Financial risk intelligent assessment method and system based on multi-mode dynamic graph combined efficiency improvement

Abstract

The invention relates to the technical field of financial risk assessment, in particular to an intelligent financial risk assessment method and system based on multi-mode dynamic graph combined efficiency improvement, and the method comprises the steps: mining the implicit correlation characteristics of an enterprise / individual in a complex financial network through the combination of multi-mode data fusion and a dynamic knowledge graph; and precise evaluation and real-time early warning are realized in combination with time sequence analysis and an external event quantitative model. According to the method, implicit associated feature information of enterprises / individuals in a complex financial network is mined by adopting a mode of combining multi-modal data fusion and a dynamic knowledge graph, and accurate assessment and real-time early warning of financial risks are realized by combining time sequence analysis and an external event influence quantitative model, so that the risk assessment efficiency is improved. The financial risk assessment efficiency and accuracy can be improved, and supervision compliance verification is supported.

Description

Technical Field

[0001] This invention relates to the fields of financial risk assessment and artificial intelligence technology, and in particular to a method and system for intelligent financial risk assessment that improves efficiency through multi-modal dynamic graphs. Background Technology

[0002] In the financial sector, risk assessment is a crucial step in ensuring the sound operation of financial institutions and preventing systemic risks. Traditional financial risk assessment methods primarily rely on static analysis of corporate financial data, such as the calculation and comparison of indicators like the debt-to-equity ratio and current ratio. However, this approach has significant limitations: First, traditional methods are often limited to single structured data sources, neglecting the risk information contained in unstructured data (such as contract terms, public opinion information, and audio recordings), leading to prominent data silos. Second, existing technologies mostly employ static modeling methods, making it difficult to capture complex dynamic network effects between enterprises, such as potential risk propagation paths in equity relationships, transactional relationships, and guarantee chains. Finally, traditional assessment processes lack transparency; the input data, intermediate calculations, and final results of the assessment model are difficult for regulatory agencies to effectively trace and verify, leading to a lack of trust and making it difficult to meet increasingly stringent regulatory compliance requirements.

[0003] With the development of big data, artificial intelligence, and blockchain technologies, the field of financial risk assessment has ushered in new technological opportunities. However, existing technologies still have significant shortcomings in integrating multi-source heterogeneous data, mining dynamic risk characteristics, and achieving transparency in the assessment process. For example, the semantic analysis capabilities of unstructured data are limited, making it difficult to efficiently extract key risk indicators; the technology for constructing and updating dynamic knowledge graphs in real time is still immature, making it difficult to quantify the risk propagation effects in complex networks; at the same time, the evidence storage and auditing mechanisms for assessment results lack technical support, failing to provide regulatory agencies with credible verification methods. Therefore, developing an intelligent assessment method and system that can combine multimodal data, dynamically capture risk propagation paths, and achieve full-process transparency has become an urgent technical challenge to be solved in the field of financial risk management. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a multi-modal dynamic graph-based intelligent financial risk assessment method and system. It employs a combination of multi-modal data fusion and dynamic knowledge graphs to uncover implicit correlation characteristics of enterprises / individuals within complex financial networks. This is combined with time-series analysis and a quantitative model of the impact of external events to achieve accurate assessment and real-time early warning of financial risks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A multi-modal dynamic graph-based intelligent financial risk assessment system, comprising: a data acquisition and preprocessing module for collecting data from multiple sources, identifying and semantically analyzing unstructured data, extracting key risk indicators, and desensitizing sensitive data using homomorphic encryption technology; a dynamic knowledge graph construction module for building a real-time updated knowledge graph with enterprises or individuals as nodes and relationships as edges, and converting the graph structure into a vector representation through a graph embedding algorithm; a risk feature extraction module for extracting static features, dynamic features, spatial features, and external event features; a risk scoring and early warning module for inputting features into an XGBoost-LightGBM hybrid model to output a risk score and setting dynamic thresholds; and a blockchain evidence storage and auditing module for putting key data from the assessment process on the blockchain and supporting regulatory agencies to verify the compliance of the assessment process through smart contracts.

[0006] In the aforementioned intelligent financial risk assessment system that enhances efficiency through multi-modal dynamic graph collaboration, the dynamic knowledge graph construction module is used to: update the node and edge weights in the knowledge graph in real time using an incremental learning algorithm, wherein the node weight update formula is:

[0007] in, This represents the weight of node i at time t. This is a smoothing coefficient used to control the degree of influence of historical weights. Let i represent the set of neighboring nodes. The edge weight update formula is: (This represents the degree of neighbor node j; the formula is:)

[0008] in, This represents the weight of edge (i,j) at time t. The attenuation coefficient is... Based on the distance between nodes Nonlinear functions (such as the Gaussian kernel function) are used to quantify the tightness of the relationship between nodes.

[0009] In the aforementioned intelligent financial risk assessment system that combines multi-modal animation for enhanced efficiency, the risk feature extraction module is used to: capture risk propagation paths using the ST-GAT model, the core formula of which is:

[0010] in, Represents a node In the Layer feature representation, For activation functions (such as ReLU). The attention weight is calculated using the following formula:

[0011] in, The exponential function maps input values ​​to the range of positive real numbers, ensuring that attention weights are positive; LeakyReLU is the activation function. This is a learnable attention vector used to calculate the attention score. It is the transpose of the vector; W is the weight matrix, used to perform linear transformation on the node features; Indicates a splicing operation; and These are the feature vectors of node i and node j, respectively.

[0012] In the aforementioned multi-modal animated graph-based intelligent financial risk assessment system, the risk scoring and early warning module is used to: perform risk scoring using the XGBoost-LightGBM hybrid model, with the prediction formula as follows:

[0013] in, These are the predicted values ​​from the mixture model; Let be the prediction value of the k-th tree in XGBoost. This is the predicted value for LightGBM. The balancing coefficient is used to adjust the contribution ratio of the two models; the dynamic threshold calculation formula is:

[0014] in, A dynamic threshold is used for risk warning; The historical average rating Standard deviation, It is a regulating factor.

[0015] The aforementioned intelligent financial risk assessment system, which enhances efficiency through multi-modal dynamic graph collaboration, also includes a training module for optimizing the dynamic knowledge graph construction module and the risk scoring and early warning module.

[0016] In the aforementioned intelligent financial risk assessment system that enhances efficiency through multi-modal animated graphs, the training module includes: a training data generation module, used to generate large-scale labeled data through a simulation environment to simulate risk propagation paths under different scenarios; and a graph structure optimization module, used to optimize the topology of the knowledge graph through graph regularization constraints, the objective function of which is:

[0017] in, Let E represent the graph regularization loss function, which measures how well the graph structure is preserved in the embedding space; E is the set of edges. Let i be the vector representation of node i; For nodes The vector representation of ; V is the set of nodes; The regularization coefficient is used. is the Euclidean norm of a vector, used to calculate the length of a vector or the difference between two vectors. Specifically, express and The square of the difference between them, express The square of the length.

[0018] The model fine-tuning module is used to adapt the XGBoost-LightGBM hybrid model to a specific domain through transfer learning. Its loss function is:

[0019] in, For mission losses, Loss due to domain differences These are the transfer learning weights.

[0020] According to another aspect of this application, a multi-modal dynamic graph-based intelligent financial risk assessment method is provided, which includes: collecting data from multiple sources and identifying and semantically analyzing unstructured data; extracting key risk indicators and desensitizing sensitive data using homomorphic encryption technology; constructing a real-time updated knowledge graph with enterprises / individuals as nodes and relationships as edges, and converting it into vector representations through graph embedding algorithms; extracting static features, dynamic features, spatial features, and external event features; inputting features into an XGBoost-LightGBM hybrid model, outputting a risk score, and setting dynamic thresholds; and putting key data of the assessment process on the blockchain to support regulatory agencies in verifying the compliance of the assessment process through smart contracts.

[0021] According to another aspect of this application, an electronic device is provided, comprising: a processor; and a memory storing computer program instructions, which, when executed by the processor, cause the processor to perform the financial risk intelligent assessment method described above for multi-modal animation joint efficiency improvement.

[0022] According to another aspect of this application, a computer-readable medium is provided having computer program instructions stored thereon, which, when executed by a processor, cause the processor to perform the financial risk intelligent assessment method described above for multi-modal animation joint efficiency improvement.

[0023] Compared with the prior art, the beneficial effects of the present invention are: This invention employs a combination of multimodal data fusion and dynamic knowledge graphs to uncover implicit correlation characteristics of enterprises / individuals within complex financial networks. By combining time-series analysis and quantitative models of external event impacts, it aims to achieve accurate assessment and real-time early warning of financial risks. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the module structure of a financial risk intelligent assessment system that improves efficiency through multi-modal animation, as provided in an embodiment of this application.

[0025] Figure 2 This is a schematic diagram of the node and edge weight update process of the dynamic knowledge graph construction module in this application embodiment.

[0026] Figure 3 This is a schematic diagram illustrating the working principle of the risk feature extraction module in the embodiments of this application.

[0027] Figure 4 This is a schematic diagram of the risk score calculation and dynamic threshold adjustment process of the risk score and early warning module in this application embodiment. Detailed Implementation

[0028] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0029] This application provides a multi-modal dynamic graph-based intelligent assessment method and system for financial risk. Its core lies in mining the implicit correlation characteristics of enterprises or individuals within complex financial networks through a combination of multi-modal data fusion and dynamic knowledge graphs. This is combined with time-series analysis, spatial propagation path modeling, and quantitative models of external event impacts to achieve accurate assessment and real-time early warning of financial risks. The following will be illustrated with the appendix... Figure 1 To be continued Figure 4 The specific implementation methods of this application are described in detail.

[0030] First, describe the overall architecture of the system and its operating principle. (See attached image) Figure 1As shown, this system includes a data acquisition and preprocessing module, a dynamic knowledge graph construction module, a risk feature extraction module, a risk scoring and early warning module, and a blockchain-based evidence storage and auditing module. These modules work together to complete the entire process from data acquisition to risk assessment.

[0031] The data acquisition and preprocessing module is responsible for acquiring various types of data from financial institutions, government open data platforms, and third-party data sources, including structured data (such as financial statements) and unstructured data (such as PDF contracts and audio call logs). For unstructured data, natural language processing techniques are used for semantic analysis, such as using the DocLLM model to parse text content and extract key risk indicators (such as default clauses and repayment commitments). To protect sensitive data, the system uses homomorphic encryption technology to anonymize the raw data, ensuring data security in subsequent processing stages.

[0032] The dynamic knowledge graph construction module is one of the core components of the system, and its operating principle is shown in the appendix. Figure 2 As shown, the dynamic knowledge graph construction module is used to build a real-time updated knowledge graph with enterprises / individuals as nodes and relationships (equity, transactions, guarantees) as edges. It transforms the graph structure into a vector representation using graph embedding algorithms (such as Node2Vec), and then quantifies node importance using the Temporal PageRank algorithm. Specifically, the Node2Vec algorithm transforms the graph structure into a series of vector representations of nodes. After obtaining the node vector representations, the system uses the Temporal PageRank algorithm, which combines time decay factors, incremental calculations, and graph embedding techniques, to dynamically adjust the node and edge weights of the financial network. This quantifies the real-time importance of nodes in the complex network, providing crucial evidence for risk scoring and early warning.

[0033] To capture the dynamic changes of nodes and edges over time, the system introduces an incremental learning algorithm to update the weights in the knowledge graph in real time. Specifically, the update formula for node weights is as follows: ,in, Let i be the weight of node i at time t; This is a smoothing coefficient used to control the degree of influence of historical weights; Let i be the set of neighboring nodes of node i; Let be the degree of neighbor node j; Let be the weight of node j, a neighbor of node i, at time t. The formula for updating edge weights is: ,in This represents the weight of edge (i,j) at time t. The attenuation coefficient is used to reduce the influence of the old weights. Based on the distance between nodes Nonlinear functions (such as the Gaussian kernel function) are used to quantify the tightness of relationships between nodes. Using the above formula, the system can dynamically adjust the topology of the knowledge graph, thereby more accurately reflecting real-time changes in the financial network.

[0034] The working principle of the risk feature extraction module is shown in the attached figure. Figure 3 As shown, the risk feature extraction module is used to extract static features (financial ratios, historical default records), dynamic features (using LSTM neural networks to analyze corporate revenue fluctuation trends), spatial features (capturing risk propagation paths in the knowledge graph through the ST-GAT model), and external event features (combining news sentiment analysis and policy text matching to quantify the intensity of policy impact). Static features include financial ratios (such as debt-to-equity ratio and current ratio) and historical default records. Financial ratios such as debt-to-equity ratio and current ratio reflect the financial health of a company; historical default records are directly related to the company's credit risk. The core formula of the ST-GAT model is... , in Indicates that node i is at the 1st rank. Layer feature representation, For activation functions (such as ReLU). Let i be the set of neighboring nodes of node i; For the first Layer weight matrix; Let be the attention weight of node i to node j, used to measure the influence of node j on node i, and its calculation formula is: ,in, The exponential function maps input values ​​to the range of positive real numbers, ensuring that attention weights are positive; LeakyReLU is the activation function. This is a learnable attention vector used to calculate the attention score. It is the transpose of the vector; W is the weight matrix, used to perform linear transformation on the node features; Indicates a splicing operation; and These are the feature vectors of node i and node j, respectively.

[0035] Using the above formula, the system can dynamically allocate attention weights, thereby more accurately capturing the key paths of risk transmission.

[0036] The operation flow of the risk scoring and early warning module is as follows: Figure 4 As shown, its core is risk scoring using the XGBoost-LightGBM hybrid model. The prediction formula for the hybrid model is... ,in, These are the predicted values ​​from the mixture model; Let be the prediction value of the k-th tree in XGBoost. This is the predicted value for LightGBM. This is a balancing coefficient used to adjust the contribution ratio of the two models. By combining the advantages of both models, the system can improve the accuracy of risk scoring. Furthermore, the system also sets a dynamic threshold to automatically adjust the high-risk warning line; its calculation formula is as follows: ,in, The historical average rating Standard deviation, This is an adjustment factor used to control the sensitivity of the threshold. By dynamically adjusting the threshold, the system can flexibly respond to changes in risk under different scenarios.

[0037] To further optimize the dynamic knowledge graph construction module and the risk scoring and early warning module, the system also includes a training module. The training module consists of a training data generation module, a graph structure optimization module, and a model fine-tuning module.

[0038] (1) The training data generation module generates large-scale labeled data through a simulation environment to simulate risk propagation paths in different scenarios, specifically including the following steps: Step 1: Define the financial network structure: Node design: Node types: enterprises, financial institutions, individual investors, regulatory agencies, etc.

[0039] Node attributes: financial indicators (debt-to-equity ratio, cash flow, etc.), industry classification, credit rating, and historical default records.

[0040] Edge design: Edge types: equity relationship, lending relationship, guarantee relationship, supply relationship, and transaction.

[0041] Edge weights: based on quantitative indicators such as transaction amount, shareholding ratio, and collateral amount.

[0042] Dynamics: Edge weights decay over time (e.g., the lending relationship expires) or strengthen (e.g., frequent transactions increase the association strength).

[0043] Network generation: Using random graph models (such as the Barabási-Albert model), heterogeneous networks with power-law distributions are generated to simulate the characteristics of a few core nodes connecting a large number of edge nodes in a real financial system.

[0044] Step 2, establish risk propagation rules: Risk triggering conditions: 1. Node default: When the enterprise's node debt ratio exceeds the threshold (e.g., 80%) or cash flow is negative for three consecutive months, a default event is triggered. 2. External shocks: Simulate the impact of events such as policy tightening (e.g., a 2% increase in loan interest rates) and market crashes (e.g., a 30% drop in stock prices) on node attributes.

[0045] Propagation Mechanisms: 1. Direct Transmission: Defaulting nodes transmit risk to related financial institutions through lending edges. Loss amount = edge weight × default probability. 2. Indirect Transmission: Propagation occurs through supply chain edges. For example, default by a core enterprise leads to a decrease in supplier revenue, triggering supplier defaults. 3. Threshold Propagation: Nodes only further propagate risk when the total amount of risk received exceeds their own capacity (e.g., 50% of their capital buffer).

[0046] Dynamic adjustment: The rate of risk transmission decays over time (e.g., daily decay coefficient). =0.9).

[0047] Step 3: Simulate different scenarios: Scenario 1: Single node default Trigger a default by a large enterprise and observe how the risk is transmitted to related banks through the guarantee chain, leading to a decrease in the banks' capital adequacy ratio. Record whether the banks trigger regulatory red lines (e.g., capital adequacy ratio <8%).

[0048] Scenario 2: Systemic Risk in the Industry The simulation demonstrates how tightening real estate policies (such as stricter purchase restrictions) could lead to defaults by multiple real estate companies, with the risk spreading through the supply chain to industries such as building materials and home appliances. This tests the model's ability to capture cross-industry risk propagation.

[0049] Scenario 3: Market Sentiment Contagion By simulating negative public opinion (such as "a certain bank is facing a liquidity crisis") through news sentiment analysis, a bank run is simulated, leading to investor panic and triggering bank liquidity risk. The impact of public opinion intensity on the risk level at each node is quantified (e.g., for every 10-point decrease in public opinion score, the probability of a bank run increases by 5%).

[0050] Step 4: Generate labeled data: Based on the simulation results, large-scale labeled data is generated. The labeled data should include key indicators such as the structural information of the financial network, node attributes, edge attributes, risk propagation paths, and the risk level of each node. This data can be used to subsequently train risk assessment models, improving their generalization ability and accuracy.

[0051] Step 5, Verification and Optimization: The risk assessment model was trained using the generated labeled data, and its performance in real-world scenarios was validated. Based on the validation results, the simulation environment and the risk assessment model were optimized and adjusted to improve their accuracy and reliability.

[0052] (2) The graph structure optimization module optimizes the topology of the knowledge graph through graph regularization constraints. Its objective function is: ,in, Let E represent the graph regularization loss function, which measures how well the graph structure is preserved in the embedding space; E is the set of edges. Let i be the vector representation of node i; For nodes The vector representation of ; V is the set of nodes; The regularization coefficient is used. is the Euclidean norm of a vector, used to calculate the length of a vector or the difference between two vectors. Specifically, express and The square of the difference between them, express The square of the length.

[0053] (3) The model fine-tuning module adapts the XGBoost-LightGBM hybrid model to the domain through transfer learning, and its loss function is: ,in, For mission losses, Loss due to domain differences These are the transfer learning weights, used to balance task loss and domain difference loss. Through the above optimization measures, the system can significantly improve the model's generalization ability and evaluation accuracy.

[0054] The blockchain-based evidence storage and auditing module leverages the immutability and traceability of blockchain to ensure the secure storage and compliant verification of critical data during the assessment process. Specific functions include: data on-chain recording and smart contract verification.

[0055] In this process, data on-chaining involves the system encrypting key data such as the original data hash value, model parameters, and scoring results during the evaluation process and then storing them on the blockchain. As crucial evidence of the evaluation process, the on-chain data cannot be tampered with once written, thus ensuring the authenticity and integrity of the data.

[0056] Smart contract verification enables regulatory agencies to automatically verify the compliance of assessment processes through smart contracts. Smart contracts embed assessment rules and standards into the blockchain in the form of code. When it is necessary to verify the results of a company's risk assessment, regulatory agencies can query on-chain data by calling the contract, and the system can automatically verify whether the assessment process complies with preset rules (such as model parameter thresholds and consistency of scoring logic).

[0057] In summary, this application achieves accurate assessment and real-time early warning of financial risks through the collaborative work of multiple modules, including multimodal data fusion, dynamic knowledge graph construction, risk feature extraction, risk scoring and early warning, and blockchain-based evidence storage and auditing. The system has performed exceptionally well in practical applications. For example, in a commercial bank's risk management project, the system successfully identified multiple potentially high-risk customers and issued early warnings, helping the bank avoid significant economic losses. Furthermore, the system possesses good scalability and adaptability, enabling customized development to meet the needs of different financial institutions, providing the financial industry with an efficient and reliable intelligent risk management solution.

Claims

1. A financial risk intelligent assessment system that enhances efficiency through multi-modal dynamic graph integration, characterized in that: It includes a data acquisition and preprocessing module, which is used to collect data from multiple sources and identify and semantically analyze unstructured data, extract key risk indicators and use homomorphic encryption technology to desensitize sensitive data; and a dynamic knowledge graph construction module, which is used to build a real-time updated knowledge graph with enterprises or individuals as nodes and relationships as edges, and to convert the graph structure into a vector representation through a graph embedding algorithm. The risk feature extraction module is used to extract static features, dynamic features, spatial features, and external event features. The risk scoring and early warning module is used to input features into the XGBoost-LightGBM hybrid model to output risk scores and set dynamic thresholds; the blockchain evidence storage and auditing module is used to put key data of the assessment process on the blockchain and support regulatory agencies to verify the compliance of the assessment process through smart contracts.

2. The financial risk intelligent assessment system with multi-modal dynamic graph joint efficiency improvement according to claim 1, characterized in that, The dynamic knowledge graph construction module is used to update the node and edge weights in the knowledge graph in real time through an incremental learning algorithm. The node weight update formula is: ,in, Let i be the weight of node i at time t; This is a smoothing coefficient used to control the degree of influence of historical weights; Let i be the set of neighboring nodes of node i; Let be the degree of neighbor node j; Let j be the weight of node i's neighbor node j at time t; is the Euclidean norm of a vector, used to calculate the length of a vector or the difference between two vectors; The edge weight update formula is as follows ,in, It is a nonlinear function based on the distance between nodes, used to quantify the tightness of the relationship between nodes; Let (i, j) be the weight of edge (i, j) at time t; This is the attenuation coefficient, used to reduce the influence of the old weights.

3. The financial risk intelligent assessment system with multi-modal dynamic graph joint efficiency improvement according to claim 1, characterized in that, The risk feature extraction module is used to capture risk propagation paths through the ST-GAT model, and its core formula is: ,in, For node i in the th... Layer feature representation; For activation functions; Let i be the set of neighboring nodes of node i; For the first Layer weight matrix; Represents a node In the Layer feature representation; The attention weight of node i to node j is calculated using the following formula: ;in, The exponential function maps input values ​​to the range of positive real numbers, ensuring that attention weights are positive; LeakyReLU is the activation function. This is a learnable attention vector used to calculate the attention score. It is the transpose of the vector; W is the weight matrix, used to perform linear transformation on the node features; Indicates a splicing operation; and These are the feature vectors of node i and node j, respectively.

4. The financial risk intelligent assessment system with multi-modal dynamic graph joint efficiency improvement according to claim 1, characterized in that, The risk scoring and early warning module is used to perform risk scoring using the XGBoost-LightGBM hybrid model, and its prediction formula is as follows: ,in, These are the predicted values ​​from the mixture model; Let be the prediction value of the k-th tree in XGBoost. This is the predicted value for LightGBM. The balancing coefficient is used to adjust the contribution ratio of the two models; the dynamic threshold calculation formula is: ,in, A dynamic threshold is used for risk warning; This is the historical average score. Standard deviation; It is a regulating factor.

5. The financial risk intelligent assessment system for multi-modal dynamic graph joint efficiency improvement according to claim 1, characterized in that, It also includes a training module for optimizing the dynamic knowledge graph construction module and the risk scoring and early warning module, specifically including: The training data generation module is used to generate large-scale labeled data through a simulation environment; The graph structure optimization module is used to optimize the topological structure of knowledge graphs through graph regularization constraints. Its objective function is: ;in, Let E represent the graph regularization loss function, which measures how well the graph structure is preserved in the embedding space; E is the set of edges. Let i be the vector representation of node i; For nodes The vector representation of ; V is the set of nodes; The regularization coefficient is used. The model fine-tuning module is used to adapt the XGBoost-LightGBM hybrid model to a specific domain through transfer learning. Its loss function is: ,in, For mission losses, Loss due to domain differences These are the transfer learning weights.

6. A method for intelligent financial risk assessment that enhances efficiency through multi-modal animated graphs, applied to the intelligent financial risk assessment system for enhancing efficiency through multi-modal animated graphs as described in any one of claims 1-5, characterized in that, This includes collecting data from multiple sources and identifying and semantically analyzing unstructured data; extracting key risk indicators and using homomorphic encryption technology to desensitize sensitive data; constructing a real-time updated knowledge graph with enterprises or individuals as nodes and relationships as edges, and converting it into a vector representation through graph embedding algorithms; and extracting static features, dynamic features, spatial features, and external event features. Input features into the XGBoost-LightGBM hybrid model to output risk scores and set dynamic thresholds; put key data of the assessment process on the blockchain and support regulatory agencies to verify the compliance of the assessment process through smart contracts.

7. An electronic device, characterized in that, It includes a processor and a memory, in which computer program instructions are stored. When the processor executes the computer program instructions, the processor causes the processor to perform the intelligent financial risk assessment method for improving efficiency through multi-modal animation as described in claim 6.

8. A computer-readable medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the intelligent financial risk assessment method for multi-modal animation joint efficiency improvement as described in claim 6.

Images