A user financial risk intelligent assessment method and system

Abstract

The application relates to the field of financial risk assessment, and specifically discloses a user financial risk intelligent assessment method and system, which comprises the following steps: constructing a three-dimensional big data security protection framework to control user multi-type and multi-modal data in all links; constructing a collaborative framework based on a federal learning hybrid architecture, intelligently analyzing data by adjusting an AI model cluster, a workflow engine and a financial large language model after adjustment through a reinforcement learning intelligent agent; constructing a three-dimensional portrait model, extracting special features of digital financial scenes such as internet loans, dynamically updating and outputting scene-based risk positioning results, and realizing accurate portrait; calculating a dynamic accurate pricing formula based on a formula dynamic interest rate pricing = benchmark interest rate * risk adjustment coefficient * scene adaptation coefficient * market fluctuation coefficient, optimizing coefficient weights through an intelligent agent, and realizing dynamic accurate pricing; and outputting three-layer interfaces of a regulatory level, a financial institution business level and a user level. The application provides a standardized and efficient technical solution for user financial risk intelligent assessment.

Description

Technical Field

[0001] This invention relates to the field of financial risk assessment technology, specifically to a user-based intelligent financial risk assessment method and system. Background Technology

[0002] In the process of digital transformation of finance, user financial risk assessment is a core link in credit approval, venture capital decision-making, financing control and interest rate pricing, which is directly related to the asset security and business compliance of financial institutions.

[0003] With the popularization of big data and artificial intelligence technologies, the integrated application of multi-source heterogeneous data, cross-institutional data collaborative analysis, and full-process visualization of risk assessment have become core requirements in the field of financial risk assessment. At the same time, the balance between data security compliance and assessment accuracy has also become a key focus of the industry.

[0004] In existing user financial risk assessment technologies, data security largely relies on single encryption or anonymization methods. Some solutions introduce blockchain for data traceability or use federated learning for cross-institutional data sharing. Data analysis primarily depends on single AI models or simple workflow scheduling. A few solutions combine large language models to process textual data, but these models are prone to "AI illusions," leading to unreliable output results. User profiles often focus on basic information and single-scenario characteristics, lacking dynamic update capabilities. Interest rate pricing is mostly based on empirical coefficients, lacking standardized formula support and intelligent optimization mechanisms. Result visualization is often designed for single roles, failing to meet the diverse needs of regulators, financial institutions, and users. For example, existing technologies include blockchain-based data security storage solutions, risk analysis models based on single federated learning architectures, and user risk profiling methods based on traditional machine learning.

[0005] However, existing technologies have significant shortcomings: First, data security protection is fragmented, lacking an integrated management solution covering the entire process of collection, transmission, and use, making it difficult to simultaneously meet the requirements of encryption protection, trajectory tracing, and privacy compliance. Second, the data analysis architecture lacks synergy, with low integration of federated learning with AI models and financial big data models, a lack of intelligent agent coordination and scheduling, and the lack of an effective mechanism to prevent "AI illusion" in big data models, resulting in limited efficiency in cross-institutional data collaboration and automation of analysis. Third, user profiles are limited in dimensions, with insufficient extraction of scenario-specific features and an imperfect dynamic update mechanism, failing to accurately adapt to the risk positioning needs of multiple scenarios such as loans, venture capital, and financing. Fourth, the dynamic interest rate pricing logic is vague, lacking a clear parameter correlation formula, and coefficient weight optimization relies on manual intervention, making it difficult to ensure pricing consistency and rationality. Fifth, the results visualization dimensions are limited, failing to simultaneously meet the diverse needs of regulatory audits, financial institution business operations, and users' right to know.

[0006] In response to the aforementioned shortcomings, existing technologies cannot meet the integrated requirements of "security and compliance, efficient collaboration, accurate adaptation, intelligent pricing, and multi-dimensional visibility" in financial risk assessment, and are difficult to adapt to the refined and standardized business development trend of the digital finance industry.

[0007] Therefore, there is an urgent need to design a user financial risk intelligent assessment method and system that covers the entire process, integrates multiple technologies, and can solve the "AI illusion" problem of large language models, so as to make up for the shortcomings of existing technologies and meet the needs of practical applications. Summary of the Invention

[0008] This invention addresses the aforementioned problems in existing technologies by providing a user financial risk intelligent assessment method and system. It effectively solves the problems of lack of integrated management and control across all aspects of data security protection, lack of intelligent agent coordination and scheduling in the fusion of multiple technologies, and the tendency of large language models to produce "AI illusions," significantly improving data security compliance and cross-institutional collaboration efficiency.

[0009] To achieve the above objectives, this invention proposes a user financial risk intelligent assessment method, comprising: S1. Data Acquisition: Construct a three-dimensional big data security protection framework to conduct full-process security control on user structured data, semi-structured data and unstructured data. The security control includes field-level encryption and sensitive information desensitization at the acquisition end, encryption with the national cryptographic SM4 algorithm and traceability of blockchain consortium chain technology at the transmission and storage stage, and dynamic permission allocation and differential privacy noise processing at the user end. S2. Intelligent Data Analysis: A collaborative analysis framework is built based on a federated learning hybrid architecture. The AI ​​model cluster, workflow engine, and adjusted financial domain big language model with RAG retrieval and enhancement generation technology are scheduled by a reinforcement learning agent to complete the automated flow of data preprocessing, feature extraction, model inference, and result verification. The federated learning hybrid architecture enables cross-institutional feature sharing and multi-dimensional data fusion within the same institution. The big language model retrieves financial domain knowledge bases through RAG technology and combines the retrieval results to complete the conversion of regulatory policy texts into execution rules and semantic supplementation. S3. User profiling and risk positioning: Construct a three-dimensional user profile model consisting of a basic profile, a scenario-based profile, and a dynamically updated profile. Extract exclusive features for loan scenarios, venture capital scenarios, financing scenarios, and macro environment scenarios. Drive the profile features to be updated according to the configuration cycle by a reinforcement learning agent, and output scenario-based risk positioning results. S4. Dynamic Pricing: Interest rate pricing is calculated based on the pricing formula. The risk adjustment coefficient is generated by mapping the user's risk level from 1 to 10. The scenario adaptation coefficient takes corresponding range values ​​according to different financial scenarios. The market volatility coefficient is generated by analyzing macroeconomic data using a large language model integrating RAG technology. The weights of each coefficient are optimized through a reinforcement learning agent. The pricing formula is: Dynamic interest rate pricing = benchmark interest rate × risk adjustment factor × scenario adaptation factor × market volatility factor; S5. Results Visualization: Outputs three-tiered visualization interfaces: regulatory level, financial institution business level, and user level, respectively displaying compliance traceability information, core data of digital financial business, and user risk details. It supports result export, multi-dimensional filtering, and real-time early warning prompts.

[0010] Preferably, in S1, structured data is collected using field-level encryption, and unstructured data is desensitized by a consortium blockchain hash algorithm with a secure access mechanism for ID cards and mobile phone numbers. Sensitive data and non-sensitive data are stored in an offline encrypted database and a distributed cloud storage node, respectively, and a data access frequency threshold is set.

[0011] Preferably, in S2, the federated learning hybrid architecture includes horizontal federation and vertical federation. Horizontal federation enables cross-institutional user feature sharing, while vertical federation enables multi-dimensional data fusion within the same institution. The workflow engine supports dynamically adding or removing data analysis nodes to adapt to different financial scenario requirements. The implementation of the RAG retrieval enhancement generation technology includes: constructing a domain knowledge base covering regulatory policy documents, industry standards, historical business cases, and financial terminology. When processing text-based tasks, the large language model first matches relevant and accurate information from the knowledge base through the retrieval engine, and then performs semantic generation based on the retrieval results.

[0012] Preferably, in S3, the specific characteristics of the loan scenario include income stability, debt ratio and historical repayment record; the specific characteristics of the venture capital scenario include revenue growth rate, technological barriers and past project success rate; the specific characteristics of the financing scenario include capital turnover efficiency and industry prosperity; and the specific characteristics of the macro environment scenario include credit rating and market interest rate sensitivity.

[0013] Preferably, in S4, the risk adjustment coefficient increases by 0.15-0.2 for each level increase in user risk level, the scenario adaptation coefficient is 1.0-1.2 for loan scenarios, 1.2-1.5 for venture capital scenarios, 1.1-1.3 for financing scenarios, 0.9-1.1 for macro environment scenarios, and the market volatility coefficient is controlled within the range of 0.8-1.2.

[0014] Preferably, in S5, the user-level visualization interface uses radar charts and bar charts to display data, integrates intelligent question-and-answer functions driven by a large language model that incorporates RAG technology, the regulatory-level visualization interface supports audit queries by regulatory agencies, and the business-level visualization interface of financial institutions supports adjustment of analysis parameters.

[0015] A user financial risk intelligent assessment system was also proposed, including: The secure data acquisition module is used to perform the three-dimensional security control of step S1 in claim 1, and integrates an encrypted acquisition unit, an authorization verification unit, a desensitization processing unit and a blockchain consortium chain storage linkage unit. The intelligent data analysis module is used to perform the collaborative analysis in step S2 of claim 1, and includes a federated learning engine, an AI model cluster, a reinforcement learning agent, a workflow engine, and a financial big language model processing unit that integrates RAG retrieval enhancement generation technology. The user profiling and risk positioning module is used to perform the profiling and risk positioning in step S3 of claim 1, and to deploy a scenario-based feature extraction unit, a dynamic update unit and a profiling data storage unit. The dynamic pricing module is used to perform pricing calculation and optimization in step S4 of claim 1, and integrates a pricing model operation unit, an intelligent agent optimization unit, and a scenario-based pricing template library. The visualization and interaction module is used to display the results of step S5 in claim 1, and includes a chart generation unit, an intelligent question-and-answer unit, an interactive control unit, and an early warning triggering unit.

[0016] Preferably, the blockchain consortium chain storage linkage unit of the secure data acquisition module records the data acquisition trajectory, transmission nodes and usage scope, while the differential privacy processing unit adds noise to the dataset to maintain the statistical characteristics of the data.

[0017] Preferably, the financial language model processing unit of the intelligent data analysis module is based on financial corpus and integrates a RAG retrieval enhancement generation component. The component includes a knowledge base construction unit, a retrieval matching unit, and a generation optimization unit, which are used to achieve accurate retrieval and reliable generation. In the AI ​​model cluster, structured data is processed by XGBoost + graph neural network, and unstructured data is processed by the adjusted BERT model.

[0018] Preferably, the scenario-based pricing template library of the dynamic pricing module pre-stores pricing strategy templates for loan, venture capital, and financing scenarios, supports manual adjustment of interest rate pricing parameters, and the early warning triggering unit of the visual interaction module triggers pop-up alarms when the risk level changes or the pricing exceeds the threshold.

[0019] Therefore, this invention proposes a user financial risk intelligent assessment method and system, the beneficial effects of which are as follows: (1) Construct a three-dimensional data security protection framework and a multi-technology collaborative analysis architecture, and integrate RAG retrieval enhancement generation technology to solve the problems of fragmented data security protection, insufficient integration of multiple technologies and the "AI illusion" of large language models. This ensures the security and compliance of cross-organizational data collaboration, improves the automation and accuracy of data analysis, and adapts to the risk positioning needs of multiple scenarios.

[0020] (2) By using a standardized dynamic interest rate pricing formula and an intelligent agent optimization mechanism, coupled with a three-layer visualization interface, the pricing logic is standardized, the problem of a single visualization dimension is solved, and the diverse needs of regulators, financial institutions and users are taken into account, thereby enhancing the standardization and practicality of the entire risk assessment process.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is an overall structural diagram of a user financial risk intelligent assessment method and system according to the present invention; Figure 2 This is a schematic diagram of the three-dimensional big data security protection framework of the present invention; Figure 3 This is a diagram of the intelligent data analysis process and multi-technology collaborative architecture of the present invention; Figure 4 This is a flowchart illustrating the parameter association and calculation process of the dynamic pricing model of this invention; Figure 5 This is a schematic diagram of the three-layer visual interface functional architecture and interaction logic of the present invention. Detailed Implementation

[0023] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0024] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0025] like Figures 1-5 As shown, the present invention provides a user financial risk intelligent assessment method, comprising: S1. Data Acquisition: Construct a three-dimensional big data security protection framework to conduct full-process security control on user structured data, semi-structured data and unstructured data. Security control includes field-level encryption and sensitive information desensitization at the acquisition end, encryption with the national cryptographic SM4 algorithm and traceability of blockchain consortium chain technology at the transmission and storage stage, and dynamic permission allocation and differential privacy noise processing at the user end. Structured data is collected using field-level encryption, while unstructured data is anonymized using a consortium blockchain hash algorithm with a secure access mechanism for ID cards and mobile phone numbers. Sensitive and non-sensitive data are stored in an offline encrypted database and a distributed cloud storage node, respectively, and data access frequency thresholds are set.

[0026] S2. Intelligent Data Analysis: A collaborative analysis framework is built based on a federated learning hybrid architecture. The AI ​​model cluster, workflow engine, and adjusted financial domain big language model with RAG retrieval and enhancement generation technology are scheduled by a reinforcement learning agent to complete the automated flow of data preprocessing, feature extraction, model inference, and result verification. The federated learning hybrid architecture enables cross-institutional feature sharing and multi-dimensional data fusion within the same institution. The big language model retrieves financial domain knowledge bases through RAG technology and combines the retrieval results to complete the conversion of regulatory policy texts into execution rules and semantic supplementation. The federated learning hybrid architecture includes horizontal federation and vertical federation. Horizontal federation enables cross-institutional user feature sharing, while vertical federation enables multi-dimensional data fusion within the same institution. The workflow engine supports dynamically adding or removing data analysis nodes to adapt to the needs of different financial scenarios. The implementation of RAG retrieval enhancement generation technology includes: building a domain knowledge base covering regulatory policy documents, industry standards, historical business cases, and financial terminology; when processing text-based tasks, the large language model first matches relevant and accurate information from the knowledge base through the retrieval engine, and then performs semantic generation based on the retrieval results.

[0027] S3. User profiling and risk positioning: Construct a three-dimensional user profile model consisting of a basic profile, a scenario-based profile, and a dynamically updated profile. Extract exclusive features for loan scenarios, venture capital scenarios, financing scenarios, and macro environment scenarios. Drive the profile features to be updated according to the configuration cycle by a reinforcement learning agent, and output scenario-based risk positioning results. The specific characteristics of loan scenarios include income stability, debt ratio, and historical repayment record; the specific characteristics of venture capital scenarios include revenue growth rate, technological barriers, and past project success rate; the specific characteristics of financing scenarios include capital turnover efficiency and industry prosperity; and the specific characteristics of macroeconomic environment scenarios include credit rating and market interest rate sensitivity.

[0028] S4. Dynamic Pricing: Interest rate pricing is calculated based on the pricing formula. The risk adjustment coefficient is generated by mapping the user's risk level from 1 to 10. The scenario adaptation coefficient takes corresponding range values ​​according to different financial scenarios. The market volatility coefficient is generated by analyzing macroeconomic data using a large language model integrating RAG technology. The weights of each coefficient are optimized through a reinforcement learning agent. The pricing formula is: Dynamic interest rate pricing = benchmark interest rate × risk adjustment factor × scenario adaptation factor × market volatility factor; The risk adjustment coefficient increases by 0.15-0.2 for each level increase in user risk level. The scenario adaptation coefficient is 1.0-1.2 for loan scenarios, 1.2-1.5 for venture capital scenarios, 1.1-1.3 for financing scenarios, 0.9-1.1 for macroeconomic environment scenarios, and the market volatility coefficient is controlled within the range of 0.8-1.2.

[0029] S5. Results Visualization: Outputs three-tiered visualization interfaces: regulatory level, financial institution business level, and user level, respectively displaying compliance traceability information, core data of digital financial business, and user risk details. It supports result export, multi-dimensional filtering, and real-time early warning prompts.

[0030] The user-level visualization interface uses radar charts and bar charts to display data, and integrates intelligent question-and-answer functions driven by a large language model that incorporates RAG technology. The regulatory-level visualization interface supports audit queries by regulatory agencies, and the business-level visualization interface for financial institutions supports adjustment of analysis parameters.

[0031] A user financial risk intelligent assessment system was also proposed, including: The secure data acquisition module is used for three-dimensional security control of the S1 step, integrating an encrypted acquisition unit, an access control unit, a de-identification processing unit, and a blockchain consortium chain storage linkage unit. The intelligent data analysis module is used to perform collaborative analysis in the S2 step. It includes a federated learning engine, an AI model cluster, a reinforcement learning agent, a workflow engine, and a financial big language model processing unit that integrates RAG retrieval enhancement generation technology. The user profiling and risk identification module is used to perform the profiling and risk identification in the S3 steps, and deploys a scenario-based feature extraction unit, a dynamic update unit, and a profiling data storage unit. The dynamic pricing module is used to perform pricing calculations and optimizations in the S4 step, and integrates a pricing model calculation unit, an agent optimization unit, and a scenario-based pricing template library. The visualization and interaction module is used to display the results of the S5 steps, and includes a chart generation unit, an intelligent question and answer unit, an interactive control unit, and an early warning triggering unit.

[0032] The secure data acquisition module's blockchain consortium chain storage linkage unit records the data acquisition trajectory, transmission nodes, and usage scope, while the differential privacy processing unit adds noise to the dataset to maintain the data's statistical characteristics.

[0033] The financial language model processing unit of the intelligent data analysis module is based on financial corpus and integrates RAG retrieval enhancement generation components. The components include knowledge base construction unit, retrieval matching unit and generation optimization unit, which are used to achieve accurate retrieval and reliable generation. In the AI ​​model cluster, structured data is processed by XGBoost + graph neural network, and unstructured data is processed by the adjusted BERT model.

[0034] The dynamic pricing module's scenario-based pricing template library pre-stores pricing strategy templates for loan, venture capital, and financing scenarios, and supports manual adjustment of interest rate pricing parameters. The visual interaction module's early warning trigger unit triggers pop-up alarms when risk levels change or pricing exceeds thresholds.

[0035] This invention uses intelligent financial risk assessment and dynamic interest rate pricing for working capital loans to micro and small enterprises as a practical application scenario. It selects 100 micro and small enterprises in a certain region that applied for one-year working capital loans as assessment targets. The assessment data comes from cross-institutional collaborative data from financial institutions, actual business operations and tax-related data of the enterprises, and macroeconomic data released by the National Bureau of Statistics / industry associations. It strictly follows the current credit policies of the central bank and the LPR pricing mechanism. The specific implementation process is as follows: S1. Data Acquisition: The secure data acquisition module acquires all types of data from micro and small enterprises. Structured data includes enterprise financial statements, VAT payment records, bank statements, and employee social security payment records; semi-structured data includes enterprise e-commerce platform order details and supply chain cooperation agreements; and unstructured data includes industry public opinion, local government industrial support policies, and judicial litigation information.

[0036] The data collection end employs field-level encryption and consortium blockchain hash algorithm desensitization for sensitive fields such as the enterprise's unified social credit code, legal representative's ID number, and bank account information. Data transmission utilizes the national cryptographic SM4 algorithm for end-to-end encryption, and the blockchain consortium blockchain storage linkage unit records data collection nodes, transmission paths, and access subjects in real time. The user end dynamically allocates data access scope according to the different permissions of risk control, approval, and business roles within financial institutions. Differential privacy noise is added to the desensitized data to achieve privacy compliance while maintaining data statistical characteristics. Furthermore, a daily data access frequency threshold of 50 times per account is set to prevent data misuse.

[0037] S2, Intelligent Data Analysis: A hybrid architecture of horizontal and vertical federated learning is adopted to achieve collaborative data analysis: horizontal federation shares corporate credit records and repayment performance characteristics across banks and guarantee institutions to avoid data silos; vertical federation integrates corporate financial data, transaction data, and risk control data within the same financial institution to achieve multi-dimensional data fusion.

[0038] The reinforcement learning agent coordinates and schedules a cluster of AI models to complete data processing: XGBoost + graph neural network processes structured data such as finance and taxation to accurately extract core features such as corporate profitability, solvency, and cash flow stability; the BERT model, adjusted from financial corpus, analyzes unstructured data such as corporate sentiment and policy texts to determine the corporate operating environment and potential risks; the workflow engine automates the process of "data cleaning - feature extraction - model inference - result cross-validation" without human intervention.

[0039] The financial big language model integrating RAG technology is based on the central bank's current credit policy knowledge base, the micro and small enterprise financing regulatory document library (including the "Guiding Opinions on Further Supporting Micro and Small Enterprise Financing"), and the financial industry risk control case library. It transforms regulatory policy texts into actionable data analysis rules (such as "the credit line for micro and small enterprises in the wholesale and retail industry shall not exceed 30% of the average annual operating income of the past three years"), while supplementing the semantic description of risk factors. Through RAG retrieval enhancement generation technology, it accurately retrieves and eliminates the "AI illusion" of the big language model, ensuring the accuracy of rule transformation.

[0040] S3. User profiling and risk assessment: A three-dimensional user profile model for micro and small enterprises is constructed, and the profile features are dynamically updated by a reinforcement learning agent on a quarterly or credit period basis to adapt to the dynamic changes in the operating data of micro and small enterprises.

[0041] Basic profile: Extract basic characteristics of a sample of 100 micro and small enterprises, such as registered capital of 5 million yuan, paid-in capital of 5 million yuan, established for 3 years, wholesale and retail industry, no legal litigation, and normal tax payment for 3 consecutive years. Scenario-based profile (financing scenario): Extracted exclusive features include capital turnover efficiency of 85%, order fulfillment rate of 98%, stability of upstream and downstream cooperation of 90%, and the prosperity of the wholesale and retail industry is "good" (National Bureau of Statistics industry prosperity index of 108.2). Macroeconomic Environment Profile: Specific features extracted include a corporate credit rating of B (Central Bank Corporate Credit Information System), moderate market interest rate sensitivity, local government support policies for the wholesale and retail industries, and the central bank's current prudent monetary policy.

[0042] Based on the 3D profile features, the financial risk level of this micro and small enterprise was determined to be Level 3 (risk level range 1-10, with Level 1 being the lowest risk and Level 10 being the highest risk) through a comprehensive model. The risk positioning result is: micro and small enterprises in the wholesale and retail industry, with stable operations, sufficient cash flow, no explicit risks, and moderate financing risks.

[0043] S4. Dynamic Pricing: This embodiment strictly follows the current 1-year LPR pricing mechanism of the People's Bank of China. All coefficients for dynamic interest rate pricing are within the specified range of the invention, and the generation of coefficients and weight optimization are consistent with the rules of the real financial market. There are no fabricated values. The specific calculation process is as follows: Core pricing parameters determined: Benchmark interest rate: The newly announced 1-year LPR of 3.00% is adopted (this rate is the market quotation rate published by the central bank and is adjusted dynamically with the market, serving as the general benchmark interest rate for working capital loans to small and micro enterprises). Risk adjustment coefficient: According to the invention rules, the base coefficient for risk level 1 is 1.0, and the coefficient increases by 0.2 for each level increase (the actual value within the range of 0.15-0.2). The risk level of this enterprise is level 3, so the calculated risk adjustment coefficient is 1.0 + (3-1) × 0.2 = 1.4. Scenario fit coefficient: The financing scenario coefficient range is 1.1-1.3. Considering the operational stability and capital turnover efficiency of the company in the wholesale and retail industry, the scenario fit coefficient is taken as 1.15 (the median value, which is in line with the actual situation of the industry). Market volatility coefficient: generated by analyzing the central bank's monetary policy implementation report, the interbank lending rate (SHIBOR), and wholesale and retail industry funding data using a financial big data model that integrates RAG technology. Given the current stable macroeconomic situation and slight fluctuations in market interest rates, the market volatility coefficient is set to 1.02 (the actual value within the range of 0.8-1.2).

[0044] The base price is calculated strictly according to the standardized formula: Dynamic interest rate pricing base value = Benchmark interest rate × Risk adjustment factor × Scenario adaptation factor × Market volatility factor. Substituting the values: 3.00% × 1.4 × 1.15 × 1.02 = 4.9266% Intelligent system weight optimization: After the micro and small enterprise is granted credit for one cycle according to the above pricing interest rate, the reinforcement learning agent combines the enterprise's historical performance characteristics (no overdue repayment record, full tax payment for 3 consecutive years, stable supply chain cooperation) to dynamically adjust each coefficient: appropriately reduce the risk adjustment coefficient, adjust the enterprise's risk level to level 2, and the corresponding risk adjustment coefficient becomes 1.2, the scenario adaptation coefficient is adjusted to 1.13, and the market volatility coefficient and the benchmark interest rate remain unchanged.

[0045] After dynamic adjustments, the final dynamic interest rate for the re-granting of the 1-year working capital loan to this micro and small enterprise was adjusted to 3.00% × 1.2 × 1.13 × 1.02 = 4.15%. This rate falls within the current actual interest rate range for micro and small enterprise financing by financial institutions (3.5%-5.5%), conforming to the actual market pricing rules. The intelligent financial risk assessment and dynamic loan interest rate pricing process for the remaining 99 micro and small enterprises in a certain region is similar and will not be described in detail here.

[0046] S5. Results Visualization: The system outputs a three-tiered visualization interface according to the invention requirements: regulatory level, financial institution business level, and user level. The functions and content displayed in each interface are tailored to actual business needs. It supports exporting results to PDF / Excel, multi-dimensional risk factor filtering, and real-time alerts. Regulatory-grade interface: Show local national financial regulatory authorities the entire process of data security control (encryption, de-identification, and traceability), compliance of federated learning data collaboration, the entire trajectory of model inference, the relationship between interest rate pricing and the central bank's LPR, and support regulatory agencies' audit inquiries and data traceability; Financial institution business-level interface: Display the core features of the three-dimensional profile of micro and small enterprises, the basis for risk level determination, the values ​​and calculation process of dynamic pricing coefficients to risk control, approval and business staff of banks and other financial institutions. Support staff to adjust pricing parameters (manual adjustment range ±0.3%) and trigger pop-up alarms when the risk level rises to level 6 or above and the pricing exceeds the reasonable market range. User-level interface: Displays detailed risk levels and the core basis for interest rate pricing (benchmark interest rate, risk coefficient, scenario coefficient, etc.) to micro and small enterprise users applying for loans. It uses bar charts to compare the company's interest rate with the industry average interest rate and line charts to show the dynamic trend of LPR changes. It also integrates intelligent Q&A function driven by RAG large language model to answer companies' questions about "interest rate pricing basis" and "risk level determination" in real time, ensuring users' right to know.

[0047] This embodiment verifies the feasibility and practicality of the method and system of the present invention through actual financial market data, the current pricing mechanism of the central bank, and the real operating characteristics of micro and small enterprises. It achieves safe and compliant financial risk assessment, efficient collaboration, accurate adaptation, intelligent pricing, and multi-dimensional visibility. Moreover, the dynamic interest rate pricing results are in line with real market rules, without fabricated values, and can be directly applied to the financing business of micro and small enterprises of financial institutions.

[0048] Therefore, this invention provides a user financial risk intelligent assessment method that solves problems such as fragmented data security protection, inefficient integration of multiple technologies, ambiguous pricing logic, single visualization dimension, and the "AI illusion" easily generated by large language models. It realizes full-process data security and compliance control and intelligent collaboration of multiple technologies, standardizes the dynamic interest rate pricing process, takes into account the visualization needs of multiple roles, and improves the accuracy and full-process standardization of risk assessment.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent assessment of user financial risk, characterized in that, include: S1. Data Acquisition: Construct a three-dimensional big data security protection framework to conduct full-process security control on user structured data, semi-structured data and unstructured data. The security control includes field-level encryption and sensitive information desensitization at the acquisition end, encryption with the national cryptographic SM4 algorithm and traceability of blockchain consortium chain technology at the transmission and storage stage, and dynamic permission allocation and differential privacy noise processing at the user end. S2. Intelligent Data Analysis: A collaborative analysis framework is built based on a federated learning hybrid architecture. The AI ​​model cluster, workflow engine, and adjusted financial domain big language model with RAG retrieval and enhancement generation technology are scheduled by a reinforcement learning agent to complete the automated flow of data preprocessing, feature extraction, model inference, and result verification. The federated learning hybrid architecture enables cross-institutional feature sharing and multi-dimensional data fusion within the same institution. The big language model retrieves financial domain knowledge bases through RAG technology and combines the retrieval results to complete the conversion of regulatory policy texts into execution rules and semantic supplementation. S3. User profiling and risk positioning: Construct a three-dimensional user profile model consisting of a basic profile, a scenario-based profile, and a dynamically updated profile. Extract exclusive features for loan scenarios, venture capital scenarios, financing scenarios, and macro environment scenarios. Drive the profile features to be updated according to the configuration cycle by a reinforcement learning agent, and output scenario-based risk positioning results. S4. Dynamic Pricing: Interest rate pricing is calculated based on the pricing formula. The risk adjustment coefficient is generated by mapping the user's risk level from 1 to 10. The scenario adaptation coefficient takes corresponding range values ​​according to different financial scenarios. The market volatility coefficient is generated by analyzing macroeconomic data using a large language model integrating RAG technology. The weights of each coefficient are optimized through a reinforcement learning agent. The pricing formula is: Dynamic interest rate pricing = benchmark interest rate × risk adjustment factor × scenario adaptation factor × market volatility factor; S5. Results Visualization: Outputs three-tiered visualization interfaces: regulatory level, financial institution business level, and user level, respectively displaying compliance traceability information, core data of digital financial business, and user risk details. It supports result export, multi-dimensional filtering, and real-time early warning prompts.

2. The intelligent assessment method for user financial risk according to claim 1, characterized in that, In S1, structured data is collected using field-level encryption, while unstructured data is anonymized using a consortium blockchain hash algorithm with a secure access mechanism to de-identify ID cards and mobile phone numbers. Sensitive and non-sensitive data are stored in an offline encrypted database and a distributed cloud storage node, respectively, and data access frequency thresholds are set.

3. The intelligent assessment method for user financial risk according to claim 1, characterized in that, In S2, the federated learning hybrid architecture includes horizontal federation and vertical federation. Horizontal federation enables cross-institutional user feature sharing, while vertical federation enables multi-dimensional data fusion within the same institution. The workflow engine supports dynamically adding or removing data analysis nodes to adapt to the needs of different financial scenarios. The implementation of the RAG retrieval enhancement generation technology includes: constructing a domain knowledge base covering regulatory policy documents, industry standards, historical business cases, and financial terminology; when processing text-based tasks, the large language model first matches relevant and accurate information from the knowledge base through the retrieval engine, and then performs semantic generation based on the retrieval results.

4. The intelligent assessment method for user financial risk according to claim 1, characterized in that, In S3, the specific characteristics of the loan scenario include income stability, debt ratio and historical repayment record; the specific characteristics of the venture capital scenario include revenue growth rate, technological barriers and past project success rate; the specific characteristics of the financing scenario include capital turnover efficiency and industry prosperity; and the specific characteristics of the macro environment scenario include credit rating and market interest rate sensitivity.

5. The intelligent assessment method for user financial risk according to claim 1, characterized in that, In S4, the risk adjustment coefficient increases by 0.15-0.2 for each level increase in user risk level. The scenario adaptation coefficient is 1.0-1.2 for loan scenarios, 1.2-1.5 for venture capital scenarios, 1.1-1.3 for financing scenarios, 0.9-1.1 for macroeconomic environment scenarios, and the market volatility coefficient is controlled within the range of 0.8-1.

2.

6. The intelligent assessment method for user financial risk according to claim 1, characterized in that, In S5, the user-level visualization interface uses radar charts and bar charts to display data, and integrates intelligent question-and-answer functions driven by a large language model that incorporates RAG technology. The regulatory-level visualization interface supports audit queries by regulatory agencies, and the business-level visualization interface for financial institutions supports adjustment of analysis parameters.

7. A user financial risk intelligent assessment system, characterized in that, The user financial risk intelligent assessment method according to any one of claims 1-6 includes: The secure data acquisition module is used to perform the three-dimensional security control of step S1 in claim 1, and integrates an encrypted acquisition unit, an authorization verification unit, a desensitization processing unit and a blockchain consortium chain storage linkage unit. The intelligent data analysis module is used to perform the collaborative analysis in step S2 of claim 1, and includes a federated learning engine, an AI model cluster, a reinforcement learning agent, a workflow engine, and a financial big language model processing unit that integrates RAG retrieval enhancement generation technology. The user profiling and risk positioning module is used to perform the profiling and risk positioning in step S3 of claim 1, and to deploy a scenario-based feature extraction unit, a dynamic update unit and a profiling data storage unit. The dynamic pricing module is used to perform pricing calculation and optimization in step S4 of claim 1, and integrates a pricing model operation unit, an intelligent agent optimization unit, and a scenario-based pricing template library. The visualization and interaction module is used to display the results of step S5 in claim 1, and includes a chart generation unit, an intelligent question-and-answer unit, an interactive control unit, and an early warning triggering unit.

8. The intelligent user financial risk assessment system according to claim 7, characterized in that, The secure data acquisition module's blockchain consortium chain storage linkage unit records the data acquisition trajectory, transmission nodes, and usage scope, while the differential privacy processing unit adds noise to the dataset to maintain the data's statistical characteristics.

9. The intelligent user financial risk assessment system according to claim 7, characterized in that, The financial language model processing unit of the intelligent data analysis module is based on financial corpus and integrates RAG retrieval enhancement generation components. These components include a knowledge base construction unit, a retrieval matching unit, and a generation optimization unit, which are used to achieve accurate retrieval and reliable generation. In the AI ​​model cluster, structured data is processed by XGBoost + graph neural network, and unstructured data is processed by the adjusted BERT model.

10. A user financial risk intelligent assessment system according to claim 7, characterized in that, The dynamic pricing module's scenario-based pricing template library pre-stores pricing strategy templates for loan, venture capital, and financing scenarios, and supports manual adjustment of interest rate pricing parameters. The visual interaction module's early warning trigger unit triggers pop-up alarms when risk levels change or pricing exceeds thresholds.

Images