Financial lease risk prediction method and system based on knowledge graph

Abstract

The invention relates to the technical field of data processing, and particularly discloses a financing lease risk prediction method and system based on a knowledge graph, and the method comprises the steps: obtaining structured data, unstructured data and real-time transaction flow data from a financing lease service system to form an initial knowledge unit set; identifying entities and semantic relationships from the initial knowledge unit set, performing entity alignment and knowledge fusion, and constructing a dynamically updated financing lease knowledge graph; capturing dynamic association features between entities in the financing lease knowledge graph, extracting a time sequence risk evolution mode of historical transaction data, generating a composite risk feature vector fusing static attributes and dynamic behaviors, and inputting the composite risk feature vector to an integrated model fusing a multilayer perceptron and XGBoost; the lessee default probability, the equipment asset depreciation rate and the industry risk conduction intensity are output; a lessee financing lease risk prediction result is further generated based on a model output result; and scientificity, accuracy and comprehensiveness of financing lease risk prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly relates to a financial leasing risk prediction method and system based on a knowledge graph. Background Art

[0002] At present, in today's economic environment, financial leasing, as an important financial service model, plays a key role in promoting enterprise equipment renewal and industrial upgrading. However, due to the involvement of multiple parties and complex transaction structures in financial leasing business, the potential risk assessment and prediction face many challenges.

[0003] Traditional methods cannot integrate structured data, unstructured data and real-time stream data, resulting in information silos. A cross-entity knowledge graph is not constructed, making it difficult to identify hidden risk paths. Risk feature engineering mainly focuses on static attributes (such as asset-liability ratio, equipment valuation), ignoring the evolution laws of dynamic time-series features such as transaction behavior and industry cycle fluctuations. Mainstream methods use single models such as logistic regression and random forest, which have insufficient ability to capture the non-linear relationships of high-dimensional heterogeneous data and weak model interpretability. There is a lack of dynamic modeling of industry risks transmitted to financial leasing business through the supply chain and capital chain, resulting in a lag in systemic risk warning.

[0004] Therefore, the present invention proposes a financial leasing risk prediction method and system based on a knowledge graph. Summary of the Invention

[0005] The present invention provides a financing lease risk prediction method and system based on a knowledge graph, including: obtaining various data from a financing lease business system to form an initial knowledge unit set, integrating structured data, unstructured data, and real-time transaction flow data, breaking information barriers, and providing a comprehensive data basis for risk prediction. Constructing a dynamically updated financing lease knowledge graph, introducing entity alignment algorithms and knowledge fusion technologies, establishing a multi-hop association network covering multiple parties, and dynamically updating the graph through real-time data flow, identifying implicit risk propagation paths can timely reflect business changes and new information, improving the timeliness and accuracy of risk assessment. Capturing the dynamic association features between entities and extracting risk evolution patterns, generating a composite risk feature vector, realizing the feature coupling of static attributes and dynamic behaviors, and more comprehensively and deeply depicting risk features. Using an integrated model that combines a multi-layer perceptron and XGBoost for prediction, where the multi-layer perceptron learns high-dimensional non-linear features, and XGBoost enhances the analysis of sparse features and time series trends, giving full play to the advantages of both models, improving the generalization ability of the model, and enhancing the accuracy and reliability of prediction. Introducing an industry risk conduction intensity indicator, quantifying the risk conduction probability based on the industry node topology structure in the knowledge graph, and realizing cross-level prediction from micro-enterprise default to macro-industry risk. Generating a risk prediction result based on multi-faceted risk indicators, providing a more comprehensive and accurate reference basis for financing lease decisions. The overall solution solves the deficiencies of traditional methods in three levels: data dimension, entity association modeling, and risk dynamic evolution through dynamic association mining of the knowledge graph and multi-modal data fusion; at the same time, the integrated model design takes into account both prediction accuracy and interpretability, providing panoramic early warning support for financing lease business from individual default to systemic risk. It improves the scientificity, accuracy, and comprehensiveness of financing lease risk prediction.

[0006] The present invention provides a financing lease risk prediction method based on a knowledge graph, including:

[0007] S1: Obtaining structured data, unstructured data, and real-time transaction flow data from a financing lease business system and forming an initial knowledge unit set;

[0008] S2: Identifying entities and semantic relationships from the initial knowledge unit set, performing entity alignment and knowledge fusion, and constructing a dynamically updated financing lease knowledge graph;

[0009] S3: Capturing the dynamic association features between entities in the financing lease knowledge graph, and extracting the time-series risk evolution pattern of historical transaction data, generating a composite risk feature vector that integrates static attributes and dynamic behaviors;

[0010] S4: Inputting the composite risk feature vector into an integrated model that combines a multi-layer perceptron and XGBoost, and outputting the lessee default probability, equipment asset depreciation rate, and industry risk conduction intensity;

[0011] S5: Generate the risk prediction result of the lessee's financial lease based on the lessee's default probability, the depreciation rate of equipment assets, and the intensity of industry risk transmission.

[0012] Preferably, for the financial lease risk prediction method based on a knowledge graph, S1: Obtain structured data, unstructured data, and real-time transaction flow data from the financial lease business system and form an initial knowledge unit set, including:

[0013] Obtain structured data, unstructured data, and real-time transaction flow data from the financial lease business system, including the lessee's credit record, equipment asset status data, industry economic indicators, contract clause text, and historical default records;

[0014] Clean, denoise, and standardize the obtained structured data, unstructured data, and real-time transaction flow data, and extract entity, attribute, and relationship triples to generate an initial knowledge unit set.

[0015] Preferably, for the financial lease risk prediction method based on a knowledge graph, S2: Identify entities and semantic relationships from the initial knowledge unit set, and perform entity alignment and knowledge fusion to construct a dynamically updated financial lease knowledge graph, including:

[0016] Define the core entity types and relationship types in the financial lease field based on the ontology modeling method. The entity types include lessee, leased equipment, guarantor, industry classification, and contract terms, and the relationship types include guarantee association, equipment mortgage status, and industry risk transmission path;

[0017] Adopt a joint entity relationship extraction model based on BERT to identify entities and semantic relationships from the initial knowledge unit set, calculate the entity semantic similarity using pre-trained domain word vectors, and aggregate neighborhood entity information combined with a graph attention network to generate entity alignment weights;

[0018] Through a cross-source conflict resolution algorithm, vote on the attribute conflicts of the same entity in multiple data sources and retain the attribute value with the highest confidence to obtain a dynamically updated financial lease knowledge graph.

[0019] Preferably, for the financial lease risk prediction method based on a knowledge graph, S3: Capture the dynamic association features between entities in the financial lease knowledge graph, and extract the temporal risk evolution pattern of historical transaction data to generate a composite risk feature vector that combines static attributes and dynamic behaviors, including:

[0020] Adopt a graph embedding algorithm to analyze the financial lease knowledge graph to generate low-dimensional vector representations of entities and relationships;

[0021] Concatenate the entity node features in the financial leasing knowledge graph with the timestamp embedding vector, and input it into the gated graph convolutional layer to capture the time-varying association strength between entities as the dynamic association features between entities;

[0022] Extract the local temporal patterns of historical transaction data through the sliding time window mechanism, and perform cross-modal fusion with the global graph features to obtain the temporal risk evolution pattern of historical transaction data;

[0023] Based on the low-dimensional vector representations of entities and relationships, the dynamic association features between entities, and the temporal risk evolution pattern of historical transaction data, generate a composite risk feature vector that fuses static attributes and dynamic behaviors.

[0024] Preferably, for the financial leasing risk prediction method based on the knowledge graph, S5: Generate the financial leasing risk prediction result of the lessee based on the lessee's default probability, equipment asset depreciation rate, and industry risk transmission intensity, including:

[0025] Obtain the three-party defined weights of the user for the lessee's default probability, equipment asset depreciation rate, and industry risk transmission intensity;

[0026] Based on the three-party defined weights of the user for the lessee's default probability, equipment asset depreciation rate, and industry risk transmission intensity, as well as the lessee's default probability, equipment asset depreciation rate, and industry risk transmission intensity, calculate the financial leasing risk prediction value of the lessee;

[0027] When the financial leasing risk prediction value of the lessee does not exceed the preset prediction threshold, then regard the financial leasing risk prediction value of the lessee as the financial leasing risk prediction result;

[0028] When the financial leasing risk prediction value of the lessee exceeds the preset prediction threshold, then extract the guarantee chain, equipment mortgage status, and upstream and downstream enterprises in the industry associated with the lessee from the financial leasing knowledge graph to generate a visual risk propagation subgraph;

[0029] Calculate the risk influence weight of each node in the risk propagation subgraph, label the key risk nodes and transmission paths based on the risk influence weights of all nodes in the risk propagation subgraph to form a key risk path, and generate a risk traceability report as the financial leasing risk prediction result of the lessee in combination with the financial leasing risk prediction value of the lessee.

[0030] Preferably, for the financial leasing risk prediction method based on the knowledge graph, calculate the risk influence weight of each node in the risk propagation subgraph, label the key risk nodes and transmission paths based on the risk influence weights of all nodes in the risk propagation subgraph to form a key risk path, including:

[0031] Determine the initial value of the risk influence weights of all nodes in the risk propagation subgraph based on the total number of nodes in the risk propagation subgraph;

[0032] Obtain the risk correlation value and risk propagation intensity of adjacent nodes in the risk propagation sub-graph;

[0033] Based on the risk correlation value and risk propagation intensity of adjacent nodes in the risk propagation sub-graph and a preset iteration formula, perform iterative calculation on the initial values of the risk influence weights of all nodes in the risk propagation sub-graph until the difference between the risk influence weights obtained by all nodes in the risk propagation sub-graph after the latest iteration process and the risk influence weights obtained after the previous iteration process is less than the preset threshold. Then, in the risk propagation sub-graph, all nodes whose risk influence weights obtained after the latest iteration process are not less than the preset risk influence weight threshold are marked as all key risk nodes;

[0034] Based on all key risk nodes in the risk propagation sub-graph, perform conduction path fitting to form a key risk path.

[0035] Preferably, for the financial leasing risk prediction method based on a knowledge graph, obtaining the risk correlation value and risk propagation intensity of adjacent nodes in the risk propagation sub-graph includes:

[0036] Generate a risk-related feature vector for each node based on the business transaction data of each node in the risk propagation sub-graph;

[0037] Take the similarity between the risk-related feature vectors of adjacent nodes in the risk propagation sub-graph as the risk correlation value of adjacent nodes;

[0038] Based on the total transaction amount and transaction frequency between adjacent nodes in the risk propagation sub-graph, calculate the risk propagation intensity of adjacent nodes.

[0039] Preferably, for the financial leasing risk prediction method based on a knowledge graph, the preset iteration formula includes:

[0040]

[0041] Wherein, RIW k+1 (i) is the risk influence weight of the i-th node in the risk propagation sub-graph after the (k + 1)-th iteration, α is the risk propagation tendency coefficient, n is the total number of nodes in the risk propagation sub-graph, Relevance(j, i) is the risk correlation value between the j-th node and the i-th node in the risk propagation sub-graph, M(i) is the set of all nodes pointing to the i-th node in the risk propagation sub-graph, RIW k(j) is the risk influence weight of the i-th node in the risk propagation sub-graph after the k-th iteration, Strength(j,i) is the risk propagation strength from the j-th node to the i-th node in the risk propagation sub-graph, O(j) is the set of all nodes pointed to by the j-th node in the risk propagation sub-graph, and Strength(j,k) is the risk propagation strength from the j-th node to the k-th node in the risk propagation sub-graph.

[0042] Preferably, the financing lease risk prediction method based on the knowledge graph fits the conduction paths based on all the key risk nodes in the risk propagation sub-graph to form key risk paths, including:

[0043] Fitting at least one conduction path based on the edges between all the key risk nodes in the risk propagation sub-graph;

[0044] When there is only one conduction path, the only conduction path is regarded as the key risk path;

[0045] When there are more than one conduction paths, the comprehensive conduction probability of each conduction path is calculated based on the risk correlation values and risk propagation strengths between all adjacent key risk nodes in each conduction path, and the conduction path with the maximum comprehensive conduction probability among all the conduction paths is regarded as the key risk path.

[0046] The present invention provides a financing lease risk prediction system based on the knowledge graph for executing any of the above financing lease risk prediction methods based on the knowledge graph, including:

[0047] An initial knowledge unit construction module for obtaining structured data, unstructured data, and real-time transaction flow data from the financing lease business system and forming an initial knowledge unit set;

[0048] A knowledge graph construction module for identifying entities and semantic relationships from the initial knowledge unit set, and performing entity alignment and knowledge fusion to construct a dynamically updated financing lease knowledge graph;

[0049] A composite risk feature vector generation module for capturing the dynamic association features between entities in the financing lease knowledge graph, and extracting the temporal risk evolution pattern of historical transaction data to generate a composite risk feature vector integrating static attributes and dynamic behaviors;

[0050] A model prediction and evaluation module for inputting the composite risk feature vector into an integrated model that combines a multi-layer perceptron and XGBoost, and outputting the lessee default probability, equipment asset depreciation rate, and industry risk conduction intensity;

[0051] A risk prediction module for generating a financing lease risk prediction result of the lessee based on the lessee default probability, equipment asset depreciation rate, and industry risk conduction intensity.

[0052] The beneficial effects of the present invention compared with the prior art are as follows: obtaining various data from the financial leasing business system to form an initial knowledge unit set, integrating structured data, unstructured data and real-time transaction flow data, breaking the information barrier, and providing a comprehensive data basis for risk prediction. Constructing a dynamically updated knowledge graph of financial leasing, introducing entity alignment algorithms and knowledge fusion technologies, establishing a multi-hop association network covering multiple parties, and dynamically updating the graph through real-time data flow, identifying implicit risk propagation paths can timely reflect business changes and new information, improving the timeliness and accuracy of risk assessment. Capturing the dynamic association features between entities and extracting risk evolution patterns, generating composite risk feature vectors, realizing the feature coupling of static attributes and dynamic behaviors, and more comprehensively and deeply depicting risk features. Using an integrated model that combines a multi-layer perceptron and XGBoost for prediction, the multi-layer perceptron learns high-dimensional non-linear features, and XGBoost enhances the analysis of sparse features and time series trends, giving full play to the advantages of the two models, improving the generalization ability of the model, and enhancing the accuracy and reliability of prediction. Introducing an industry risk conduction intensity index, quantifying the risk conduction probability based on the industry node topology structure in the knowledge graph, and realizing cross-level prediction from micro-enterprise default to macro-industry risk. Generating risk prediction results based on multi-faceted risk indicators, providing a more comprehensive and accurate reference basis for financial leasing decisions. The overall solution solves the defects of traditional methods in three aspects: data dimension, entity association modeling, and risk dynamic evolution through the dynamic association mining of the knowledge graph and multi-modal data fusion; at the same time, the integrated model design takes into account both prediction accuracy and interpretability, providing panoramic warning support for financial leasing business from individual default to systemic risk. It improves the scientificity, accuracy and comprehensiveness of financial leasing risk prediction.

[0053] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structure specifically pointed out in this application document.

[0054] The following will further describe the technical solutions of the present invention in detail through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0056] Figure 1 is a flowchart of a method for predicting financial leasing risks based on a knowledge graph in an embodiment of the present invention;

[0057] Figure 2Schematic diagram of a financial leasing risk prediction system based on a knowledge graph in an embodiment of the present invention. Detailed implementation manners

[0058] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustrating and explaining the present invention, and are not used to limit the present invention.

[0059] Embodiment 1:

[0060] The present invention provides a financial leasing risk prediction method based on a knowledge graph. Referring to Figure 1 , including:

[0061] S1: Obtain structured data, unstructured data, and real-time transaction flow data from a financial leasing business system and form an initial knowledge unit set;

[0062] S2: Identify entities and semantic relationships from the initial knowledge unit set, perform entity alignment and knowledge fusion, and construct a dynamically updated financial leasing knowledge graph;

[0063] S3: Capture the dynamic association features between entities in the financial leasing knowledge graph, extract the temporal risk evolution patterns of historical transaction data, and generate a composite risk feature vector that combines static attributes and dynamic behaviors;

[0064] S4: Input the composite risk feature vector into an integrated model that combines a multi-layer perceptron and XGBoost, and output the lessee default probability, equipment asset depreciation rate, and industry risk transmission intensity;

[0065] S5: Generate a financial leasing risk prediction result for the lessee based on the lessee default probability, equipment asset depreciation rate, and industry risk transmission intensity;

[0066] It is also possible to update the entity relationships and risk weights in the knowledge graph using an incremental learning algorithm according to the newly added business data and early warning feedback results, and realize the online adaptive optimization of the risk prediction model.

[0067] In this embodiment, the financial leasing business system is an information system specifically used to process and manage financial leasing-related businesses, covering a series of processes and data from lease application, contract signing to asset tracking.

[0068] In this embodiment, structured data is data with a clear format and structure, such as lease transaction records and customer information in tabular form, which can be easily processed and analyzed by computer programs.

[0069] In this embodiment, unstructured data refers to data without a fixed format or structure, such as the text description of a lease contract and the feedback from customers, which is relatively complex to process and analyze.

[0070] In this embodiment, the real-time transaction flow data is continuously generated real-time transaction-related data along with the financial leasing business, reflecting the immediate status and changes of the business.

[0071] In this embodiment, the initial knowledge unit set is a set formed after preliminary processing and integration of structured, unstructured, and real-time transaction flow data obtained from the financial leasing business system.

[0072] In this embodiment, the dynamically updated knowledge graph of financial leasing is a graph that can be continuously adjusted and improved with the acquisition of new data and changes in the business, and is used to describe various entities and relationships in the field of financial leasing.

[0073] In this embodiment, the temporal risk evolution pattern of historical transaction data is the law and characteristics of the risk changing and developing over time summarized from past lease transaction data.

[0074] In this embodiment, the composite risk feature vector that fuses static attributes and dynamic behaviors is a vector formed by comprehensively combining the fixed attributes describing the financial leasing object and its behavioral characteristics changing over time, and is used to characterize the risk.

[0075] In this embodiment, the integrated model that combines a multi-layer perceptron and XGBoost is a model that combines two machine learning models, a multi-layer perceptron and XGBoost, for risk prediction. During the training process, adversarial training is used to enhance the model's robustness to noisy data, and feature contribution analysis is performed based on the Shapley value to optimize the model's interpretability. The Focal Loss function is used to solve the problem of imbalance between positive and negative samples in the financial leasing scenario; a meta-path constraint based on the knowledge graph is introduced to limit the logical rationality of the model on the industry risk conduction path.

[0076] In this embodiment, inputting the composite risk feature vector into the integrated model that combines a multi-layer perceptron and XGBoost and outputting the lessee default probability, equipment asset depreciation rate, and industry risk conduction intensity means putting the risk feature vector that fuses static attributes and dynamic behaviors into the model combined by a multi-layer perceptron and XGBoost, and then this model gives three results through calculation and analysis, namely the numerical value of the possibility of the lessee defaulting, the numerical value of the proportion of the value reduction of the leased equipment assets, and the numerical value of the intensity of the industry risk conduction to the financial leasing business.

[0077] In this embodiment, the lessee default probability refers to the likelihood that the lessee fails to fulfill its obligations according to the contract and defaults in the financial leasing business.

[0078] In this embodiment, the depreciation rate of equipment assets is the ratio of the value decline of leased equipment during use.

[0079] In this embodiment, the intensity of industry risk transmission is the strength and impact degree of the risks existing in the industry transmitted to the financial leasing business.

[0080] In this embodiment, the prediction result of the lessee's financial leasing risk is a prediction conclusion on the risks that the lessee may face in the financial leasing business based on various data and models.

[0081] The beneficial effects of the above technology are as follows: Obtain various data from the financial leasing business system to form an initial knowledge unit set, integrate structured data, unstructured data and real-time transaction flow data, break the information barrier, and provide a comprehensive data basis for risk prediction. Build a dynamically updated financial leasing knowledge graph, introduce entity alignment algorithms and knowledge fusion technologies, establish a multi-hop association network covering multiple parties, and dynamically update the graph through real-time data flow. Identifying implicit risk propagation paths can timely reflect business changes and new information, improving the timeliness and accuracy of risk assessment. Capture the dynamic association features between entities and extract risk evolution patterns, generate composite risk feature vectors, and realize the feature coupling of static attributes and dynamic behaviors, more comprehensively and deeply depicting risk features. Use an integrated model that combines a multi-layer perceptron and XGBoost for prediction. The multi-layer perceptron learns high-dimensional non-linear features, and XGBoost enhances the analysis of sparse features and time series trends, giving full play to the advantages of the two models, improving the generalization ability of the model, and enhancing the accuracy and reliability of the prediction. Introduce an industry risk transmission intensity indicator, quantify the risk transmission probability based on the industry node topology structure in the knowledge graph, and realize cross-level prediction from micro-enterprise default to macro-industry risk. Generate risk prediction results based on multi-faceted risk indicators, providing a more comprehensive and accurate reference basis for financial leasing decisions. The overall solution solves the defects of traditional methods in three levels: data dimension, entity association modeling, and risk dynamic evolution through the dynamic association mining of the knowledge graph and multi-modal data fusion; at the same time, the integrated model design takes into account both prediction accuracy and interpretability, providing panoramic early warning support for the financial leasing business from individual default to systemic risk. It improves the scientificity, accuracy and comprehensiveness of financial leasing risk prediction.

[0082] Embodiment 2:

[0083] Based on the method for predicting financial leasing risks based on a knowledge graph in Embodiment 1, S1: Obtain structured data, unstructured data and real-time transaction flow data from the financial leasing business system and form an initial knowledge unit set, including:

[0084] Obtain structured data, unstructured data and real-time transaction flow data from the financial leasing business system, including the credit record of the lessee, equipment asset status data, industry economic indicators, contract terms and historical default records;

[0085] The acquired structured data, unstructured data and real-time transaction flow data are cleaned, denoised and standardized, and entity, attribute and relationship triplets are extracted to generate an initial set of knowledge units.

[0086] In this embodiment, the lessee's credit record refers to relevant records about the lessee's credit status, including credit score, past loan repayment status and other information.

[0087] In this embodiment, the equipment asset status data is data describing various conditions of the leased equipment, such as the service life, degree of wear and tear, maintenance status, etc. of the equipment.

[0088] In this embodiment, the industry economic indicators are various data reflecting the economic operation status and development trend of the industry, such as industry growth rate, market saturation, etc.

[0089] In this embodiment, the contract clause text is a textual description of the specific clauses contained in the contract signed by both parties in the financial leasing business.

[0090] In this embodiment, the historical default record is the relevant record of the lessee's failure to perform the provisions of the financial lease contract in the past.

[0091] In this embodiment, cleaning, denoising and standardizing the acquired structured data, unstructured data and real-time transaction flow data refers to removing erroneous, duplicate or useless information from the various acquired data, reducing noise interference, and converting the data into a unified format and standard for subsequent analysis and processing.

[0092] In this embodiment, extracting entity, attribute and relationship triples is to identify specific objects (entities), their characteristics (attributes) and their mutual associations (relationships) from the data, and organize them into a set consisting of these three elements.

[0093] The beneficial effects of the above technical solutions are: obtaining a variety of comprehensive data, including the credit records of the lessee, etc., providing a rich source of information for risk prediction. Cleaning, denoising and standardizing the acquired data improves the quality and availability of the data. Extracting entity, attribute and relationship triples to generate an initial set of knowledge units lays a good foundation for the subsequent construction of the knowledge graph. It can ensure the accuracy and consistency of the data and reduce the impact of noise and errors on risk prediction. This solution provides a high-quality and standardized data foundation for financial leasing risk prediction, which helps to improve the accuracy and reliability of prediction.

[0094] Example 3:

[0095] Based on the method for predicting financial leasing risks based on a knowledge graph in Example 1, S2: Identify entities and semantic relationships from the initial knowledge unit set, perform entity alignment and knowledge fusion, and construct a dynamically updated financial leasing knowledge graph, including:

[0096] Define the core entity types and relationship types in the financial leasing field based on the ontology modeling method. The entity types include lessees, leased equipment, guarantors, industry classifications, and contract terms. The relationship types include guarantee associations, equipment mortgage statuses, and industry risk transmission paths;

[0097] Adopt a joint entity-relationship extraction model based on BERT to identify entities and semantic relationships from the initial knowledge unit set, calculate the entity semantic similarity using pre-trained domain word vectors, and aggregate neighborhood entity information combined with a graph attention network to generate entity alignment weights;

[0098] Through a cross-source conflict resolution algorithm, vote on the attribute conflicts of the same entity in multiple data sources, retain the attribute value with the highest confidence, and obtain a dynamically updated financial leasing knowledge graph.

[0099] In this embodiment, defining the core entity types and relationship types in the financial leasing field based on the ontology modeling method means using the ontology modeling method to clarify the key entity categories (such as lessees, leased equipment, etc.) within the scope of financial leasing and their associated categories (such as guarantee associations, equipment mortgage statuses, etc.).

[0100] In this embodiment, the industry risk transmission path refers to the ways and means by which risks existing in the industry spread and transfer among different entities and links. For example, in the automotive manufacturing industry, due to a significant increase in raw material prices (this is an industry risk), it will first affect component suppliers, who may increase component prices. This is the risk transmission from the raw material supply link to the component production link. Then, the cost of automobile manufacturers increases, and they may reduce production and increase automobile prices. This is the risk transmission to the automobile manufacturing link. Subsequently, consumers may reduce purchases due to the increase in automobile prices, resulting in a decline in automobile sales and affecting the profits of automobile dealers. This is the risk transmission to the sales link. The entire process from the increase in raw material prices to the impact on automobile sales is the path and method of risk transmission in the automotive manufacturing industry.

[0101] In this embodiment, adopting a joint entity-relationship extraction model based on BERT to identify entities and semantic relationships from the initial knowledge unit set means using a specific model based on BERT to distinguish specific entities (such as lessees, guarantors, etc.) and their semantic associations from the initially constructed knowledge unit set.

[0102] In this embodiment, the semantic similarity of entities is calculated using pre-trained domain word vectors, and the neighborhood entity information is aggregated by combining the graph attention network to generate entity alignment weights. This means that by using pre-trained domain-specific word vectors, the semantic similarity of entities is measured, and the graph attention network is used to integrate the information of surrounding entities, thereby obtaining the weight value of entity alignment. Suppose in the financial domain, there are two entities, "stock investment" and "securities investment". Through the pre-trained financial domain word vectors, it can be calculated that they have a relatively high semantic similarity. At the same time, in the knowledge graph, there are also neighborhood entities such as "stock trading" and "stock analysis" around the entity "stock investment". Using the graph attention network, it will pay more attention to the neighborhood entities closely related to "stock investment", such as "stock trading", and aggregate more of their information. Combining the aggregation results of semantic similarity and neighborhood entity information, the weight value of the alignment between the two entities "stock investment" and "securities investment" is finally obtained. For example, a higher weight value indicates a stronger association and alignment degree between them in a specific context.

[0103] In this embodiment, through the cross-source conflict resolution algorithm, a voting decision is made on the attribute conflicts of the same entity in multiple data sources, and the attribute value with the highest confidence is retained to obtain a dynamically updated knowledge graph of financial leasing. This means using an algorithm that can resolve conflicts from different data sources. For the conflicting attributes of the same entity in multiple data sources, a voting method is used to decide, and finally the most credible attribute value is retained, thereby obtaining a knowledge graph of financial leasing that can be continuously updated.

[0104] The beneficial effects of the above technical solutions are as follows: Defining the core entities and relationship types in the financial leasing domain provides a clear framework and structure for the construction of the knowledge graph. Adopting advanced joint entity relationship extraction models and technologies improves the accuracy and efficiency of entity and semantic relationship recognition. Using entity semantic similarity calculation and neighborhood entity information aggregation to achieve more accurate entity alignment. Solving attribute conflicts through the cross-source conflict resolution algorithm ensures the consistency and reliability of the data in the knowledge graph. Obtaining a dynamically updated knowledge graph can timely reflect business changes and provide the latest and most accurate information for risk prediction. This solution improves the scientificity, accuracy, and dynamic adaptability of the construction of the financial leasing knowledge graph, and helps to conduct risk prediction more effectively.

[0105] Embodiment 4:

[0106] Based on the method for predicting financial leasing risks based on the knowledge graph in Embodiment 1, S3: Capture the dynamic association features between entities in the financial leasing knowledge graph, extract the temporal risk evolution pattern of historical transaction data, and generate a composite risk feature vector that combines static attributes and dynamic behaviors, including:

[0107] Use the graph embedding algorithm to analyze the financial leasing knowledge graph and generate low-dimensional vector representations of entities and relationships;

[0108] Concatenate the entity node features in the financial leasing knowledge graph with the timestamp embedding vector, and input it into the gated graph convolutional layer to capture the time-varying association strength between entities as the dynamic association features between entities;

[0109] Extract the local temporal patterns of historical transaction data through the sliding time window mechanism, and perform cross-modal fusion with the global graph features to obtain the temporal risk evolution pattern of historical transaction data;

[0110] Based on the low-dimensional vector representations of entities and relationships, the dynamic association features between entities, and the temporal risk evolution pattern of historical transaction data, generate a composite risk feature vector that fuses static attributes and dynamic behaviors.

[0111] In this embodiment, the entity node features refer to various characteristics and attributes representing entities in the financial leasing knowledge graph. For example, in the financial leasing knowledge graph, "a certain company" is an entity node, and its features may include: the company's name, establishment time, registered capital, industry, credit rating, past lease transaction records, etc. Another example is the entity node "a certain equipment", and its features can be the equipment model, value, service life, maintenance status, etc.

[0112] In this embodiment, the timestamp embedding vector is to convert time information into vector form for easy model processing and analysis.

[0113] In this embodiment, the gated graph convolutional layer is a neural network layer that performs convolutional operations on graph-structured data and has a gating mechanism to control the flow and screening of information.

[0114] In this embodiment, the time-varying association strength between entities is a measure that describes the tightness or importance of the relationship between entities over time. Suppose in the financial leasing field, there are two entities, "lessee A" and "leased equipment B". In the first year, lessee A pays the rent on time, and the association strength with leased equipment B is relatively strong, indicating a close relationship between the two parties and normal equipment use. In the second year, lessee A begins to pay the rent overdue. At this time, the association strength with leased equipment B may become weaker, meaning that the relationship between the two parties becomes tense and there may be risks. After another half a year, lessee A defaults completely, and the association strength becomes very weak, indicating that the relationship between the two parties is on the verge of breaking. Through this measure of time-varying association strength, the dynamic changes in the relationship between these two entities can be clearly understood.

[0115] In this embodiment, the entity node features in the financial lease knowledge graph are concatenated with the timestamp embedding vector and input into the gated graph convolutional layer to capture the time-varying association strength between entities as the dynamic association features between entities. This means putting the vector composed of the features of the entities and the time information in the graph into the gated graph convolutional layer, so as to obtain the tightness of the relationship between entities changing over time and use it as the dynamic association feature.

[0116] In this embodiment, the sliding time window mechanism is a method of selecting data by moving according to a certain time interval.

[0117] In this embodiment, extracting the local temporal pattern of historical transaction data through the sliding time window mechanism means using this way of moving the time window to obtain the time series law within a specific time period from the historical transaction data.

[0118] In this embodiment, the local temporal pattern of historical transaction data is the regular time series feature presented by the historical transaction data within a local time range.

[0119] In this embodiment, the global graph feature is the overall feature of the entire financial lease knowledge graph. For example, in a financial lease knowledge graph, the global graph features may include: a large number of entities in the whole, indicating a large business scale; most of the connections between entities are relatively tight, reflecting relatively close cooperation relationships in the industry; the distribution of different types of entities (such as lessees, lessors, equipment suppliers, etc.) is relatively balanced, showing a relatively reasonable market structure; there are some core entities in the graph, which are connected to many other entities, reflecting the important status of these core entities in the entire financial lease business. These are all the overall features presented by the entire financial lease knowledge graph.

[0120] In this embodiment, extracting the local temporal pattern of historical transaction data through the sliding time window mechanism and performing cross-modal fusion with the global graph feature to obtain the temporal risk evolution pattern of historical transaction data means using the sliding time window to obtain the local time pattern and then fusing it with the features of the entire graph in different modes, so as to obtain the law of risk changing over time in the historical transaction data. Suppose there is a 5-year financial lease historical transaction data. Set a sliding time window of 1 year. Within the window of the first year, it is found that the lessees in a certain industry have a high willingness to pay when leasing equipment at the beginning of the year, but the willingness to pay decreases significantly at the end of the year. This is the local temporal pattern. At the same time, the global graph feature shows that the overall credit rating of this industry is relatively low and the industry competition is fierce. Fusing this local temporal pattern with the global graph feature, for example, it is found that the fierce industry competition leads to the operating difficulties of the lessees, thus affecting their willingness to pay for the leased equipment. This obtains the law of risk changing over time in the historical transaction data, that is, the lessees in this industry are more likely to default at the end of the year.

[0121] In this embodiment, based on the low-dimensional vector representation of entities and relationships, the dynamic association features between entities, and the temporal risk evolution pattern of historical transaction data, generating a composite risk feature vector that fuses static attributes and dynamic behaviors means constructing a risk feature vector that combines static attributes (such as the inherent characteristics of entities) and dynamic behaviors (such as changes in relationships and evolutions in time series) according to the above-mentioned various features and patterns.

[0122] The beneficial effects of the above technical solutions are as follows: Using the graph embedding algorithm to generate low-dimensional vector representations reduces the data dimension, improves the computational efficiency and data processing ability. Utilizing the gated graph convolutional layer to capture dynamic association features can more accurately capture the relationship strength that changes over time between entities. Extracting temporal patterns through the sliding time window mechanism and performing cross-modal fusion fully exploits the risk evolution information in historical transaction data. Generating a composite risk feature vector that fuses static attributes and dynamic behaviors makes the description of risk features more comprehensive and accurate. It helps to more deeply understand and analyze the risks in the financial leasing business, improving the accuracy and reliability of risk prediction. This solution enhances the effect of risk feature extraction and fusion, providing strong support for accurate risk prediction.

[0123] Embodiment 5:

[0124] Based on the financial leasing risk prediction method using a knowledge graph in Embodiment 1, S5: Generating the financial leasing risk prediction result for the lessee based on the lessee's default probability, the depreciation rate of equipment assets, and the intensity of industry risk transmission, including:

[0125] Obtaining the three-party defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the intensity of industry risk transmission;

[0126] Based on the three-party defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the intensity of industry risk transmission, as well as the lessee's default probability, the depreciation rate of equipment assets, and the intensity of industry risk transmission, calculating the financial leasing risk prediction value for the lessee;

[0127] When the financial leasing risk prediction value for the lessee does not exceed the preset prediction threshold, then taking the financial leasing risk prediction value as the financial leasing risk prediction result for the lessee;

[0128] When the financial leasing risk prediction value for the lessee exceeds the preset prediction threshold, then extracting the guarantee chain, the equipment mortgage status, and the upstream and downstream enterprises in the industry associated with the lessee from the financial leasing knowledge graph to generate a visualized risk propagation subgraph;

[0129] Calculate the risk influence weights of each node in the risk propagation sub-graph, label the key risk nodes and the conduction paths based on the risk influence weights of all nodes in the risk propagation sub-graph, form the key risk paths, and generate a risk tracing report in combination with the risk prediction value of the lessee's financial lease as the risk prediction result of the lessee's financial lease.

[0130] In this embodiment, obtaining the three-party defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the industry risk conduction intensity means obtaining the defined weights of the user for these three aspects of the lessee's default probability, the depreciation rate of equipment assets, and the industry risk conduction intensity. For example, the user defines the weight of the lessee's default probability as 0.4, the weight of the depreciation rate of equipment assets as 0.3, and the weight of the industry risk conduction intensity as 0.3.

[0131] In this embodiment, based on the three-party defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the industry risk conduction intensity, as well as the lessee's default probability, the depreciation rate of equipment assets, and the industry risk conduction intensity, the risk prediction value of the lessee's financial lease can be calculated using the weighted summation method. For example, the user defines the weight of the lessee's default probability as 0.4, the weight of the depreciation rate of equipment assets as 0.3, and the weight of the industry risk conduction intensity as 0.3;

[0132] Suppose the lessee's default probability is 0.2, the depreciation rate of equipment assets is 0.1, and the industry risk conduction intensity is 0.15;

[0133] Then the risk prediction value of the lessee's financial lease = 0.4×0.2 + 0.3×0.1 + 0.3×0.15 = 0.08 + 0.03 + 0.045 = 0.155.

[0134] In this embodiment, the preset prediction threshold is a numerical standard set in advance for judging the risk degree of the lessee's financial lease.

[0135] In this embodiment, the risk prediction value of the lessee's financial lease is a quantitative value obtained through model calculation, representing the risk faced by the lessee in the financial lease business.

[0136] In this embodiment, extracting the guarantee chain, equipment mortgage status, and upstream and downstream enterprises in the industry associated with the lessee from the financial lease knowledge graph to generate a visual risk propagation sub-graph means selecting information such as the guarantee relationship, equipment mortgage situation, and upstream and downstream enterprises in the industry related to the lessee from the overall financial lease knowledge graph, and displaying the possible risk propagation paths and scopes in a graphical manner.

[0137] In this embodiment, the risk propagation sub-graph is a local graph specifically extracted from the entire knowledge graph for risk propagation, used to more focusedly analyze the risk propagation situation.

[0138] In this embodiment, the risk influence weight of a node is a value that measures the magnitude of the risk propagation influence of each node in the risk propagation sub-graph.

[0139] In this embodiment, a key risk node is a node that plays an important role and has a greater influence on risk propagation in the risk propagation sub-graph.

[0140] In this embodiment, a conduction path is the specific route through which risk propagates between nodes.

[0141] In this embodiment, a key risk path is a path that plays a key role and has a greater influence on risk propagation among numerous conduction paths.

[0142] In this embodiment, generating a risk traceability report by combining the risk prediction value of the lessee's financial lease means integrating the risk prediction value of the lessee's financial lease and the analysis results of risk propagation to form a report for tracing the sources and causes of risks.

[0143] The beneficial effects of the above technical solutions are as follows: Obtaining the user-defined tripartite weights makes the risk prediction results more in line with the user's needs and key concerns. Calculating the risk prediction value of the lessee's financial lease provides a quantitative indicator for risk assessment. When the risk prediction value exceeds the threshold, a visual risk propagation sub-graph is generated to intuitively display the risk propagation situation. Calculating the risk influence weights and marking the key risk nodes and conduction paths helps to accurately identify the risk sources and key links. Generating a risk traceability report provides a detailed and clear basis for risk management and decision-making. This solution improves the personalization, visualization, and traceability of financial lease risk prediction, and helps to effectively respond to and manage risks.

[0144] Embodiment 6:

[0145] Based on the financial lease risk prediction method using a knowledge graph on the basis of Embodiment 5, calculating the risk influence weights of each node in the risk propagation sub-graph, and marking the key risk nodes and conduction paths based on the risk influence weights of all nodes in the risk propagation sub-graph to form a key risk path, including:

[0146] Determining the initial values of the risk influence weights of all nodes in the risk propagation sub-graph based on the total number of nodes in the risk propagation sub-graph;

[0147] Obtaining the risk correlation values and risk propagation intensities of adjacent nodes in the risk propagation sub-graph;

[0148] Iteratively calculate the initial values of the risk influence weights of all nodes in the risk propagation subgraph based on the risk correlation values and risk propagation intensities of adjacent nodes in the risk propagation subgraph and a preset iteration formula until the difference between the risk influence weights obtained by all nodes in the risk propagation subgraph after the latest iteration process and the risk influence weights obtained after the previous iteration process is less than a preset threshold. Then, in the risk propagation subgraph, label all nodes whose risk influence weights obtained after the latest iteration process are not less than the preset risk influence weight threshold as all key risk nodes;

[0149] Based on all the key risk nodes in the risk propagation subgraph, perform conduction path fitting to form a key risk path.

[0150] In this embodiment, determining the initial values of the risk influence weights of all nodes in the risk propagation subgraph based on the total number of nodes in the risk propagation subgraph means taking the reciprocal of the number of nodes in the risk propagation subgraph as the initial risk influence weight value of each node.

[0151] In this embodiment, the risk correlation values and risk propagation intensities of adjacent nodes in the risk propagation subgraph refer to the numerical values measuring the risk association degree between adjacent nodes in the risk propagation subgraph and the measure of the magnitude of risk propagation between these adjacent nodes.

[0152] In this embodiment, the preset threshold is a numerical boundary set in advance for judging or comparing the difference in the risk influence weights obtained by the same node after adjacent iteration processes.

[0153] In this embodiment, the preset risk influence weight threshold is a weight numerical boundary preset to judge the magnitude of the node risk influence.

[0154] The beneficial effects of the above technical solutions are as follows: By determining the initial values of the risk influence weights, a starting point is provided for subsequent iterative calculations. Obtaining the risk correlation values and propagation intensities of adjacent nodes provides important parameters for accurately calculating the risk influence weights. Using iterative calculations to continuously optimize the risk influence weights improves the accuracy and stability of weight calculations. By screening key risk nodes through the preset threshold and weight threshold, nodes that have an important impact on risk propagation can be accurately identified. Based on the key risk nodes, conduction path fitting is performed to construct a key risk path, which helps to clearly grasp the key links of risk propagation. This solution improves the accuracy and reliability of risk propagation analysis and provides strong support for the effective management of financial leasing risks.

[0155] Embodiment 7:

[0156] Based on the financial leasing risk prediction method using a knowledge graph in Embodiment 6, obtaining the risk correlation values and risk propagation intensities of adjacent nodes in the risk propagation subgraph includes:

[0157] Generate a risk-related feature vector for each node based on the business transaction data of each node in the risk propagation sub-graph;

[0158] Take the similarity between the risk-related feature vectors of adjacent nodes in the risk propagation sub-graph as the risk correlation value of the adjacent nodes;

[0159] Calculate the risk propagation intensity of adjacent nodes based on the total transaction amount and transaction frequency between adjacent nodes in the risk propagation sub-graph.

[0160] In this embodiment, the business transaction data of a node refers to various information and data related to specific business transactions associated with each node in the risk propagation sub-graph.

[0161] In this embodiment, generating a risk-related feature vector for each node based on the business transaction data of each node in the risk propagation sub-graph is achieved by processing and analyzing the business transaction data of each node and transforming it into a vector form that can characterize the risk characteristics of the node. Suppose there is a node "Enterprise A" in a risk propagation sub-graph. Its business transaction data includes: transaction amounts, transaction frequencies, and on-time performance of transactions with multiple suppliers in the past year. Process these data, such as calculating the average value, standard deviation of the transaction amount, change rate of the transaction frequency, and good rate of the performance. Then combine these processed values to form a vector, for example, [average transaction amount, standard deviation of transaction amount, change rate of transaction frequency, good rate of performance], which is the risk-related feature vector of the node "Enterprise A" and is used to characterize its characteristics in terms of risk.

[0162] In this embodiment, the risk-related feature vector of a node is a vector used to describe the characteristics of the node in terms of risk and can reflect the risk status of the node. For example, in a risk propagation sub-graph of financial leasing, there is a node representing "a certain manufacturing enterprise". Its risk-related feature vector may be [0.8, 0.3, 0.6, 0.1]. The first element 0.8 may indicate that the enterprise has a relatively high debt ratio, meaning a greater risk of debt repayment; the second element 0.3 may represent the proportion of the recent decline in its market share, reflecting the operating risk; the third element 0.6 may indicate that the cooperation stability with suppliers is average, presenting a supply chain risk; the fourth element 0.1 may represent a relatively small degree of adverse impact of recent policies on its industry. Such a vector can comprehensively describe the risk status of this node (i.e., the manufacturing enterprise).

[0163] In this embodiment, the similarity between the risk-related feature vectors of adjacent nodes is a measure of the similarity in characteristics between the risk-related feature vectors of two adjacent nodes and can be calculated through the cosine similarity or Euclidean distance between the vectors.

[0164] In this embodiment, the risk propagation intensity between adjacent nodes in the risk propagation subgraph is calculated based on the total transaction amount and transaction frequency between adjacent nodes, that is, the degree of risk propagation between these two adjacent nodes is determined according to the total transaction amount and the frequency of transactions between adjacent nodes, which is: the product of the quotient of the total transaction amount between two adjacent nodes and the average value of the total assets of the two adjacent nodes, and the transaction frequency between adjacent nodes is regarded as the risk propagation intensity between adjacent nodes.

[0165] The beneficial effects of the above technical solutions are as follows: Generate risk-related feature vectors, which provide a basis for quantifying risk correlation and propagation intensity. Using the similarity of the risk-related feature vectors of adjacent nodes as the risk correlation value can accurately reflect the degree of risk association between nodes. Calculating the risk propagation intensity through the transaction amount and transaction frequency objectively evaluates the possibility and influence of risk propagation from a business perspective. It can obtain the key information of adjacent nodes in the risk propagation subgraph more scientifically and accurately, providing strong support for subsequent risk analysis and assessment. This solution improves the ability to understand and evaluate the node relationship in the risk propagation subgraph, and helps to construct the key risk path more accurately.

[0166] Embodiment 8:

[0167] Based on the risk prediction method for financial leasing of the knowledge graph in Embodiment 6, a preset iterative formula is included:

[0168]

[0169] In the formula, RIW k+1 (i) is the risk influence weight of the i-th node in the risk propagation subgraph after the (k + 1)-th iteration, α is the risk propagation tendency coefficient, n is the total number of nodes in the risk propagation subgraph, Relevance(j, i) is the risk correlation value between the j-th node and the i-th node in the risk propagation subgraph, M(i) is the set of all nodes pointing to the i-th node in the risk propagation subgraph, RIW k (j) is the risk influence weight of the i-th node in the risk propagation subgraph after the k-th iteration, Strength(j, i) is the risk propagation intensity from the j-th node to the i-th node in the risk propagation subgraph, O(j) is the set of all nodes pointed to by the j-th node in the risk propagation subgraph, and Strength(j, k) is the risk propagation intensity from the j-th node to the k-th node in the risk propagation subgraph.

[0170] In this embodiment, the risk propagation tendency coefficient is a custom influence factor with a value range between 0 and 1, which is used to adjust the proportion of influence transmitted from other nodes, that is, to balance random risk propagation and risk propagation based on node relationships. It represents the probability of propagation depending on the actual relationships between nodes during the risk propagation process and is dimensionless.

[0171] The beneficial effects of the above technical solutions are as follows: A clear preset iteration formula is provided, which provides a standardized and quantifiable method for calculating the risk influence weight. Multiple factors such as the risk propagation tendency coefficient, the total number of nodes, the risk correlation value between nodes, and the risk propagation intensity are considered in the formula, making the calculation results more comprehensive and accurate. Through iterative calculation, it can dynamically reflect the change of the risk influence weight and adapt to the dynamics and complexity of risk propagation. It helps to more accurately determine the risk influence of nodes, so as to more accurately label key risk nodes and construct key risk paths. This solution improves the scientificity and accuracy of risk propagation analysis and provides a more reliable basis for financial leasing risk prediction and management.

[0172] Embodiment 9:

[0173] Based on the method for predicting financial leasing risks based on a knowledge graph in Embodiment 6, conduction path fitting is performed based on all key risk nodes in the risk propagation subgraph to form a key risk path, including:

[0174] Fitting at least one conduction path based on the edges between all key risk nodes in the risk propagation subgraph;

[0175] When there is only one conduction path, the only conduction path is regarded as the key risk path;

[0176] When there is more than one conduction path, the comprehensive conduction probability of each conduction path is calculated based on the risk correlation value and the risk propagation intensity between all adjacent key risk nodes in each conduction path, and the conduction path with the maximum comprehensive conduction probability among all conduction paths is regarded as the key risk path.

[0177] In this embodiment, fitting at least one conduction path based on the edges between all key risk nodes in the risk propagation subgraph means deriving at least one possible risk propagation route according to the connection relationships between key risk nodes.

[0178] In this embodiment, calculating the comprehensive conduction probability of each conduction path based on the risk correlation values and risk propagation intensities between all adjacent key risk nodes in each conduction path means comprehensively considering the degree of risk association and the strength of risk propagation between adjacent key risk nodes on each path, so as to obtain the comprehensive possibility value of risk propagation on each conduction path. For example, a preset weight can be used to perform weighted summation processing on the risk correlation values and risk propagation intensities between adjacent key risk nodes to obtain a parameter between adjacent nodes, and then the parameters of all groups of adjacent nodes on each conduction path are sequentially summed to obtain the comprehensive conduction probability of each conduction path.

[0179] In this embodiment, the comprehensive conduction probability of a conduction path is a value used to represent the overall possibility that risk can smoothly propagate on a specific conduction path.

[0180] The beneficial effects of the above technical solutions are as follows: It can comprehensively consider various possible conduction paths between key risk nodes. When there is only one conduction path, it is directly determined, simplifying the processing process. When there are multiple conduction paths, the path with the highest probability is selected as the key risk path by calculating the comprehensive conduction probability, improving the accuracy and rationality of path selection. It helps to more accurately determine the key path of risk propagation, providing a clear direction for risk management and control. This solution improves the scientificity and effectiveness of fitting the key risk path, enhancing the ability of financial lease risk prediction and management.

[0181] Embodiment 10:

[0182] The present invention provides a financial lease risk prediction system based on a knowledge graph, which is used to execute any one of the financial lease risk prediction methods based on a knowledge graph in Embodiments 1 to 9, refer to Figure 2 , including:

[0183] An initial knowledge unit construction module, which is used to obtain structured data, unstructured data, and real-time transaction flow data from a financial lease business system and form an initial knowledge unit set;

[0184] A knowledge graph construction module, which is used to identify entities and semantic relationships from the initial knowledge unit set, and perform entity alignment and knowledge fusion to construct a dynamically updated financial lease knowledge graph;

[0185] A composite risk feature vector generation module, which is used to capture the dynamic association features between entities in the financial lease knowledge graph, extract the temporal risk evolution pattern of historical transaction data, and generate a composite risk feature vector that integrates static attributes and dynamic behaviors;

[0186] A model prediction and evaluation module, which is used to input the composite risk feature vector into an integrated model that combines a multi-layer perceptron and XGBoost, and outputs the default probability of the lessee, the depreciation rate of equipment assets, and the industry risk transmission intensity;

[0187] A risk prediction module, which is used to generate a risk prediction result for the lessee's financial leasing based on the default probability of the lessee, the depreciation rate of equipment assets, and the industry risk transmission intensity.

[0188] The beneficial effects of the above technologies are as follows: Obtain various data from the financial leasing business system to form an initial knowledge unit set, integrate structured data, unstructured data, and real-time transaction flow data, break the information barrier, and provide a comprehensive data basis for risk prediction. Build a dynamically updated knowledge graph for financial leasing, introduce entity alignment algorithms and knowledge fusion technologies, establish a multi-hop association network covering multiple parties, and dynamically update the graph through real-time data flow. Identify hidden risk propagation paths, which can timely reflect business changes and new information, and improve the timeliness and accuracy of risk assessment. Capture the dynamic association features between entities and extract risk evolution patterns, generate composite risk feature vectors, and achieve the feature coupling of static attributes and dynamic behaviors, so as to more comprehensively and deeply depict risk characteristics. Use an integrated model that combines a multi-layer perceptron and XGBoost for prediction. The multi-layer perceptron learns high-dimensional non-linear features, and XGBoost enhances the analysis of sparse features and time series trends, giving full play to the advantages of the two models, improving the generalization ability of the model, and enhancing the accuracy and reliability of prediction. Introduce an industry risk transmission intensity indicator, quantify the risk transmission probability based on the industry node topology structure in the knowledge graph, and achieve cross-level prediction from micro-enterprise default to macro-industry risk. Generate risk prediction results based on various risk indicators, providing a more comprehensive and accurate reference basis for financial leasing decisions. The overall solution solves the defects of traditional methods in three aspects: data dimension, entity association modeling, and risk dynamic evolution through the dynamic association mining of the knowledge graph and multi-modal data fusion; at the same time, the integrated model design takes into account both prediction accuracy and interpretability, providing panoramic warning support for financial leasing business from individual default to systemic risk. It improves the scientificity, accuracy, and comprehensiveness of financial leasing risk prediction.

[0189] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A risk prediction method for financial leasing based on a knowledge graph, characterized in that, Including: S1: Obtain structured data, unstructured data, and real-time transaction flow data from the financial leasing business system and form an initial knowledge unit set; S2: Identify entities and semantic relationships from the initial knowledge unit set, perform entity alignment and knowledge fusion, and construct a dynamically updated financial leasing knowledge graph; S3: Capture the dynamic association features between entities in the financial leasing knowledge graph, extract the temporal risk evolution patterns of historical transaction data, and generate a composite risk feature vector that integrates static attributes and dynamic behaviors; S4: Input the composite risk feature vector into an integrated model that combines a multi-layer perceptron and XGBoost, and output the default probability of the lessee, the depreciation rate of equipment assets, and the intensity of industry risk transmission; S5: Generate the financial leasing risk prediction result of the lessee based on the default probability of the lessee, the depreciation rate of equipment assets, and the intensity of industry risk transmission.

2. The method for predicting financing lease risks based on a knowledge graph according to claim 1, wherein S1: Obtain structured data, unstructured data, and real-time transaction flow data from the financial leasing business system and form an initial knowledge unit set, including: Obtain structured data, unstructured data, and real-time transaction flow data from the financial leasing business system, including the credit records of the lessee, the status data of equipment assets, industry economic indicators, contract clause texts, and historical default records; Clean, denoise, and standardize the obtained structured data, unstructured data, and real-time transaction flow data, and extract entity, attribute, and relationship triples to generate an initial knowledge unit set.

3. The method for predicting financial leasing risks based on a knowledge graph according to claim 1, wherein S2: Identify entities and semantic relationships from the initial knowledge unit set, perform entity alignment and knowledge fusion, and construct a dynamically updated financial leasing knowledge graph, including: Define the core entity types and relationship types in the financial leasing domain based on the ontology modeling method. The entity types include lessees, leased equipment, guarantors, industry classifications, and contract clauses, and the relationship types include guarantee associations, equipment mortgage status, and industry risk transmission paths; Adopt a joint entity relationship extraction model based on BERT to identify entities and semantic relationships from the initial knowledge unit set, calculate the entity semantic similarity using pre-trained domain word vectors, and aggregate neighborhood entity information combined with a graph attention network to generate entity alignment weights; Through a cross-source conflict resolution algorithm, vote on the attribute conflicts of the same entity in multiple data sources, retain the attribute value with the highest confidence, and obtain a dynamically updated financial leasing knowledge graph.

4. The method for predicting financial leasing risks based on a knowledge graph according to claim 1, wherein S3: Capture the dynamic association features between entities in the financial leasing knowledge graph, extract the temporal risk evolution patterns of historical transaction data, and generate a composite risk feature vector that integrates static attributes and dynamic behaviors, including: Adopt a graph embedding algorithm to analyze the financial leasing knowledge graph to generate low-dimensional vector representations of entities and relationships; Concatenate the entity node features in the financial leasing knowledge graph with the timestamp embedding vector, input it into the gated graph convolutional layer, and capture the time-varying association strength between entities as the dynamic association feature between entities; Extract the local temporal patterns of historical transaction data through a sliding time window mechanism and perform cross-modal fusion with the global graph features to obtain the temporal risk evolution patterns of historical transaction data; Generate a composite risk feature vector that fuses static attributes and dynamic behaviors based on the low-dimensional vector representation of entities and relationships, the dynamic association features between entities, and the temporal risk evolution pattern of historical transaction data.

5. The method for predicting financial leasing risks based on a knowledge graph according to claim 1, wherein S5: Generate the risk prediction result of the lessee's financial leasing based on the lessee's default probability, the depreciation rate of equipment assets, and the industry risk transmission intensity, including: Obtain the tripartite defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the industry risk transmission intensity; Calculate the risk prediction value of the lessee's financial leasing based on the tripartite defined weights of the user for the lessee's default probability, the depreciation rate of equipment assets, and the industry risk transmission intensity, as well as the lessee's default probability, the depreciation rate of equipment assets, and the industry risk transmission intensity; When the risk prediction value of the lessee's financial leasing does not exceed the preset prediction threshold, then regard the risk prediction value of the lessee's financial leasing as the risk prediction result of the lessee's financial leasing; When the risk prediction value of the lessee's financial leasing exceeds the preset prediction threshold, then extract the guarantee chain, equipment mortgage status, and upstream and downstream enterprises in the industry associated with the lessee from the financial leasing knowledge graph to generate a visual risk propagation subgraph; Calculate the risk influence weight of each node in the risk propagation subgraph, label the key risk nodes and the conduction path based on the risk influence weights of all nodes in the risk propagation subgraph to form a key risk path, and generate a risk traceability report in combination with the risk prediction value of the lessee's financial leasing as the risk prediction result of the lessee's financial leasing.

6. The method for predicting financial leasing risks based on a knowledge graph according to claim 5, wherein Calculate the risk influence weight of each node in the risk propagation subgraph, label the key risk nodes and the conduction path based on the risk influence weights of all nodes in the risk propagation subgraph to form a key risk path, including: Determine the initial value of the risk influence weight of all nodes in the risk propagation subgraph based on the total number of nodes in the risk propagation subgraph; Obtain the risk correlation value and risk transmission intensity of adjacent nodes in the risk propagation subgraph; Iteratively calculate the initial value of the risk influence weight of all nodes in the risk propagation subgraph based on the risk correlation value and risk transmission intensity of adjacent nodes in the risk propagation subgraph and the preset iteration formula until the difference between the risk influence weight obtained by all nodes in the risk propagation subgraph after the latest iteration process and the risk influence weight obtained after the previous iteration process is less than the preset threshold, then label all nodes in the risk propagation subgraph whose risk influence weight obtained after the latest iteration process is not less than the preset risk influence weight threshold as all key risk nodes; Conduct conduction path fitting based on all key risk nodes in the risk propagation subgraph to form a key risk path.

7. The method for predicting financial leasing risks based on a knowledge graph according to claim 6, wherein Obtain the risk correlation value and risk transmission intensity of adjacent nodes in the risk propagation subgraph, including: Generate a risk-related feature vector for each node based on the business transaction data of each node in the risk propagation subgraph; Regard the similarity between the risk-related feature vectors of adjacent nodes in the risk propagation subgraph as the risk correlation value of adjacent nodes; Calculate the risk transmission intensity of adjacent nodes based on the total transaction amount and transaction frequency between adjacent nodes in the risk propagation subgraph.

8. The method for predicting financing lease risks based on a knowledge graph according to claim 6, wherein, Preset iteration formula, including: where RIW k+1 (i) is the risk influence weight of the i-th node in the risk propagation subgraph after the (k + 1)-th iteration, α is the risk propagation tendency coefficient, n is the total number of nodes in the risk propagation subgraph, Relevance(j, i) is the risk correlation value between the j-th node and the i-th node in the risk propagation subgraph, M(i) is the set of all nodes pointing to the i-th node in the risk propagation subgraph, RIW k (j) is the risk influence weight of the i-th node in the risk propagation subgraph after the k-th iteration, Strength(j, i) is the risk propagation strength from the j-th node to the i-th node in the risk propagation subgraph, O(j) is the set of all nodes pointed to by the j-th node in the risk propagation subgraph, Strength(j, k) is the risk propagation strength from the j-th node to the k-th node in the risk propagation subgraph.

9. The method for predicting financial leasing risks based on a knowledge graph according to claim 6, characterized in that, Conduct path fitting is performed based on all key risk nodes in the risk propagation subgraph to form a key risk path, including: Fitting at least one conduction path based on the edges between all key risk nodes in the risk propagation subgraph; When there is only one conduction path, the only conduction path is regarded as the key risk path; When there is more than one conduction path, the comprehensive conduction probability of each conduction path is calculated based on the risk correlation value and risk propagation intensity between all adjacent key risk nodes in each conduction path, and the conduction path with the maximum comprehensive conduction probability among all conduction paths is regarded as the key risk path.

10. A financial leasing risk prediction system based on a knowledge graph, characterized in that, Used to execute the knowledge graph-based financial leasing risk prediction method described in any one of claims 1 to 9, including: An initial knowledge unit construction module for obtaining structured data, unstructured data, and real-time transaction flow data from the financial leasing business system and forming an initial knowledge unit set; A knowledge graph construction module for identifying entities and semantic relationships from the initial knowledge unit set, performing entity alignment and knowledge fusion, and constructing a dynamically updated financial leasing knowledge graph; A composite risk feature vector generation module for capturing the dynamic association features between entities in the financial leasing knowledge graph, extracting the temporal risk evolution pattern of historical transaction data, and generating a composite risk feature vector that fuses static attributes and dynamic behaviors; A model prediction and evaluation module for inputting the composite risk feature vector into an integrated model that fuses a multi-layer perceptron and XGBoost, and outputting the lessee's default probability, equipment asset depreciation rate, and industry risk conduction intensity; A risk prediction module for generating a financial leasing risk prediction result for the lessee based on the lessee's default probability, equipment asset depreciation rate, and industry risk conduction intensity.

Images