Financial risk prediction system and method based on AI analysis
By integrating multimodal data fusion and constructing dynamic knowledge graphs, and combining GNN and Transformer models, the problems of data isolation and insufficient interpretation in financial risk prediction are solved, enabling real-time, panoramic, and transparent prediction of enterprise risks, thereby improving the efficiency and accuracy of risk management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGBEI UNIVERSITY OF FINANCE AND ECONOMICS
- Filing Date
- 2026-01-23
- Publication Date
- 2026-04-17
AI Technical Summary
Existing technologies for financial risk prediction suffer from problems such as limited data dimensions, isolated analytical perspectives, delayed risk insights, and insufficient model interpretability, making it difficult to effectively integrate multimodal data and dynamically understand the complex relationships between entities.
It employs a multimodal data real-time fusion module, a dynamic association knowledge graph construction module, and a hybrid AI risk inference engine, combined with GNN and Transformer models, to construct a dynamic knowledge graph and perform risk prediction and visualization analysis.
It enables real-time fusion and dynamic correlation of multi-dimensional data, captures risk signals in advance, provides transparent and interpretable risk prediction results, improves the comprehensiveness of risk identification and prediction accuracy, and supports penetrating insight into systemic risks and the credibility of decision-making basis.
Smart Images
Figure CN121883170A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of financial risk prediction technology, specifically relating to a financial risk prediction system and method based on AI analysis. Background Technology
[0002] The stability and healthy development of financial markets depend on the accurate identification and effective management of potential risks. For a long time, the assessment and prediction of financial risks have primarily relied on the analysis of structured financial data, such as corporate balance sheets, income statements, cash flow statements, and market data like stock prices and trading volumes. While these traditional methods can reflect a company's historical operating conditions and market performance to some extent, their inherent limitations are becoming increasingly apparent.
[0003] First, traditional financial data has a significant lag. Corporate financial reports are usually released quarterly or annually, which means that by the time risk signals appear in the financial reports, the risk event may have already occurred or developed for quite some time, resulting in a slow response to risk management measures and missing the best intervention window.
[0004] Secondly, traditional methods are often isolated and one-dimensional. They focus on analyzing the financial indicators of individual business entities, while ignoring the intricate relationships between businesses in a modern market economy. A company's risk may stem from production disruptions at its upstream suppliers, financial crises at its downstream core customers, or the break in an inconspicuous node in a guarantee chain. These implicit risks existing in the industrial chain, supply chain, guarantee chain, and equity chain are difficult for traditional models to capture.
[0005] In recent years, with the development of big data and artificial intelligence technologies, some new attempts have emerged, such as using public opinion analysis to assist in risk assessment. However, most of these methods remain at a superficial level of data application. They may analyze a single type of unstructured data (such as news), but fail to effectively integrate this "alternative data" with traditional financial data. More importantly, they lack a framework capable of dynamically depicting and understanding the complex relationships between entities, resulting in fragmented analysis results that fail to form a comprehensive and forward-looking insight into the transmission path and systemic impact of risks. When a model issues a high-risk warning, it often fails to clearly explain the root cause of the risk, key driving factors, and possible evolutionary paths, becoming a "black box" that is difficult for decision-makers to accept and adopt.
[0006] Therefore, existing technologies generally suffer from problems such as limited data dimensions, isolated analytical perspectives, delayed risk insights, and insufficient model interpretability. The market urgently needs a next-generation financial risk prediction solution that can integrate multimodal data, deeply understand the dynamic relationships between entities, and provide transparent and interpretable predictive results. Summary of the Invention
[0007] To overcome the above-mentioned technical problems, the present invention provides a financial risk prediction system and method based on AI analysis.
[0008] The present invention adopts the following technical solution: A financial risk prediction system based on AI analysis includes: A multimodal data real-time fusion module for collecting and processing structured financial data and unstructured alternative data from multiple data sources; A dynamic association knowledge graph construction module is used to transform the data processed by the multimodal data real-time fusion module into a dynamic knowledge graph composed of entity nodes and relation edges, and to update the knowledge graph in real time according to newly collected data; A hybrid AI risk inference engine is connected to the dynamic knowledge graph construction module, which integrates at least two AI models for inferring and predicting the financial risks of target entities based on the dynamic knowledge graph. A risk panorama visualization and attribution early warning module is used to present the prediction results of the hybrid AI risk inference engine in a visual form and generate a risk attribution report.
[0009] Preferably, the unstructured alternative data includes at least one or more of the following: public opinion news data, supply chain and logistics data, industrial policy and regulatory document data, and human resource mobility data.
[0010] Preferably, the AI model integrated within the hybrid AI risk simulation engine includes: A systemic risk transmission model based on GNN is used to predict the transmission path and probability of risk between different entity nodes on the dynamic knowledge graph. A multi-dimensional time-series prediction model based on the Transformer architecture is used to integrate multi-dimensional time-series data of a single entity node to predict its future risk indicators.
[0011] Preferably, the hybrid AI risk inference engine further includes a risk factor fusion and reinforcement learning decision model, which integrates the outputs of the systemic risk transmission model and the multidimensional time series prediction model to give a comprehensive risk score and identify the key driving factors that cause changes in the risk score.
[0012] Preferably, the risk panorama visualization and attribution early warning module can generate risk attribution reports in natural language format and display the predicted risk transmission path in a visualized knowledge graph slice by highlighting nodes and indicative lines.
[0013] Preferably, it includes the following steps: Real-time collection and fusion of structured financial data and unstructured alternative data; Based on the fused data, a dynamic association knowledge graph consisting of entity nodes and relation edges is constructed, and the knowledge graph is continuously and dynamically updated according to newly collected data. A hybrid AI risk inference engine is used to perform calculations on the dynamically related knowledge graph to predict the financial risk of the target entity; Generate and display a visual report that includes risk scores, risk attribution analysis, and risk transmission paths, and trigger an alert when the risk reaches a preset threshold.
[0014] Preferably, the step of collecting and fusing unstructured alternative data specifically includes collecting data on public opinion news, supply chain logistics, industrial policy supervision, and human resource flow, and using natural language processing technology to perform entity recognition and sentiment quantification analysis on the text data.
[0015] Preferably, in the step of constructing a dynamic relational knowledge graph, entity nodes represent companies, executives, or events, and relation edges represent relationships such as investment, guarantee, supply chain, or legal proceedings. Nodes and edges in the graph are generated, modified, or deleted in real time based on newly occurring data events.
[0016] Preferably, the step of making predictions using a hybrid AI risk inference engine specifically includes: By running a GNN model, the cross-entity transmission of systemic risk is extrapolated on the knowledge graph; By running a Transformer model, we can analyze multidimensional time series data of a single entity to predict its own risk evolution. The prediction results of the two models are then merged to generate a final comprehensive risk assessment.
[0017] Preferably, the step of generating and displaying the visualization report specifically includes explaining the main causes of the risk in natural language and dynamically displaying the predicted propagation path of the risk from the source node to the target node on a graphical interface.
[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention breaks through the limitations of relying solely on lagging financial data. Through a "real-time multimodal data fusion module," it incorporates high-frequency, real-time alternative data such as public opinion, supply chain, policy, and personnel changes into its analysis. This allows the system to capture early risk signals beyond financial statements, such as the escalation of negative news or the disruption of cooperation with core suppliers, thus significantly advancing the risk warning window and buying valuable time for risk response. The multidimensionality of the data also constructs a 360-degree panoramic risk profile of the target entity, improving the comprehensiveness of risk identification.
[0019] The core of this invention lies in the "dynamic relational knowledge graph construction module." It not only stores data but also connects isolated data points into a dynamically evolving relationship network, intuitively depicting the complex relationships between enterprises, individuals, and events, such as investment, guarantees, and supply chains. Based on this graph, the system can discover and deduce hidden risk transmission paths that traditional methods cannot reach. For example, the system can predict the potential impact of a crisis originating from a third-tier supplier, transmitted through the supply chain, on core enterprises. This penetrating insight into systemic and interconnected risks is unparalleled by existing technologies.
[0020] This invention employs a hybrid AI risk inference engine, organically combining multiple advanced models such as GNN and Transformer. GNN excels at learning and inferring risk transmission patterns on knowledge graphs, while the Transformer model can deeply mine the evolutionary patterns in the multi-dimensional time series data of an entity itself. This dual perspective analysis of "individual + network" enables the model to understand both the systemic shocks of the external environment and the heterogeneous risks within the entity. The combination of the two produces prediction accuracy and robustness far exceeding that of a single model.
[0021] The risk panorama visualization and attribution early warning module of this invention not only provides a risk score, but more importantly, it can clearly explain "why there is risk," "where the risk comes from," and "how the risk will evolve" in the form of visualized graphs and natural language reports. Users can intuitively see the predicted risk transmission path and key risk driving factors, making the AI analysis results transparent, credible, and traceable. This enhances the trust of financial practitioners in AI models, transforming abstract risk warnings into concrete and actionable decision-making basis, thereby achieving a leap from "risk perception" to "intelligent decision-making." Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation
[0023] The embodiments of the present invention are described in detail below. Unless otherwise specified, the methods in the embodiments are conventional methods in the art. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0024] The financial risk prediction system in this embodiment mainly consists of four core modules: a multimodal data real-time fusion module, a dynamic association knowledge graph construction module, a hybrid AI risk inference engine, and a risk panorama visualization and attribution early warning module.
[0025] I. Real-time Multimodal Data Fusion Module This module forms the data foundation of the entire system, responsible for collecting and processing data related to business operations from multiple dimensions and sources in real time.
[0026] First, at the data acquisition level, the system not only accesses traditional structured data, such as quarterly and annual financial statements of enterprises, stock market trading data, credit history records, and macroeconomic indicators; more importantly, it acquires large-scale unstructured and semi-structured alternative data through customized web crawlers and API interfaces. This alternative data mainly includes: Public opinion and news data: Real-time capture of all information about the target company, its executives, its industry, and its upstream and downstream partners from mainstream financial media, industry websites, social media platforms, and other channels.
[0027] Supply chain and logistics data: By connecting with data interfaces of customs, port logistics platforms, large e-commerce platforms and enterprise resource planning (ERP) systems, we can obtain real business activity data such as import and export records, warehousing and circulation, and order fulfillment.
[0028] Industry policies and regulatory documents: Automatically tracks the latest industry policies, environmental regulations, regulatory penalty notices, etc. issued by government departments. These documents are often important indicators of industry-wide risks.
[0029] Human Resources Mobility Data: By partnering with mainstream recruitment websites, we anonymously track the recruitment needs and departure dynamics of core technical personnel and senior managers within enterprises, using this as a measure of internal stability.
[0030] At the data processing level, this module cleans, aligns, and standardizes the collected raw data. For unstructured text data, the system uses natural language processing technology to perform entity recognition, such as identifying company names, personal names, and product names, and performs sentiment analysis to determine whether each piece of information has a positive, neutral, or negative impact on the relevant entities, quantifying it as a sentiment score. For supply chain data, it performs correlation matching to build clear relationships between trading counterparties.
[0031] II. Dynamically Related Knowledge Graph Construction Module This module is one of the core components of this invention. It is responsible for transforming the multidimensional data processed by the previous module into a dynamically evolving enterprise-related knowledge graph.
[0032] In this graph, each node represents an entity, such as a company, an executive, a product, a policy, or a specific event. The edges between nodes represent the relationships between them, such as investment relationships, guarantee relationships, upstream and downstream relationships in the supply chain, joint shareholding, executive concurrent positions, technological cooperation, or legal proceedings.
[0033] Unlike static knowledge graphs, the knowledge graph in this system is "dynamic." The graph updates in real time as the multimodal data fusion module captures new information. For example, when the system detects news that Company A and Company B have signed a large procurement contract, a "supply" edge from Company B to Company A will be generated or strengthened on the graph, and attributes such as contract amount and time will be assigned to this edge. Conversely, if the system detects that Company A has defaulted on payments to its supplier Company B, the weight or attributes of this edge will be updated to reflect this potential risk.
[0034] In this way, the system weaves isolated data points into a vast, dynamically changing network of business relationships, laying the foundation for revealing the transmission path of risks.
[0035] III. Hybrid AI Risk Inference Engine This is the system's "brain," which contains a set of collaborative AI models that perform risk calculations and deductions based on a dynamic knowledge graph.
[0036] A systemic risk transmission model based on Graph Neural Networks (GNNs) operates directly on a dynamic knowledge graph. It learns not the isolated risks of individual enterprises, but rather the transmission patterns of risk throughout the network. When a node in the graph exhibits risk symptoms due to its own reasons, the GNN model can predict the probability, pathways, and duration of the risk's impact on related enterprises, based on the graph's topology, edge weights, and transmission patterns learned from historical data. This is crucial for early warning of cascading supply chain or guarantee chain risks.
[0037] A multi-dimensional time-series forecasting model based on the Transformer architecture focuses on in-depth risk assessment of a single enterprise node. It integrates all relevant data for an enterprise, including time-varying financial indicators, public sentiment scores, and employee turnover indicators, into a multi-dimensional time series. Leveraging the Transformer model's powerful ability to capture long-range dependencies, this model can reveal complex interactions between different dimensions of data and predict specific indicators such as the probability of default and the risk of cash flow depletion for the enterprise in the next quarter or six months. For example, the model may find a strong correlation between "increased turnover of core technical personnel," "product development delays three months later," and "revenue decline six months later."
[0038] Risk Factor Fusion and Reinforcement Learning Decision Model: The risk prediction results from the two models mentioned above, along with the original feature data, are input into a final reinforcement learning model. This model acts as a "risk decision-maker," not only integrating all information to provide a final comprehensive risk score, but more importantly, outputting the key driving factors that lead to changes in the risk rating. Through reinforcement learning training, the model learns to weigh the importance of different risk signals in complex situations, thus providing judgments closer to those of human experts.
[0039] IV. Risk Panorama Visualization and Attribution Early Warning Module This module is the system's user interface, which presents complex analysis results to risk managers in an intuitive and user-friendly manner.
[0040] When a user searches for a company, the system displays a dynamic risk dashboard. The dashboard not only shows the company's current overall risk score (e.g., from 0 to 100), but also provides a risk attribution report in natural language, such as: "Warning: The target company's risk score has increased by 15 points in the past week, primarily due to export restrictions imposed by its upstream key raw material supplier, 'Company C,' in its country. This is expected to lead to a 20% increase in the company's production costs over the next two months. The systemic risk transmission model predicts that there is a 75% probability that this cost pressure will be passed on to its downstream customers, 'Company D' and 'Company E'." Meanwhile, the interface will display relevant knowledge graph slices in a visual way. Risk nodes will be highlighted in red, and predicted risk transmission paths will be marked with flashing arrows, allowing users to see the ins and outs of risks at a glance.
[0041] When any enterprise's risk score exceeds the preset threshold, or when the system predicts that a high-risk transmission path is about to form, the system will send a proactive warning to the relevant person in charge via email, SMS or instant messaging tools to ensure that the risk can be detected and intervened in a timely manner at the bud stage.
[0042] In summary, this invention successfully addresses many pain points of traditional risk control models by constructing a closed-loop system encompassing data fusion, graph construction, AI inference, and visual early warning. It not only provides accurate profiles of individual enterprises but also reveals the systemic risk transmission mechanisms within complex business networks, achieving a qualitative leap from "post-event attribution" to "pre-event prediction." This system has significant application value in areas such as bank credit approval, supply chain finance, portfolio management, and insurance underwriting, and can significantly improve the risk management level and decision-making efficiency of financial institutions.
[0043] Although embodiments of the present invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to the above embodiments without departing from the principles and spirit of the present invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A financial risk prediction system based on AI analysis, characterized in that, include: A multimodal data real-time fusion module for collecting and processing structured financial data and unstructured alternative data from multiple data sources; A dynamic association knowledge graph construction module is used to transform the data processed by the multimodal data real-time fusion module into a dynamic knowledge graph composed of entity nodes and relation edges, and to update the knowledge graph in real time according to newly collected data; A hybrid AI risk inference engine is connected to the dynamic knowledge graph construction module, which integrates at least two AI models for inferring and predicting the financial risks of target entities based on the dynamic knowledge graph. A risk panorama visualization and attribution early warning module is used to present the prediction results of the hybrid AI risk inference engine in a visual form and generate a risk attribution report.
2. The system according to claim 1, characterized in that, The unstructured alternative data includes at least one or more of the following: public opinion news data, supply chain and logistics data, industrial policy and regulatory document data, and human resource mobility data.
3. The system according to claim 1, characterized in that, The AI models integrated within the hybrid AI risk simulation engine include: A systemic risk transmission model based on GNN is used to predict the transmission path and probability of risk between different entity nodes on the dynamic knowledge graph. A multi-dimensional time-series prediction model based on the Transformer architecture is used to integrate multi-dimensional time-series data of a single entity node to predict its future risk indicators.
4. The system according to claim 3, characterized in that, The hybrid AI risk inference engine also includes a risk factor fusion and reinforcement learning decision model, which integrates the outputs of the systemic risk transmission model and the multidimensional time series prediction model to give a comprehensive risk score and identify the key driving factors that cause changes in the risk score.
5. The system according to claim 1, characterized in that, The risk panorama visualization and attribution early warning module can generate risk attribution reports in natural language format and display the predicted risk transmission path in the visualized knowledge graph slices by highlighting nodes and indicative lines.
6. A financial risk prediction method based on AI analysis, characterized in that, Includes the following steps: Real-time collection and fusion of structured financial data and unstructured alternative data; Based on the fused data, a dynamic association knowledge graph consisting of entity nodes and relation edges is constructed, and the knowledge graph is continuously and dynamically updated according to newly collected data. A hybrid AI risk inference engine is used to perform calculations on the dynamically related knowledge graph to predict the financial risk of the target entity; Generate and display a visual report that includes risk scores, risk attribution analysis, and risk transmission paths, and trigger an alert when the risk reaches a preset threshold.
7. The method according to claim 6, characterized in that, The steps of collecting and integrating unstructured alternative data specifically include collecting data on public opinion news, supply chain logistics, industrial policy supervision, and human resource flow, and using natural language processing technology to perform entity recognition and sentiment quantification analysis on the text data.
8. The method according to claim 6, characterized in that, In the steps of constructing a dynamic relational knowledge graph, entity nodes represent companies, executives, or events, and relationship edges represent relationships such as investment, guarantee, supply chain, or legal proceedings. Nodes and edges in the graph are generated, modified, or deleted in real time based on newly occurring data events.
9. The method according to claim 6, characterized in that, The steps for making predictions using a hybrid AI risk inference engine specifically include: By running a GNN model, the cross-entity transmission of systemic risk is extrapolated on the knowledge graph; By running a Transformer model, we can analyze multidimensional time series data of a single entity to predict its own risk evolution. The prediction results of the two models are then merged to generate a final comprehensive risk assessment.
10. The method according to claim 6, characterized in that, The steps of generating and displaying the visualization report specifically include explaining the main causes of the risk in natural language and dynamically displaying the predicted propagation path of the risk from the source node to the target node on a graphical interface.