Real estate business and financial business risk control management method based on big data model

Abstract

The invention discloses a real estate business financial business risk control management method based on a big data model, and relates to the technical field of financial science and technology and risk management, and the method comprises the steps: collecting multi-dimensional original data, carrying out the fusion cleaning, and generating a structured risk control basic data set; extracting three feature sequences of application subject static portraits, real estate dynamic values and associated enterprise risks from the application subject static portraits and the real estate dynamic values; inputting the multi-source risk control feature fusion model into a pre-trained multi-source risk control feature fusion model for interactive modeling, and generating a comprehensive risk feature map; constructing a hierarchical risk decision network by using the map, and outputting a multi-dimensional vector containing credit and collateral risks and associated party conduction risk scores; and based on the vector, generating an adaptive credit line, a guarantee rate floating and a post-loan monitoring scheme through a dynamic strategy matching engine, and converting the adaptive credit line, the guarantee rate floating and the post-loan monitoring scheme into an executable instruction set to be issued and executed. According to the invention, the deep correlation analysis of the multi-source risk information and the automatic and accurate generation of the risk management strategy are realized.

Description

Technical Field

[0001] This invention belongs to the field of financial technology and risk management technology, specifically a risk control management method for real estate financial business based on big data models. Background Technology

[0002] Currently, financial institutions generally adopt a modular, independent assessment framework in their real estate finance risk control. Credit risk assessment relies on static financial and credit data models of the applicant; collateral risk management is based on independent valuation using external assessment reports or market price indices; and related-party risks are managed statically through manual verification of equity relationships and the creation of lists. These technical methods place core risk elements such as borrowers, collateral, and related parties in separate assessment processes.

[0003] Existing solutions have shortcomings. Risk data and assessment conclusions from different dimensions are fragmented, failing to construct a unified risk view to quantify the dynamic transmission effects of risk across multiple entities and assets. Traditional methods lack effective technical pathways to deeply integrate and correlate borrower credit changes, collateral market fluctuations, and related network risk events. Furthermore, risk assessment results are severely disconnected from the formulation of specific business strategies. The determination of key strategy parameters such as credit limits, loan-to-value ratios, and post-loan monitoring intensity relies heavily on expert experience for manual judgment and matching, resulting in delayed strategy responses, inconsistent standards, and difficulty in addressing complex and ever-changing interconnected risk scenarios.

[0004] A risk management method is needed that can achieve deep fusion of multi-source risk information and automated strategy generation. This method needs to solve the technical challenges of moving from isolated data assessment to comprehensive risk correlation insights, and from risk quantification to the automatic generation of differentiated strategy instructions, in order to improve the accuracy of risk identification and the efficiency of strategy execution. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a risk control management method for real estate and financial businesses based on a big data model, comprising: Collect multi-dimensional raw data related to real estate and financial businesses; The multi-dimensional raw data is fused and cleaned to generate a structured risk control basic dataset in a unified format; From the structured risk control basic dataset, extract the static profile feature sequence of the credit applicant, the dynamic value feature sequence of the target real estate, and the associated risk feature sequence of related enterprises; The static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence are input into a pre-trained multi-source risk control feature fusion model for interactive modeling to generate a comprehensive risk feature map. Using the comprehensive risk feature map, a hierarchical risk decision network is constructed to generate a multi-dimensional risk assessment vector for the credit applicant. The multi-dimensional risk assessment vector includes credit dimension score, collateral risk dimension score and related party transmission risk dimension score. Based on the multidimensional risk assessment vector, a dynamic strategy matching engine generates an adaptive risk management strategy, which includes a credit limit configuration scheme, a collateral ratio floating scheme, and a post-loan monitoring scheme. The adaptive risk management strategy is converted into an executable risk control instruction set, and the risk control instruction set is sent to the financial business system for execution.

[0006] Furthermore, the process of fusing and cleaning the multi-dimensional raw data to generate a structured risk control dataset in a unified format includes: The multi-dimensional raw data includes the credit records of the credit applicant, the historical transaction records and market valuation fluctuation series of the target real estate, the operating cash flow data of related enterprises, and legal risk disclosure texts; The credit records are standardized and parsed to extract time-series data on the credit status of individuals or enterprises; Align and denoise the historical transaction records and market valuation fluctuation sequences to generate real estate value evolution trajectory data; Pattern recognition is performed on the aforementioned operational flow data to extract indicators of enterprise operational stability; Keyword extraction and sentiment analysis are performed on the legal risk disclosure text to generate a quantitative risk warning vector; The credit status time series data, real estate value evolution trajectory data, enterprise operation stability indicators and quantitative risk warning vectors are associated, aligned and structured according to a unified spatiotemporal coordinate to form the structured risk control basic dataset.

[0007] Furthermore, from the structured risk control dataset, static profile feature sequences of credit applicants are extracted, including: Analyze the historical debt records of credit applicants in the structured risk control dataset to generate debt burden characteristics; Analyze the income stability records of credit applicants in the structured risk control dataset to generate income stability features; Identify the associated real estate holding records of the credit applicant in the structured risk control basic dataset and generate asset diversity features; The debt burden characteristics, income stability characteristics, and asset diversity characteristics are time-series encoded to form the static profile feature sequence.

[0008] Furthermore, from the structured risk control dataset, the dynamic value feature sequence of the target real estate is extracted, including: The value evolution trajectory data of target real estate in the structured risk control basic dataset is analyzed to identify the periodic patterns and long-term trends of value fluctuations. Correlation analysis is performed on the regional market supply and demand data of the target real estate in the structured risk control basic dataset to generate market liquidity prediction features; Integrate the physical attributes of the target real estate to generate property value decay or gain adjustment features; The cyclical patterns are fused and encoded with long-term trends, market liquidity prediction characteristics, and underlying asset value decay or gain adjustment characteristics to form the dynamic value feature sequence.

[0009] Furthermore, the step of inputting the static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence into a pre-trained multi-source risk control feature fusion model for interactive modeling to generate a comprehensive risk feature map includes: The multi-source risk control feature fusion model includes multiple interactive sub-networks, which respectively process the static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence; A feature attention bridging mechanism is established between interactive sub-networks to calculate the risk correlation between the static profile feature sequence and the dynamic value feature sequence, as well as the risk transmission strength of the correlated risk feature sequence to the static profile feature sequence. Based on the calculated risk correlation and risk transmission strength, the various feature sequences are weighted, fused, and information diffused to generate the comprehensive risk feature map. The comprehensive risk feature map uses map nodes to represent risk entities and edge weights to represent risk correlation strength.

[0010] Furthermore, the step of constructing a hierarchical risk decision network using the comprehensive risk feature map to generate a multi-dimensional risk assessment vector for the credit applicant includes: The hierarchical risk decision-making network includes a credit risk analysis layer, a collateral risk analysis layer, and an association risk analysis layer; The credit risk analysis layer receives the graph substructures related to the credit of the credit applicant from the comprehensive risk feature graph and generates the credit dimension score. The collateral risk analysis layer receives the substructures of the comprehensive risk feature map that are related to the value and liquidity of the target real estate, and generates the collateral risk dimension score. The associated risk analysis layer receives the graph substructures related to the risk transmission of associated enterprises from the comprehensive risk feature graph, and generates the associated party transmission risk dimension score; The credit dimension score, collateral risk dimension score, and related party transmission risk dimension score are combined to form the multidimensional risk assessment vector.

[0011] Furthermore, the collateral risk analysis layer receives the substructures related to the target real estate value and liquidity from the comprehensive risk feature map, and generates the collateral risk dimension score, including: From the graph substructure related to the value and liquidity of the target real estate, the value fluctuation nodes, market liquidity nodes and their connection relationships are analyzed. Under different market stress scenarios, the changes in the state of the value fluctuation nodes affect the market liquidity nodes through their connection relationships, and the potential value decay path and liquidity difficulty of the collateral are calculated. Based on the potential value decay path and the difficulty of realization, the risk dimension score of the collateral is calculated through the quantitative scoring model embedded in the collateral risk analysis layer.

[0012] Furthermore, the step of generating an adaptive risk management strategy based on the multidimensional risk assessment vector through a dynamic strategy matching engine includes: The multidimensional risk assessment vector is input into the dynamic strategy matching engine; The dynamic strategy matching engine presets multiple basic risk management strategy templates, and each basic risk management strategy template is associated with a risk vector threshold range; The multidimensional risk assessment vector is matched with the threshold range of each risk vector to determine the closest basic risk management strategy template; Based on the specific scores of each dimension in the multidimensional risk assessment vector, the strategy parameters in the closest basic risk management strategy template are fine-tuned to generate the adaptive risk management strategy. The adaptive risk management strategy specifically includes a quantified credit limit allocation scheme, a score-based collateral ratio floating scheme, and a targeted post-loan monitoring scheme.

[0013] Furthermore, the step of fine-tuning the strategy parameters in the closest basic risk management strategy template based on the specific numerical values ​​of the scores for each dimension in the multidimensional risk assessment vector to generate the adaptive risk management strategy includes: For the credit limit allocation scheme, a credit limit calculation function is established, and the input variables of the credit dimension score and the collateral risk dimension score are included; For the aforementioned floating mortgage ratio scheme, a mortgage ratio adjustment rule is established, which is based primarily on the risk dimension score of the collateral and modified with reference to the risk dimension score of related parties. For the post-loan monitoring scheme, a monitoring frequency and key monitoring indicators are set. The monitoring frequency is negatively correlated with the risk dimension score of the related party transmission. The key monitoring indicators are selected based on the dimension with the lowest score in the multi-dimensional risk assessment vector. Based on the credit limit calculation function, the collateral ratio adjustment rules, and the set monitoring frequency and key monitoring indicators, specific strategy parameter values ​​are calculated to complete the fine-tuning of the basic risk management strategy template.

[0014] Furthermore, the step of converting the adaptive risk management strategy into an executable risk control instruction set includes: Analyze the credit limit configuration scheme in the adaptive risk management strategy to generate corresponding credit approval instructions and credit limit locking instructions; Analyze the loan-to-value ratio floating scheme in the adaptive risk management strategy, and generate collateral value revaluation instructions and contract terms update instructions; The post-loan monitoring scheme in the adaptive risk management strategy is analyzed to generate instructions for setting data collection frequency, risk indicator threshold alarm instructions, and periodic review task creation instructions. All generated instructions are arranged according to the business logic execution order to form a complete set of risk control instructions that can be processed and executed sequentially by the financial business system.

[0015] Compared with the prior art, the beneficial effects of the present invention are: The pre-trained multi-source risk control feature fusion model performs deep interactive modeling of static profiles, dynamic value, and associated risk feature sequences, achieving unified representation and correlation mining of heterogeneous risk information. This model captures the complex nonlinear relationships and temporal dependencies between different features through algorithms, fusing previously fragmented credit status, asset value fluctuations, and associated network events to generate a structured, comprehensive risk feature map. This technology enables the system to identify cross-risk signals that cannot be revealed in single-dimensional assessments, quantifying the potential transmission paths and strengths of risk among entity credit, collateral assets, and related enterprise networks, thereby overcoming the problem of a one-dimensional risk view in traditional modular assessments.

[0016] A hierarchical risk decision-making network built upon a comprehensive risk feature map can perform path-specific reasoning calculations and output a structured, multi-dimensional risk assessment vector. This vector clearly separates and quantifies core dimensions such as credit risk, collateral risk, and related-party transmission risk. The dynamic strategy matching engine parses and maps this vector, simultaneously driving the coordinated generation and adjustment of a series of strategy parameters, including credit limit allocation, collateral ratio fluctuations, and post-loan monitoring plans. This process achieves automatic conversion and precise matching from risk quantification to differentiated, executable management strategies.

[0017] The hierarchical network and dynamic strategy engine work together to seamlessly connect risk identification and strategy execution. Based on real-time evolving multi-dimensional risk assessment results, the system can automatically and instantly generate a complete set of risk control instructions, including credit limits, guarantee conditions, and monitoring priorities. This eliminates the subjectivity and lag of manual strategy matching, ensuring dynamic adaptation of risk management measures to the current risk situation and improving the consistency and responsiveness of strategy execution. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the risk control management method for real estate and financial businesses based on a big data model, as described in this invention. Figure 2 This is a flowchart for extracting feature sequences from static images. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1The system collects multi-dimensional raw data related to real estate financial transactions from internal and external data sources. This data covers multiple risk dimensions, including loan applicants, target collateral, and related companies. The acquired multi-dimensional raw data is then fused and cleaned, transforming raw data from different sources and formats into a unified, standardized, and directly processable structured risk control dataset. From this structured dataset, feature engineering methods are used to extract static profile feature sequences representing the creditworthiness of the loan applicant, dynamic value feature sequences reflecting the market value and liquidity trends of the target real estate, and related risk feature sequences revealing the risk status of related companies. These three feature sequences are then fed into a pre-trained multi-source risk control feature fusion model. This model can perform deep interaction and fusion modeling of features from different sources, outputting a comprehensive risk feature map that reflects the complex risk relationships between the applicant, collateral, and related parties. Using this comprehensive risk feature map, a hierarchical risk decision-making network is further constructed. This network deconstructs and quantifies comprehensive risk from different levels, such as credit risk, collateral risk, and related-party transmission risk, ultimately generating a multi-dimensional risk assessment vector containing scores for multiple risk dimensions. Based on this multi-dimensional risk assessment vector, a dynamic strategy matching engine performs intelligent matching and fine-tuning, outputting a quantifiable risk management strategy highly adapted to the current business risk situation. This strategy specifically includes credit limit allocation schemes, collateral ratio floating schemes, and post-loan monitoring schemes. Finally, the aforementioned management strategy in text or parameter form is decomposed and transformed into a set of instructions with clear operational logic that can be recognized and executed by the financial business system—the risk control instruction set—and issued to relevant business systems for execution, thus completing the closed loop from risk analysis to the implementation of risk control measures.

[0021] In one embodiment of the present invention, the multi-dimensional raw data includes the credit records of the credit applicant, historical transaction records and market valuation fluctuation sequences of the target real estate, operating cash flow data of related enterprises, and legal risk disclosure texts. After data collection, the system initiates corresponding cleaning and parsing procedures for each type of raw data, performs standardized parsing of credit records, and extracts time-series data of the credit status of individuals or enterprises. Credit records are usually obtained from data sources such as the Credit Reference Center of the People's Bank of China and Baihang Credit in PDF, XML, or structured message formats. The parsing procedure identifies and extracts key fields from the report, including the total number of credit accounts, current overdue amount, repayment status in the last 24 months, maximum credit card limit, and total outstanding loan amount. These discrete record points are organized into a time-ordered credit status sequence according to the report date or data update date, where each time point contains a snapshot of the credit overview of the credit applicant at that moment.

[0022] In practice, the historical transaction records and market valuation fluctuation sequences of the target real estate are aligned and denoised to generate real estate value evolution trajectory data. Historical transaction records can come from online transaction data registered with the real estate transaction center, while market valuation fluctuation sequences can come from listing prices, valuation model outputs, or regional housing price indices released by the government from third-party real estate information platforms. Due to differences in time granularity and statistical caliber among different data sources, the alignment and denoising process includes: uniformly sampling time series data from different channels to the same time frequency, such as monthly; identifying and removing outliers far from the mean range by calculating the moving average and standard deviation of data at adjacent time points; and uniformly mapping the identifiers corresponding to the same real estate target across all data sources, for example, by combining "community name-building number-unit number-room number" into a unified real estate unit code, thus aggregating data from different sources under the same target property to form a continuous and smooth value evolution trajectory data. This trajectory data not only includes the original price points but also includes processed trend lines, volatility, and other derived indicators.

[0023] In some embodiments, pattern recognition is performed on the operating cash flow data of related enterprises to extract enterprise operating stability indicators. The operating cash flow data can be a summary of quarterly or monthly corporate account transactions, including the recipient, sender, amount, transaction time, and summary. The pattern recognition process analyzes the regularity of incoming payments, such as calculating the total monthly operating revenue and its coefficient of variation to measure the degree of revenue fluctuation; it analyzes the composition of outgoing payments, identifying fixed and variable expenses such as wages, taxes, raw material purchases, and rent, and calculates the proportion of fixed expenses to total revenue, i.e., the operating rigid expenditure ratio; it observes seasonal cycles, such as the peak in revenue during holidays in the retail industry, and extracts the intensity of seasonal components through time series decomposition. The monthly revenue coefficient of variation, operating rigid expenditure ratio, and seasonal component intensity are combined to form the enterprise operating stability indicators.

[0024] In some embodiments, keyword extraction and sentiment analysis are performed on legal risk disclosure texts to generate a quantitative risk warning vector. The legal risk disclosure texts originate from channels such as enterprise information query platforms, court announcement websites, and the China Securities Regulatory Commission's information disclosure website. The text content includes judgments, court hearing announcements, administrative penalty decisions, equity freeze information, announcements of abnormal business operations, and related news reports. The text processing program first performs named entity recognition, extracting key entities and events such as "enforced party," "dishonest enforced party," "restriction on high consumption," "administrative penalty," "equity pledge," and "litigation amount," and assigning a time tag to each event. Then, sentiment analysis is performed to determine the severity of the risk events described in the text; for example, "serious violation" is marked as high risk, and "general violation" as medium risk. By statistically analyzing the frequency of events at different risk levels within a preset observation window and weighting them by the event's value, a multi-dimensional quantitative risk warning vector is finally generated. For example, the risk warning vector can be represented as... ,in Indicates the risk count of high-frequency events. Represents the total weighted risk value. Indicates the time interval between the most recent risk occurrences. Weighted total risk value. From the formula:

[0025] in: Iterate through all identified risk events. Indicates the first Preset weighting coefficients for each risk event. Indicates the first Each risk event involves a normalized value of the underlying asset amount.

[0026] It is understandable that after the above data have been processed and feature extracted, the final step in generating a structured risk control basic dataset in a unified format is to perform association alignment and structured storage. The core of the association alignment operation is to establish a unified spatiotemporal coordinate system. The time coordinate is uniformly set to Greenwich Mean Time or Beijing timestamp, while the spatial coordinate is the unique identifier of all data subjects and targets. For example, the ID number of the credit applicant, the real estate unit code of the target real estate, and the unified social credit code of the associated enterprise are used as association keys. The system integrates the credit status time-series data, real estate value evolution trajectory data, enterprise operational stability indicators, and quantitative risk warning vectors generated in the previous steps into the same wide table or graph database node according to their corresponding subject identifiers and collection time points. Each time point record in the credit status time-series data is associated with the valuation in the real estate value evolution trajectory data, the current data in the enterprise operational stability indicators, and the quantitative risk warning vector generated before that time point in the legal risk disclosure text. The final structured risk control basic dataset is stored in a row structure, with each row recording a multi-dimensional state snapshot of a core subject at a specific time point, or in a graph structure, where nodes represent subjects and real estate, attributes include various indicators, and edges represent the holding relationship between the subject and the real estate, and the relationship between the subject and the enterprise.

[0027] See Figure 2In one embodiment of the present invention, a static profile feature sequence of the credit applicant is extracted from the structured risk control basic dataset. This process involves analyzing the historical debt records of the credit applicant in the structured risk control basic dataset, including the total outstanding loan principal, the average monthly repayment amount over the past six months, the total credit card limit used, and the outstanding balance of external guarantees, to calculate derived indicators such as the debt-to-income ratio and the overall asset-liability ratio, generating a numerical sequence representing the characteristics of its debt burden. By analyzing the income stability records of the credit applicant in the structured risk control basic dataset, such as the past thirty-six months of salary bank statements, personal income tax payment records, and corporate account income statements of operating enterprises, the average monthly income, the moving average of income growth rate, and the standard deviation of income volatility are calculated to generate a numerical sequence representing the characteristics of its income stability. By identifying the related real estate holding records of the credit applicant in the structured risk control basic dataset, such as the registration and valuation information of other residential properties, shops, office buildings, and other real estate registered under their personal or family names, the number of properties held, the total valuation, and the regional distribution dispersion are statistically analyzed to generate a numerical sequence representing the characteristics of its asset diversity. The aforementioned debt burden characteristics, income stability characteristics, and asset diversity characteristics are aligned and integrated along the same time axis. The feature values ​​at each time point are standardized and normalized to form a static profile feature sequence of the credit applicant. This sequence can be represented as a multi-dimensional time series matrix, where each row of the matrix represents a time point and each column represents a specific static feature dimension.

[0028] In practice, the dynamic value feature sequence of the target real estate is extracted from the same structured risk control dataset. This process involves analyzing the value evolution trajectory data of the target real estate in the structured risk control dataset. This trajectory data includes the time series points of historical transaction prices and market appraisal prices for various periods. Using time series decomposition techniques, long-term trend components, seasonal cyclical components, and residual components are separated from the value evolution trajectory data to identify the cyclical patterns and long-term trends of value fluctuations. For example, it identifies annual cyclical fluctuations in property value and a long-term compound annual growth rate. By correlation analysis of market supply and demand data in the region where the target real estate is located in the structured risk control dataset, such as monthly new listings, monthly viewings, monthly sales volume, and sales cycle days for similar properties in the region, Granger causality tests or vector autoregression models are performed on these supply and demand indicators and the historical price data of the target real estate itself to generate a market liquidity prediction feature that reflects the ease of real estate realization in a certain future period. By integrating the physical attributes of the target real estate, such as building age, floor location, orientation, unit type, and decoration condition, and combining them with industry depreciation rate standards and maintenance cost data of similar properties, the impact of physical depreciation and functional depreciation on value is calculated, generating property value decay or gain adjustment characteristics. The mathematical expressions of identified periodic patterns and long-term trends, the probability distribution parameters of market liquidity prediction characteristics, and the quantitative values ​​of property value decay or gain adjustment characteristics are then fused and encoded to form a dynamic value characteristic sequence of the target real estate. This sequence is also a multi-dimensional time-series data structure, reflecting the time-varying characteristics and risk factors of the collateral value.

[0029] In some embodiments, the system extracts the associated risk feature sequence of related enterprises from the structured risk control dataset. This process is accomplished through time-series analysis of the operational stability indicators and quantitative risk warning vectors of the related enterprises. The operational stability indicators include time-series values ​​such as monthly revenue variation coefficient and operating rigid expenditure ratio, while the quantitative risk warning vectors include time-series values ​​such as high-frequency event risk count, weighted total risk value, and time interval between recent risk occurrences. The system performs sliding window analysis on these time-series indicators to identify the downward trend of operational indicators and the upward trend of risk vectors. It then maps the deterioration of operational stability indicators with the strengthening of quantitative risk warning vectors, forming an associated risk feature sequence that reflects changes in the overall risk status of related enterprises. Subsequently, the static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence are input into the pre-trained multi-source risk control feature fusion model. The multi-source risk control feature fusion model contains three parallel interactive sub-networks. The first interactive sub-network is a temporal convolutional network, which is specifically used to process the static profile feature sequence of the credit applicant. The second interactive sub-network is a long short-term memory network, which is specifically used to process the dynamic value feature sequence of the target real estate. The third interactive sub-network is a graph attention network, which is specifically used to process the associated enterprise risk relationship graph constructed by the associated enterprise risk feature sequence of associated enterprises.

[0030] In some embodiments, during feature extraction in the interactive sub-network, the multi-source risk control feature fusion model establishes a feature attention bridging mechanism between the interactive sub-networks to calculate the risk correlation between the static profile feature sequence and the dynamic value feature sequence. The risk correlation is calculated by the following formula:

[0031] in: This represents the risk correlation weight matrix. This represents the hidden state representation of a static image feature sequence in a certain network layer. This represents the hidden state representation of the dynamic value feature sequence in the corresponding network layer. and It is a learnable linear transformation weight matrix. This is the scaling factor. Simultaneously, the feature attention bridging mechanism calculates the risk transmission strength of the associated risk feature sequence to the static profile feature sequence. This is achieved by performing cross-modal attention calculation between the graph representation of the associated risk feature sequence and the representation of the static profile feature sequence. Based on the calculated risk correlation weight matrix and risk transmission strength, the multi-source risk control feature fusion model performs weighted fusion and information diffusion on the hidden state representations of each feature sequence. For example, the risk correlation weight matrix is ​​applied to the representation of the dynamic value feature sequence, and then the weighted representation is added to the representation of the static profile feature sequence, realizing the fusion of information from collateralized logistics to the subject.

[0032] It is understandable that, through interactive modeling of the multi-source risk control feature fusion model, a comprehensive risk feature map is ultimately generated. This map represents risk entities using map nodes, including credit applicant nodes, target real estate nodes, related enterprise nodes, and other key entity nodes extracted from the original data. The comprehensive risk feature map uses edge weights to represent the strength of risk correlation. These edge weights are quantified and assigned based on the risk correlation and risk transmission strength calculated by the feature attention bridging mechanism. For example, between the credit applicant node and the target real estate node, the edge weight is the risk correlation. An aggregated scalar value; between the associated enterprise node and the credit applicant node, the edge weight is a quantified value of the risk transmission strength. In this way, the diverse and heterogeneous static profile feature sequences, dynamic value feature sequences, and associated risk feature sequences are uniformly encoded in a graph structure containing rich semantic relationships, forming a comprehensive risk feature map for subsequent hierarchical risk decision-making.

[0033] In one embodiment of the present invention, the hierarchical risk decision-making network comprises three independent but interconnected analysis layers: a credit risk analysis layer, a collateral risk analysis layer, and a related risk analysis layer. The credit risk analysis layer receives a graph substructure related to the creditworthiness of the loan applicant from a comprehensive risk feature graph. This graph substructure includes a loan applicant node, directly connected historical credit record nodes, income flow nodes, asset and liability nodes, and edges connecting these nodes. The credit risk analysis layer analyzes and calculates this graph substructure. The analysis path includes traversing all first-degree-of-degree-of-connection nodes originating from the loan applicant node, aggregating the attribute values ​​of these connected nodes and the weights of the connecting edges, and performing nonlinear transformation and mapping on the aggregated information through a multilayer perceptron network. Finally, a scalar score is output, which represents the credit dimension score of the applicant's credit risk.

[0034] In practical implementation, the collateral risk analysis layer receives the substructures related to the value and liquidity of the target real estate from the comprehensive risk characteristic map. These substructures include the target real estate node, the historical value node of the target real estate, the regional market node where the target real estate is located, similar property nodes in the same area, and the edges connecting these nodes. The collateral risk analysis layer analyzes and calculates this substructure. Specifically, it extracts nodes representing value fluctuations, nodes representing market liquidity, and the connections between them from the substructure related to the value and liquidity of the target real estate. Nodes representing value fluctuations include the target real estate node itself and its historical value nodes, while nodes representing market liquidity include regional market nodes and similar property nodes. The collateral risk analysis layer simulates the impact of changes in the state of value fluctuation nodes on market liquidity nodes through connections under different market stress scenarios. Different market stress scenarios include preset scenarios such as macroeconomic downturn, regional policy adjustments, and sudden events. The simulation process adopts the message passing mechanism of graph neural networks. Under the preset stress impact, the feature vector of the value fluctuation node will undergo preset changes. This change propagates and diffuses to the market liquidity node along the connection edges in the graph substructure according to the edge weight ratio. The potential value decay path and the difficulty of realization of collateral under various simulated scenarios are calculated. The potential value decay path is the value estimation sequence of the target real estate node under different simulation step sizes. The difficulty of realization is the state indicator of the market liquidity node after the stress transmission, such as the number of days of the simulated sales cycle.

[0035] In some embodiments, the collateral risk analysis layer, through comprehensive simulation calculations of the potential value decay path and liquidity difficulty, calculates a collateral risk dimension score using a quantitative scoring model embedded within the collateral risk analysis layer. The quantitative scoring model can be a regression model, whose inputs are key statistics of the potential value decay path extracted after simulation, such as maximum drawdown rate and value volatility, as well as a quantitative value of liquidity difficulty, such as the average clearance period under stress scenarios. The quantitative scoring model maps these input features to a risk score range, such as 0 to 100 points, with higher scores indicating lower collateral risk. The score output by the quantitative scoring model embedded within the collateral risk analysis layer is the collateral risk dimension score. The associated risk analysis layer receives a graph substructure related to risk transmission among associated enterprises from the comprehensive risk feature graph. This graph substructure includes associated enterprise nodes, associated enterprise operational risk nodes, associated enterprise legal risk nodes, edges connecting associated enterprise nodes to operational and legal risk nodes, and risk transmission path edges from associated enterprise nodes to credit applicant nodes. The associated risk analysis layer analyzes and calculates this graph substructure, aggregates information about associated enterprise nodes and their associated risk nodes through a graph attention network, and pays special attention to the weights of the transmission path edges pointing to the credit application subject node, calculates the cumulative effect of risk transmission, and finally outputs an associated risk transmission dimension score representing the degree of associated risk transmission.

[0036] Understandably, after the hierarchical risk decision-making network completes three parallel analysis processes, the credit risk analysis layer outputs a credit dimension score, the collateral risk analysis layer outputs a collateral risk dimension score, and the related-party transmission risk dimension score outputs a related-party transmission risk dimension score. Finally, the credit dimension score from the credit risk analysis layer, the collateral risk dimension score from the collateral risk analysis layer, and the related-party transmission risk dimension score from the related-party risk analysis layer are combined. This combination operation arranges the three independent scalar scores in a preset order to form a numerical vector with fixed dimensions, such as a three-dimensional vector.

[0037] in: This indicates a credit score. This indicates the risk score of the collateral. This represents the risk score based on related-party transmission. This three-dimensional vector is a multi-dimensional risk assessment vector for the credit applicant. It quantifies the risk status of a real estate financial business application from three orthogonal dimensions: credit, collateral, and related-party transmission.

[0038] In one embodiment of the present invention, a multidimensional risk assessment vector is input into a dynamic strategy matching engine. This multidimensional risk assessment vector is a numerical vector containing credit dimension scores, collateral risk dimension scores, and related party transmission risk dimension scores. For example... The first value, 85, represents the credit score; the second value, 70, represents the collateral risk score; and the third value, 60, represents the related-party risk score. The dynamic strategy matching engine pre-sets multiple basic risk management strategy templates. Each template is a strategy framework with a complete parameter structure, and each template is associated with a risk vector threshold range. These threshold ranges define the value range of the multidimensional risk assessment vector in each scoring dimension, dividing the continuous risk scoring space into discrete strategy matching regions. The dynamic strategy matching engine compares the input multidimensional risk assessment vector with the internally stored risk vector threshold ranges to determine the basic risk management strategy template closest to the multidimensional risk assessment vector. The matching process involves calculating the Euclidean or Mahalanobis distance between the input vector and the center point of each threshold range, selecting the basic risk management strategy template associated with the threshold range with the smallest distance.

[0039] In practical implementation, based on the specific scores of each dimension in the multi-dimensional risk assessment vector, the strategy parameters in the closest basic risk management strategy template are fine-tuned to generate a final, suitable risk management strategy. This suitable risk management strategy specifically includes a quantified credit limit allocation scheme, a score-based collateral ratio floating scheme, and a targeted post-loan monitoring scheme. For the credit limit allocation scheme, a credit limit calculation function is established. The input variables of this function include the credit dimension score and the collateral risk dimension score. The function defines how these two scores jointly determine the final credit limit ceiling. An exemplary linear form of the credit limit calculation function is as follows:

[0040] in: This represents the calculated maximum credit limit. This indicates a credit score. This indicates the risk score of the collateral. It is the intercept term. These are the weighting coefficients for the credit score. These are the weighting coefficients for the risk dimension scoring of collateral. (Coefficient) , , The model training phase is based on historical business data. When the multidimensional risk assessment vector is... At that time, , Substituting the values ​​into the function allows you to calculate the specific credit limit. .

[0041] In some embodiments, for a floating loan-to-value (LTV) ratio scheme, a LTV adjustment rule is established. This rule primarily relies on the collateral risk dimension score, which reflects the value stability and liquidity of the collateral under stress scenarios; a lower score indicates higher risk. The LTV adjustment rule specifies a base LTV ratio, which is then adjusted downwards based on the range of the collateral risk dimension score. Simultaneously, the LTV adjustment rule is revised with reference to the related-party risk transmission dimension score, which reflects the likelihood of risk being transmitted from related parties to the loan applicant; a lower score indicates higher transmission risk. The revision method involves further reducing the LTV ratio based on the range of the related-party risk transmission dimension score, in addition to the LTV ratio adjustment determined by the collateral risk dimension score. For example, refer to Table 1 for parameter adjustments in the floating LTV ratio scheme.

[0042] Table 1: Adjustment Table for Mortgage Rate Floating Parameters Collateral risk dimension scoring range Base mortgage rate fluctuation (%) Related Party Transmission Risk Dimension Scoring Range Additional correction float (%) [80,100] +0 [80,100] +0 [60,80) -5 [60,80) -2 [0,60) -10 [0,60) -5 Optionally, when the collateral risk dimension score in the multidimensional risk assessment vector is 70 and the related party transmission risk dimension score is 60, referring to the table above, the collateral risk dimension score of 70 falls within the range [60, 80), corresponding to a base loan-to-value ratio fluctuation of -5%; the related party transmission risk dimension score of 60 falls within the range [0, 60), corresponding to an additional correction fluctuation of -5%. Therefore, the final loan-to-value ratio is reduced by a total of 10 percentage points from the base loan-to-value ratio.

[0043] In some embodiments, for post-loan monitoring schemes, monitoring frequency and key monitoring indicators are set. Monitoring frequency is negatively correlated with the related-party risk transmission dimension score; the lower the related-party risk transmission dimension score, the greater the likelihood of related-party risk transmission, thus requiring a higher monitoring frequency to closely track risk changes. A simple mapping relationship is that for every preset score range decrease in the related-party risk transmission dimension score, the post-loan monitoring frequency increases by one level, for example, from quarterly monitoring to monthly monitoring, and then to bi-weekly monitoring. Key monitoring indicators are selected based on the dimension with the lowest score in the multi-dimensional risk assessment vector, focusing on the dimension with the most prominent risk. If the credit dimension score is the lowest, the key monitoring indicators are set as the new debt situation of the loan applicant, changes in credit reports, and income stability; if the collateral risk dimension score is the lowest, the key monitoring indicators are set as the real-time average listing price in the area where the collateral is located, the transaction cycle of similar properties, and regional real estate market policy trends; if the related-party risk transmission dimension score is the lowest, the key monitoring indicators are set as the operating status of related enterprises, legal proceedings information, and changes in public market credit ratings.

[0044] Understandably, based on the credit limit calculation function, collateral ratio adjustment rules, and the set monitoring frequency and key monitoring indicators, specific strategy parameter values ​​are calculated to fine-tune the basic risk management strategy template. This involves adjusting the specific scores within the multi-dimensional risk assessment vector. , Substitute into the credit limit calculation function Calculate the specific credit limit amount. Scoring based on the risk dimension of the collateral. Risk transmission dimension scoring with related parties Based on the loan-to-value (LTV) ratio adjustment rules, the applicable LTV ratio after the collateral valuation is calculated. For example, if the base LTV ratio is 70%, the final LTV ratio after adjustment is 60%. This is then assessed based on the related-party risk transmission dimension. Given the interval [0, 60), the monitoring frequency is determined to be monthly. Based on the multidimensional risk assessment vector... The lowest-scoring dimension was the related-party risk transmission dimension, with a score of 60. The key monitoring indicators identified were the operating conditions of related enterprises, legal proceedings information, and changes in public market credit ratings. The credit limit calculated above... The specific parameters, such as the final mortgage rate of 60%, the monitoring frequency of "monthly", and the key monitoring indicators of "related enterprise operations, litigation, and rating", are filled into the corresponding parameter positions of the closest basic risk management strategy template, thus generating a highly customized and adaptable risk management strategy for this business.

[0045] In one embodiment of the present invention, the credit limit configuration scheme in the adaptive risk management strategy is parsed. The credit limit configuration scheme is a structured data object containing parameters such as the calculated specific credit limit value, the effective date of the credit limit, the expiration date of the credit limit, the currency of the credit limit, and the applicable product type. Based on these parameters, the system generates a corresponding credit approval instruction. The credit approval instruction is a set of formatted operation commands, explicitly including the "execute credit approval" action, the credit application number, the approved credit limit value, and the approver (or system) identifier. Simultaneously, the system generates a credit limit locking instruction, which is a set of formatted operation commands, explicitly including the "lock credit limit" action, the credit application number, the locked credit limit value, and the customer number. The credit approval instruction and the credit limit locking instruction are sent separately or jointly to the credit approval system and the core accounting system.

[0046] In practical implementation, the system analyzes and adapts the risk management strategy's loan-to-value (LTV) floating scheme. This scheme is a structured data object containing a calculated LTV value, collateral revaluation requirement flags, and contract update requirement flags. Based on these parameters, the system generates a collateral revaluation instruction. This instruction is a set of formatted commands explicitly including the "Initiate Collateral Revaluation" action, the collateral's unique identifier, the target LTV value, and the expected deadline for completion. This instruction is sent to the collateral management system or an external appraisal agency interface. Simultaneously, the system generates a contract terms update instruction. This instruction is also a set of formatted commands explicitly including the "Update Contract Terms" action, the contract agreement number, the fields to be updated, and the updated terms. This instruction is sent to the electronic contract system or legal system to specify the recalculated LTV, loan amount, and other key terms in the standard or electronic contract.

[0047] In some embodiments, the system parses the post-loan monitoring scheme within the adaptive risk management strategy. The post-loan monitoring scheme is a structured data object containing the monitoring frequency, a list of key monitoring indicators, and risk thresholds for each indicator. Based on these parameters, the system generates a data collection frequency setting instruction. This instruction is a set of formatted operation commands, explicitly including the "set data collection task" action, task name, data source type, data collection cycle, and specific data fields to be collected. The monitoring frequency "monthly" is converted to a collection cycle "P1M". This instruction is then sent to the data platform or an external data procurement and scheduling system. The system also generates a risk indicator threshold alarm instruction. This instruction is a set of formatted operation commands, explicitly including the "configure risk alarm rules" action, rule name, monitoring indicator name, warning threshold, alarm level, and a list of alarm recipients. For example, for the key monitoring indicator "number of related enterprise lawsuits," the instruction sets the threshold to "more than 1 new case" and the alarm level to "medium." This instruction is then sent to the risk monitoring and early warning platform. The system generates a periodic review task creation instruction, which is a set of formatted operation commands. Its content clearly includes the "create review task" action, task title, task description, executor role, planned start date, planned completion date, and associated business number. This instruction will be sent to the bank's internal work order system or process management system.

[0048] In some embodiments, all generated instructions are arranged according to the business logic execution order to form a complete set of risk control instructions that can be processed and executed sequentially by the financial business system. The business logic execution order follows the actual credit business process: first, credit line approval and locking; second, collateral management and contract signing; and finally, the establishment of post-loan monitoring. The arrangement process involves placing the previously generated instructions into an instruction list and assigning a sequence number or priority identifier to each instruction. The structured representation of the instruction list can be:

[0049] in: This represents a complete set of risk control instructions. Indicates the first in the instruction list Specific instructions, Indicates the total number of instructions. Each instruction It includes fields such as instruction action, instruction parameters, target system, and execution order. The orchestration logic ensures that the execution order of instructions satisfies business dependencies. For example, the execution of a collateral value revaluation instruction depends on the result of a credit approval instruction, and the execution of a contract term update instruction depends on the result of a collateral value revaluation instruction. The orchestrated risk control instruction set will be encapsulated in standard JSON or XML message format.

[0050] Understandably, once a complete risk control instruction set is generated, it will be sent to the corresponding financial business system for execution via the enterprise service bus or API gateway. Credit approval instructions and quota locking instructions are sent to the credit approval system and core accounting system, triggering online automated approval processes or generating pending tasks on the approval personnel's workbench. Collateral value revaluation instructions are sent to the collateral management system, which automatically creates a pending collateral revaluation task. Contract term update instructions are sent to the electronic contract system, which automatically fills in the new loan-to-value ratio and amount at the designated location in the contract template. Data collection frequency setting instructions are sent to the data platform, which schedules data collection tasks according to the newly set cycle. Risk indicator threshold alarm instructions are sent to the risk monitoring platform, which loads new monitoring rules and begins real-time calculations. Periodic review task creation instructions are sent to the work order system, generating a post-loan review work order in the to-do list of designated personnel. By transforming text strategies into a series of atomic operation instructions that can be directly parsed and executed by machines, an automated closed loop of risk management strategy from analysis and decision-making to business operations is achieved.

[0051] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. 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 be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A risk control management method for real estate and financial businesses based on a big data model, characterized in that, The method includes: Collect multi-dimensional raw data related to real estate and financial businesses; The multi-dimensional raw data is fused and cleaned to generate a structured risk control basic dataset in a unified format; From the structured risk control basic dataset, extract the static profile feature sequence of the credit applicant, the dynamic value feature sequence of the target real estate, and the associated risk feature sequence of related enterprises; The static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence are input into a pre-trained multi-source risk control feature fusion model for interactive modeling to generate a comprehensive risk feature map. Using the comprehensive risk feature map, a hierarchical risk decision network is constructed to generate a multi-dimensional risk assessment vector for the credit applicant. The multi-dimensional risk assessment vector includes a credit dimension score, a collateral risk dimension score, and a related party transmission risk dimension score. Based on the multidimensional risk assessment vector, a dynamic strategy matching engine generates an adaptive risk management strategy, which includes a credit limit configuration scheme, a collateral ratio floating scheme, and a post-loan monitoring scheme. The adaptive risk management strategy is converted into an executable risk control instruction set, and the risk control instruction set is sent to the financial business system for execution.

2. The risk control management method for real estate and financial businesses based on a big data model according to claim 1, characterized in that, The process of fusing and cleaning the multi-dimensional raw data to generate a structured risk control dataset in a unified format includes: The multi-dimensional raw data includes the credit records of the credit applicant, the historical transaction records and market valuation fluctuation series of the target real estate, the operating cash flow data of related enterprises, and legal risk disclosure texts; The credit records are standardized and parsed to extract time-series data on the credit status of individuals or enterprises; Align and denoise the historical transaction records and market valuation fluctuation sequences to generate real estate value evolution trajectory data; Pattern recognition is performed on the aforementioned operational flow data to extract indicators of enterprise operational stability; Keyword extraction and sentiment analysis are performed on the legal risk disclosure text to generate a quantitative risk warning vector; The credit status time series data, real estate value evolution trajectory data, enterprise operation stability indicators and quantitative risk warning vectors are associated, aligned and structured according to a unified spatiotemporal coordinate to form the structured risk control basic dataset.

3. The risk control management method for real estate and financial businesses based on a big data model according to claim 2, characterized in that, From the structured risk control dataset, extract static profile feature sequences of credit applicants, including: Analyze the historical debt records of credit applicants in the structured risk control dataset to generate debt burden characteristics; Analyze the income stability records of credit applicants in the structured risk control dataset to generate income stability features; Identify the associated real estate holding records of the credit applicant in the structured risk control basic dataset and generate asset diversity features; The debt burden characteristics, income stability characteristics, and asset diversity characteristics are time-series encoded to form the static profile feature sequence.

4. The risk control management method for real estate and financial businesses based on a big data model according to claim 3, characterized in that, From the structured risk control dataset, extract the dynamic value feature sequence of the target real estate, including: The value evolution trajectory data of target real estate in the structured risk control basic dataset is analyzed to identify the periodic patterns and long-term trends of value fluctuations. Correlation analysis is performed on the regional market supply and demand data of the target real estate in the structured risk control basic dataset to generate market liquidity prediction features; Integrate the physical attributes of the target real estate to generate property value decay or gain adjustment features; The cyclical patterns are fused and encoded with long-term trends, market liquidity prediction characteristics, and underlying asset value decay or gain adjustment characteristics to form the dynamic value feature sequence.

5. The risk control management method for real estate and financial businesses based on a big data model according to claim 4, characterized in that, The step of inputting the static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence into a pre-trained multi-source risk control feature fusion model for interactive modeling to generate a comprehensive risk feature map includes: The multi-source risk control feature fusion model includes multiple interactive sub-networks, which respectively process the static profile feature sequence, dynamic value feature sequence, and associated risk feature sequence; A feature attention bridging mechanism is established between interactive sub-networks to calculate the risk correlation between the static profile feature sequence and the dynamic value feature sequence, as well as the risk transmission strength of the correlated risk feature sequence to the static profile feature sequence. Based on the calculated risk correlation and risk transmission strength, the various feature sequences are weighted, fused, and information diffused to generate the comprehensive risk feature map. The comprehensive risk feature map uses map nodes to represent risk entities and edge weights to represent risk correlation strength.

6. The risk control management method for real estate and financial businesses based on a big data model according to claim 5, characterized in that, The process of constructing a hierarchical risk decision-making network using the comprehensive risk feature map to generate a multi-dimensional risk assessment vector for the credit applicant includes: The hierarchical risk decision-making network includes a credit risk analysis layer, a collateral risk analysis layer, and an association risk analysis layer; The credit risk analysis layer receives the graph substructures related to the credit of the credit applicant from the comprehensive risk feature graph and generates the credit dimension score. The collateral risk analysis layer receives the substructures of the comprehensive risk feature map that are related to the value and liquidity of the target real estate, and generates the collateral risk dimension score. The associated risk analysis layer receives the graph substructures related to the risk transmission of associated enterprises from the comprehensive risk feature graph, and generates the associated party transmission risk dimension score; The credit dimension score, collateral risk dimension score, and related party transmission risk dimension score are combined to form the multidimensional risk assessment vector.

7. The risk control management method for real estate and financial businesses based on a big data model according to claim 6, characterized in that, The collateral risk analysis layer receives the substructures of the comprehensive risk feature map related to the value and liquidity of the target real estate, and generates the collateral risk dimension score, including: From the graph substructure related to the value and liquidity of the target real estate, the value fluctuation nodes, market liquidity nodes and their connection relationships are analyzed. Under different market stress scenarios, the changes in the state of the value fluctuation nodes affect the market liquidity nodes through their connection relationships, and the potential value decay path and liquidity difficulty of the collateral are calculated. Taking into account the potential value decay path and the difficulty of realization, the risk dimension score of the collateral is calculated through the quantitative scoring model embedded in the collateral risk analysis layer.

8. The risk control management method for real estate and financial businesses based on a big data model according to claim 7, characterized in that, The process of generating adaptive risk management strategies based on the multidimensional risk assessment vector through a dynamic strategy matching engine includes: The multidimensional risk assessment vector is input into the dynamic strategy matching engine; The dynamic strategy matching engine presets multiple basic risk management strategy templates, and each basic risk management strategy template is associated with a risk vector threshold range; The multidimensional risk assessment vector is matched with the threshold range of each risk vector to determine the closest basic risk management strategy template; Based on the specific scores of each dimension in the multidimensional risk assessment vector, the strategy parameters in the closest basic risk management strategy template are fine-tuned to generate the adaptive risk management strategy. The adaptive risk management strategy specifically includes a quantified credit limit allocation scheme, a score-based collateral ratio floating scheme, and a targeted post-loan monitoring scheme.

9. A risk control management method for real estate and financial businesses based on a big data model, as described in claim 8, is characterized in that... The step of fine-tuning the strategy parameters in the closest basic risk management strategy template based on the specific scores of each dimension in the multidimensional risk assessment vector to generate the adaptive risk management strategy includes: For the credit limit allocation scheme, a credit limit calculation function is established, and the input variables of the credit dimension score and the collateral risk dimension score are included; For the aforementioned floating mortgage ratio scheme, a mortgage ratio adjustment rule is established, which is based primarily on the risk dimension score of the collateral and modified with reference to the risk dimension score of related parties. For the post-loan monitoring scheme, a monitoring frequency and key monitoring indicators are set. The monitoring frequency is negatively correlated with the risk dimension score of the related party transmission. The key monitoring indicators are selected based on the dimension with the lowest score in the multi-dimensional risk assessment vector. Based on the credit limit calculation function, the collateral ratio adjustment rules, and the set monitoring frequency and key monitoring indicators, specific strategy parameter values ​​are calculated to complete the fine-tuning of the basic risk management strategy template.

10. A risk control management method for real estate and financial businesses based on a big data model, as described in claim 9, is characterized in that... The step of converting the adaptive risk management strategy into an executable risk control instruction set includes: Analyze the credit limit configuration scheme in the adaptive risk management strategy to generate corresponding credit approval instructions and credit limit locking instructions; Analyze the loan-to-value ratio floating scheme in the adaptive risk management strategy, and generate collateral value revaluation instructions and contract terms update instructions; The post-loan monitoring scheme in the adaptive risk management strategy is analyzed to generate instructions for setting data collection frequency, risk indicator threshold alarm instructions, and periodic review task creation instructions. All generated instructions are arranged according to the business logic execution order to form a complete set of risk control instructions that can be processed and executed sequentially by the financial business system.

Images