamic knowledge graph, and the graph neural network is a heterogeneous graph neural network. The analyzing the business network knowledge graph using the graph neural network to obtain the association patterns between enterprises includes: Using the heterogeneous graph neural network to learn independent embedding representations for each type of node and edge in the business network knowledge graph; And when performing information aggregation, designing different message passing functions according to the types of nodes and edges; And combining the attention mechanism to dynamically learn the importance of different neighbor nodes and different types of edges in the information aggregation process, thereby obtaining the association patterns between enterprises.

6. The evaluation method according to claim 1, characterized in that The performing risk quantification on the business network knowledge graph using a risk quantification model to obtain the risk assessment results corresponding to the enterprises includes: Using causal inference technology to analyze the causal relationships between different risk factors in the business network; Using the risk quantification model to perform risk quantification on the causal relationships to obtain the risk assessment results corresponding to the enterprises.

7. The evaluation method according to claim 1, wherein The method further includes: Performing in-depth analysis on the key business data of enterprise customers to obtain the business conditions and risk levels of the enterprises.

8. The evaluation method according to claim 1, characterized in that After obtaining the enterprise quality score, based on the funds and payment collection situations of multiple downstream customers of the main company, comprehensively considering the status, influence, and risk level of the enterprise in the business network, dynamically adjusting the loan amount of the main company, or stopping the loan.

9. The evaluation method according to claim 1, characterized in that The using a business network dynamic quality evaluation model to combine the risk assessment results, the association patterns, and the business network knowledge graph to perform enterprise quality scoring includes: Using the business network dynamic quality evaluation model to calculate the node influence in the business network knowledge graph; Using the business network dynamic quality evaluation model to calculate the weight value of each transaction; Using the business network dynamic quality evaluation model to calculate the comprehensive influence ratio of each node. Calculate the quality score of each node using the dynamic quality evaluation model of the business network; Calculate the quality of the business ecosystem of the enterprise to be evaluated using the dynamic quality evaluation model of the business network; Calculate the status score of the enterprise to be evaluated in the business ecosystem using the dynamic quality evaluation model of the business network; Calculate the comprehensive status and influence score of the enterprise to be evaluated using the dynamic quality evaluation model of the business network.

10. The evaluation method according to claim 9, characterized in that, The calculation of the quality score of each node using the dynamic quality evaluation model of the business network includes: Use expert scoring and machine learning model scoring to perform quality scoring on each node, and normalize the scoring results to obtain the node quality score.

Images