Data observation system based on artificial intelligence

Abstract

The invention discloses a data observation system based on artificial intelligence, and belongs to the field of data observation systems, and the system comprises a data collection module which is used for obtaining the identity information, application behavior data, credit data and biological characteristic data of a personal loan subject, the business information, the financial data, the tax data, the associated enterprise data and the fund flow data of the enterprise loan main body are acquired; the correlation analysis module is used for constructing a relation graph based on a graph calculation technology and carrying out cross-dimension correlation mapping on the collected data; the abnormity identification module carries a pre-trained AI classification model and identifies personal malicious loan behaviors, enterprise malicious debt evasion behaviors and career debt dormitory features; and the early warning response module is used for generating a graded early warning signal according to the abnormal recognition result and triggering a corresponding disposal instruction. According to the invention, omnibearing and dynamic observation of personal and enterprise loan subjects is realized, various financial risk behaviors are accurately identified, and graded early warning and intelligent response are carried out.

Description

Technical Field

[0001] This invention relates to the field of data observation systems, specifically to data observation systems based on artificial intelligence. Background Technology

[0002] In financial lending, malicious lending, debt evasion, and organized financial fraud (such as professional debtors and shell companies committing loan fraud) are the main risks faced by financial institutions. Traditional risk control systems rely heavily on rule engines based on static, single-point data (such as central bank credit reports and manually reviewed materials), which has significant drawbacks: First, the rule system is rigid and struggles to identify complex and ever-changing new fraud patterns; second, information silos are severe, making it impossible to effectively integrate and analyze cross-institutional and cross-dimensional information about borrowers; third, post-loan fund monitoring is weak, making it difficult to effectively track whether fund flows are compliant; and finally, there is a lack of predictive ability for potential risks, with most measures being reactive, leading to significant losses for financial institutions.

[0003] With the development of artificial intelligence technology, some solutions have attempted to introduce machine learning models for risk identification, but most solutions are still limited to a single scenario (such as anti-fraud or credit assessment only), lacking a comprehensive and systematic dynamic observation solution that integrates data collection, correlation analysis, multi-scenario identification, early warning response and model iteration.

[0004] Therefore, those skilled in the art have provided artificial intelligence-based data observation systems to address the problems raised in the background section. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based data observation system that enables comprehensive and dynamic observation of individual and corporate borrowers, accurately identifies various financial risk behaviors, and provides tiered early warning and intelligent response, thereby constructing a proactive, intelligent, and efficient financial risk control closed loop to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Artificial intelligence-based data observation systems include: The data collection module is used to acquire the identity information, application behavior data, credit data and biometric data of individual borrowers, as well as the business registration information, financial data, tax data, related enterprise data and fund flow data of corporate borrowers. The correlation analysis module uses graph computing technology to construct a relationship graph containing borrowers, devices, addresses, related persons, and enterprise nodes, and performs cross-dimensional correlation mapping on the collected data; The anomaly detection module, equipped with a pre-trained AI classification model, identifies individual malicious loan behavior, corporate malicious debt evasion behavior, and characteristics of professional debtors. The early warning response module generates tiered early warning signals and triggers corresponding handling instructions based on the anomaly identification results; both the correlation analysis module and the anomaly identification module are iteratively optimized through deep learning models to achieve dynamic identification and early warning of fraudulent behavior.

[0007] As a further aspect of the present invention: the anomaly identification module is configured with a first identification unit for identity fraud and fraudulent application behavior in malicious loans at the individual level, the first identification unit comprising: The identity verification subunit is used to compare the collected ID card information with the data in the public security household registration system to verify the authenticity of the identity. The biometric verification subunit collects the applicant's biometric data through micro-expression recognition and voiceprint recognition technologies, and matches it with the historical biometric data associated with the identity information; The identification logic of the first identification unit is as follows: when the information of the ID card is inconsistent or the biometric matching degree is lower than the preset threshold, it is determined to be a risk of identity fraud and triggers a first-level warning.

[0008] As a further aspect of the present invention: the anomaly identification module is configured with a second identification unit for individual-level financial fraud and fraudulent application activities, the second identification unit comprising: The application frequency statistics sub-unit provides real-time statistics on the number and amount of loan applications submitted by individual borrowers to different financial institutions within a preset time window. The document verification sub-unit verifies the consistency between the applicant's salary slips, proof of employment, and social security and housing provident fund payment records through the government data platform; The second identification unit is equipped with an abnormal frequency judgment model. When the number of applications exceeds a preset threshold and the number of applicant institutions is greater than the institution threshold, or the material verification inconsistency rate exceeds the material threshold, it is judged as a risk of financial fraud or false material application and a secondary warning is generated.

[0009] As a further aspect of the present invention: the anomaly identification module is configured with a third identification unit for shell company loan activities at the enterprise level, the third identification unit comprising: The business status verification subunit collects the company's water and electricity payment records, recruitment information, and tax declaration data to determine whether the company has actual business activities. The material traceability subunit tracks the transaction records of upstream and downstream enterprises corresponding to the trade contracts submitted by enterprises when applying for loans, and verifies the authenticity of the contracts; The third identification unit integrates operational indicators and material authenticity data through a logistic regression model. When the missing rate of operational activity indicators exceeds 80% and there is no corresponding real transaction in the contract, it is determined to be a shell company loan risk, triggering a level three warning.

[0010] As a further aspect of the present invention: the anomaly identification module is configured with a fourth identification unit for enterprise-level misappropriation of loan funds, the fourth identification unit comprising: The fund flow tracking sub-unit uses blockchain technology to record every step of the loan fund transfer, forming a traceable fund chain; The purpose comparison sub-unit matches the actual flow of funds with the purpose agreed upon in the loan application (such as procurement or project investment); The fourth identification unit sets a deviation index for the use of funds. When funds flow into high-risk areas such as the stock market or real estate within 24 hours after disbursement, or flow to accounts unrelated to the agreed purpose, it is determined to be a risk of misappropriation of funds, generating a level four warning and pushing a fund freezing instruction.

[0011] As a further aspect of the present invention: the anomaly identification module is configured with a fifth identification unit for enterprise-level fraudulent activities such as cash grabs, bankruptcy fraud, and cross-guarantee loan fraud by related enterprises. The fifth identification unit includes: The asset transfer analysis subunit tracks the transfer path of a company's core assets (such as equipment and equity) over the past six months through a correlation graph, and identifies abnormal asset transfers to related parties. The guarantee chain identification sub-unit analyzes the guarantee relationships between multiple enterprises, constructs the guarantee network, and calculates the network vulnerability. When it is detected that the transfer of a company's core assets accounts for more than 60% of its total assets without a reasonable business reason, or when the guarantee chain identification subunit constructs a guarantee network by analyzing the guarantee relationships between multiple companies and calculates the network vulnerability, and finds that the probability of default transmission under the state of a single company's default is ≥80% and the predicted number of affected companies is ≥3, it is judged as a risk of malicious debt evasion and triggers a level 5 warning. The network vulnerability is calculated by the guarantee chain identification subunit by integrating the company's debt repayment ability coefficient, the proportion of guarantee amount, and the dependence of related companies. Specifically, it is to weight and quantify the credit rating, asset-liability ratio, and guarantee liability amount of each node company in the guarantee network to clarify the risk transmission path and impact scope to other related nodes after a single node defaults.

[0012] As a further aspect of the present invention: the anomaly identification module is configured with a sixth identification unit for professional debtors, the sixth identification unit comprising: The network mining subunit calculates the risk correlation between nodes based on the relationship graph. The risk correlation calculation formula is: R=α×Nc / Nt+β×Mc / Mt, where R is the risk correlation, α and β are weight coefficients (α+β=1), Nc is the number of applicants sharing the same device / phone number, Nt is the total number of applicants, Mc is the number of emergency contacts that are reused, and Mt is the total number of emergency contacts. The group feature extraction subunit identifies applicant clusters with concentrated residential addresses and unified operating patterns; when R≥0.7, it is determined that there are professional debtor intermediaries or groups, triggering a level 6 warning and marking all related applicants.

[0013] As a further aspect of the present invention: the data acquisition module also includes a behavioral feature acquisition subunit, used to record the operational behavior data of the individual borrower during the APP application process, including typing speed, number of form modifications, and term reading time; The anomaly identification module is configured with a seventh identification unit for the operational behavior data. The seventh identification unit learns the behavior patterns of normal users through a convolutional neural network model and calculates the similarity between the applicant's behavior and the typical behavior of a professional debtor. The similarity calculation formula is: S=γ×Ds / Dt+δ×Ts / Tt, where S is the behavior similarity, γ and δ are weight coefficients (γ+δ=1), Ds is the deviation value of the operation time between the applicant and the typical behavior, Dt is the average time of the typical behavior, Ts is the deviation value of the number of modifications between the applicant and the typical behavior, and Tt is the average number of modifications of the typical behavior. When S≥0.85, the applicant is identified as a suspected professional debtor and a manual review instruction is sent.

[0014] As a further embodiment of the present invention: the early warning response module includes: The entrusted payment verification submodule performs entrusted payment verification on large loan funds to ensure that the funds are directly transferred to the counterparty's account as agreed in the contract, rather than the applicant's personal account; The abnormal transaction analysis submodule monitors the fund transfer behavior of the account after the loan is issued in real time and identifies abnormal transfer patterns such as "small amount multiple transactions" and "centralized transfer to personal account"; When evasive behavior or abnormal transfer patterns of entrusted payments are detected, an early warning is immediately triggered and a notification for early loan recovery is generated.

[0015] As a further aspect of the present invention, it also includes a model iteration and penalty linkage module, wherein the model iteration and penalty linkage module includes: The predictive early warning subunit constructs an LSTM prediction model based on the company's operating indicators and public opinion information to predict the risk of business deterioration three months before the company officially defaults. The blacklist sharing sub-unit will synchronize the characteristic information (after hash processing) of identified professional debtors and malicious debt evasion enterprises to the industry fraud risk information sharing platform; The model iteration and punishment linkage module retrains the AI ​​classification model quarterly based on newly identified fraud case data, updates the identification features and thresholds, and simultaneously links with the judicial system to push fraud clues, thus achieving a closed loop of technical identification and legal punishment.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By constructing a multi-module collaborative intelligent risk control system, the system achieves full-process, dynamic, and accurate identification and early warning of various malicious lending and debt evasion behaviors at the individual and corporate levels (including identity fraud, financial fraud, shell company loan fraud, fund misappropriation, bankruptcy fraud, and professional debtor gangs). The system relies on technologies such as graph computing, biometric recognition, deep learning, and blockchain to achieve cross-dimensional data correlation analysis, automatic fraud pattern mining, and quantitative risk assessment. This significantly improves financial institutions' ability to detect and intervene in highly concealed and professional criminal activities in the early stages. Furthermore, through model iteration optimization and an industry blacklist sharing mechanism, a closed-loop risk control ecosystem linking technical defense and judicial punishment has been formed, fundamentally improving the security of financial assets, reducing credit losses, and demonstrating significant economic and social benefits. Attached Figure Description

[0017] Figure 1 This is a block diagram of an artificial intelligence-based data observation system. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0019] As mentioned in the background section of this application, research has found that existing data observation systems are limited to single scenarios (such as anti-fraud or credit assessment only), lacking a comprehensive and systematic dynamic observation solution that integrates data collection, correlation analysis, multi-scenario identification, early warning response, and model iteration, and thus have certain shortcomings.

[0020] To address the aforementioned shortcomings, this application discloses an artificial intelligence-based data observation system that enables comprehensive and dynamic observation of individual and corporate borrowers, accurately identifies various financial risk behaviors, and provides tiered early warning and intelligent response, thereby constructing a proactive, intelligent, and efficient financial risk control closed loop.

[0021] The following will describe in detail, with reference to the accompanying drawings, how the solution of this application solves the above-mentioned technical problems.

[0022] Please see Figure 1 In this embodiment of the invention, the artificial intelligence-based data observation system includes: a data acquisition module for acquiring the identity information, application behavior data, credit data, and biometric data of individual borrowers, as well as the business registration information, financial data, tax data, related enterprise data, and capital flow data of corporate borrowers; a correlation analysis module for constructing a relationship graph containing nodes of borrowers, devices, addresses, related persons, and enterprises based on graph computing technology, and performing cross-dimensional correlation mapping on the acquired data; an anomaly identification module equipped with a pre-trained AI classification model for identifying malicious loan behavior by individuals, malicious debt evasion behavior by enterprises, and characteristics of professional debtors; and an early warning response module for generating tiered early warning signals and triggering corresponding handling instructions based on the anomaly identification results. Both the correlation analysis module and the anomaly identification module are iteratively optimized through deep learning models to achieve dynamic identification and early warning of fraudulent behavior. This application constructs an integrated AI risk control platform that, through the collaborative work of four modules—data acquisition, correlation analysis, anomaly identification, and early warning response—achieves comprehensive, dynamic, and intelligent monitoring and early warning of fraudulent behavior by individual and corporate borrowers.

[0023] In this embodiment, the anomaly detection module is configured with a first detection unit to target identity fraud and fraudulent loan applications at the individual level. The first detection unit includes: an identity verification subunit, used to compare collected ID card information with data from the public security household registration system to verify the authenticity of the identity; and a biometric verification subunit, which collects the applicant's biometric data through micro-expression recognition and voiceprint recognition technologies and matches it with historical biometric features associated with the identity information. The detection logic of the first detection unit is as follows: when the ID card information comparison is inconsistent, or the biometric matching degree is lower than a preset threshold, it is determined to be an identity fraud risk, triggering a level-one warning. By integrating authoritative data sources (the public security system) and biometric recognition technology, a highly accurate identity verification defense line is constructed, which can effectively identify and block loan applications using false identities or stealing others' identities.

[0024] In this embodiment, the anomaly identification module is equipped with a second identification unit for individual-level financial fraud and fraudulent application behavior. The second identification unit includes: an application frequency statistics subunit, which counts in real time the number and amount of loan applications submitted by an individual borrower to different financial institutions within a preset time window; and a material verification subunit, which verifies the consistency of the applicant's salary slips, employment certificates, and social security and housing provident fund payment records through a government data platform. The second identification unit is equipped with an anomaly frequency judgment model. When the number of applications exceeds a preset threshold and the number of applying institutions exceeds an institution threshold, or the inconsistency rate of material verification exceeds a material threshold, it is determined to be a risk of financial fraud or fraudulent application, generating a level-two warning. By monitoring the spatiotemporal concentration of application behavior and verifying the authenticity of application materials, organized and high-frequency fraudulent application patterns are identified from both behavioral and material dimensions, preventing credit arbitrage and material forgery risks.

[0025] In this embodiment, the anomaly identification module is configured with a third identification unit for shell company loan activities at the enterprise level. This third identification unit includes: an operational status verification subunit, which collects the enterprise's utility bill payment records, recruitment information, and tax declaration data to determine whether the enterprise has actual business operations; and a material tracing subunit, which tracks the upstream and downstream enterprise transaction records corresponding to the trade contracts submitted by the enterprise when applying for a loan to verify the authenticity of the contracts. The third identification unit integrates operational indicators and material authenticity data through a logistic regression model. When the missing rate of operational activity indicators exceeds 80% and the contract lacks a corresponding real transaction, it is determined to be a shell company loan risk, triggering a level-three warning. By analyzing the enterprise's actual operational traces and the authenticity of its trade background, the module penetrates superficial materials to accurately identify shell companies with no actual business operations that rely on falsified trade backgrounds to fraudulently obtain loans, thus preventing fraudulent financing risks in the corporate lending sector.

[0026] In this embodiment, the anomaly identification module is configured with a fourth identification unit to address the misappropriation of loan funds at the enterprise level. This fourth identification unit includes: a fund flow tracking subunit, which uses blockchain technology to record every flow node of loan funds, forming a traceable fund chain; a purpose comparison subunit, which matches the actual flow of funds with the purpose agreed upon in the loan application (such as procurement or project investment); and a fund purpose deviation index. When funds flow into high-risk areas such as the stock market or real estate within 24 hours of disbursement, or flow to accounts unrelated to the agreed purpose, it is determined as a risk of misappropriation, generating a level four warning and sending a fund freezing order. By using blockchain and other technologies, the entire process of loan funds can be monitored traceably. Through intelligent comparison of the agreed purpose with the actual flow, it ensures that credit funds are used for their designated purpose, and promptly detects and prevents funds from flowing into illegal or high-risk areas.

[0027] In this embodiment, the anomaly identification module is equipped with a fifth identification unit for enterprise-level fraudulent activities such as evasion of debt, bankruptcy fraud, and cross-guarantee loan fraud by related enterprises. The fifth identification unit includes: an asset transfer analysis subunit, which tracks the transfer path of the enterprise's core assets (such as equipment and equity) in the past 6 months through a correlation graph to identify abnormal asset transfers to related parties; and a guarantee chain identification subunit, which analyzes the guarantee relationships between multiple enterprises, constructs a guarantee network, and calculates network vulnerability. When it is detected that the proportion of core asset transfers of an enterprise exceeds 60% of its total assets without reasonable commercial reasons, or when the guarantee chain identification subunit constructs a guarantee network and calculates network vulnerability by analyzing the guarantee relationships between multiple enterprises and finds that the probability of default transmission under the state of default of a single enterprise is ≥80% and the predicted number of affected enterprises is ≥3, it is judged as a risk of malicious debt evasion and triggers a level 5 warning. Among them, the network vulnerability is calculated by the guarantee chain identification subunit by integrating the enterprise's debt repayment ability coefficient, the proportion of guarantee amount, and the dependence of related enterprises. Specifically, it is to weight and quantify the credit rating, asset-liability ratio, and guarantee liability amount of each node enterprise in the guarantee network to clarify the risk transmission path and impact scope to other related nodes after the default of a single node. By analyzing the abnormal transfer paths of corporate assets and the vulnerability of guarantee networks through correlation graph analysis, we can provide early warnings of the risks of premeditated asset transfers and debt evasion by enterprises, as well as the systemic risks that may be triggered by the breakage of guarantee chains.

[0028] In this embodiment, the anomaly identification module is configured with a sixth identification unit for professional debtors. This sixth identification unit includes: a network mining subunit, which calculates the risk correlation between nodes based on a relationship graph. The risk correlation calculation formula is: R = α × Nc / Nt + β × Mc / Mt, where R is the risk correlation, α and β are weighting coefficients (α + β = 1), Nc is the number of applicants sharing the same device / phone number, Nt is the total number of applicants, Mc is the number of shared emergency contacts, and Mt is the total number of emergency contacts; and a gang feature extraction subunit, which identifies applicant clusters with concentrated residential addresses and unified operating patterns. When R ≥ 0.7, it is determined that there is a professional debtor intermediary or gang, triggering a level six warning and marking all related applicants. Through quantitative network analysis, debtors controlled by the same intermediary and possessing highly correlated characteristics are accurately identified from a massive number of applicants, thereby dismantling organized loan fraud gangs.

[0029] In this embodiment, the data acquisition module further includes a behavioral feature acquisition subunit, used to record the operational behavior data of individual borrowers during the APP application process, including typing speed, number of form modifications, and term reading time. The anomaly identification module is configured with a seventh identification unit based on this operational behavior data. This seventh identification unit learns the behavioral patterns of normal users through a convolutional neural network model and calculates the similarity between the applicant's behavior and the typical behavior of professional debtors. The similarity calculation formula is: S=γ×Ds / Dt+δ×Ts / Tt, where S is the behavioral similarity, γ and δ are weighting coefficients (γ+δ=1), Ds is the deviation value of the applicant's operation time from typical behavior, Dt is the average time of typical behavior, Ts is the deviation value of the number of modifications between the applicant and typical behavior, and Tt is the average number of modifications of typical behavior. When S≥0.85, the applicant is identified as a suspicious professional debtor, and a manual review instruction is sent. By capturing and analyzing the subtle behavioral characteristics of users during the application process, and using an AI model to determine whether their behavioral patterns are highly similar to known fraud patterns, this serves as a covert and effective auxiliary identification method.

[0030] In this embodiment, the early warning response module includes: a entrusted payment verification submodule, which performs entrusted payment verification on large loan funds to ensure that funds are directly transferred to the counterparty's account as stipulated in the contract, rather than the applicant's personal account; and a transaction anomaly analysis submodule, which monitors the fund transfer behavior of the account after the loan is issued in real time and identifies abnormal transfer patterns such as "multiple small transactions" and "centralized transfers to personal accounts"; when entrusted payment evasion behavior or abnormal transfer patterns are detected, an early warning is immediately triggered and a notification for early loan recovery is generated. Through mandatory entrusted payment and continuous transaction analysis, a "checkpoint" is set up at the fund outflow stage to proactively prevent the misappropriation of funds and to quickly take preservation measures such as freezing or recovery when anomalies are detected.

[0031] This embodiment also includes a model iteration and punishment linkage module, which comprises: a predictive early warning subunit, which constructs an LSTM predictive model based on the company's operating indicators and public opinion information to predict the risk of business deterioration three months before the company officially defaults; and a blacklist sharing subunit, which synchronizes the characteristic information (after hash processing) of identified professional debtors and malicious debt evaders to the industry fraud risk information sharing platform; the model iteration and punishment linkage module retrains the AI ​​classification model quarterly based on newly identified fraud case data, updates the identification features and thresholds, and simultaneously links with the judicial system to push fraud clues, realizing a closed loop of technical identification and legal punishment. By providing early warnings of risks months in advance through predictive models, expanding the scope of defense through industry blacklist sharing, improving identification accuracy through model iteration, and increasing the cost of fraud through judicial linkage, a powerful and deterrent intelligent risk control ecosystem is ultimately formed, capable of continuous learning and collaborative operation.

[0032] This invention constructs a multi-module collaborative intelligent risk control system, enabling full-process, dynamic, and accurate identification and early warning of various malicious lending and debt evasion behaviors at the individual and corporate levels (including identity theft, financial fraud, shell company loan fraud, fund misappropriation, bankruptcy fraud, and professional debtor gangs). The system relies on technologies such as graph computing, biometric recognition, deep learning, and blockchain to achieve cross-dimensional data correlation analysis, automatic fraud pattern mining, and quantitative risk assessment. This significantly improves financial institutions' ability to detect and intervene in highly concealed and professional criminal activities in the early stages. Furthermore, through model iteration optimization and an industry blacklist sharing mechanism, a closed-loop risk control ecosystem linking technical defense and judicial punishment is formed, fundamentally improving the security of financial assets, reducing credit losses, and demonstrating significant economic and social benefits.

[0033] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0034] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence based data observation system characterized in that, Comprise: a data collection module for obtaining identity information, application behavior data, credit data and biological feature data of individual loan subjects, and business information, financial data, tax data, associated enterprise data and fund flow data of enterprise loan subjects; an association analysis module for constructing a relationship graph containing loan subjects, equipment, addresses, associated persons and enterprise nodes based on graph computing technology, and performing cross-dimension association mapping on the collected data; an abnormality identification module for loading a pre-trained AI classification model to identify individual malicious loan behavior, enterprise malicious debt evasion behavior and professional debt characteristics; an early warning response module for generating a graded early warning signal and triggering corresponding disposal instructions according to the abnormality identification result.

2. The artificial intelligence-based data observation system of claim 1, wherein, The abnormality identification module is configured with a first identification unit for identity fraud and false application behavior of individual malicious loan at the individual level, and the first identification unit comprises: an identity verification sub-unit for comparing the collected identity document information with the public security household system data to verify the identity authenticity; a biological feature verification sub-unit for collecting the biological feature data of the applicant through micro-expression recognition and voiceprint recognition technology, and matching the historical biological features associated with the identity information; The identification logic of the first identification unit is that when the identity document information comparison is inconsistent, or the biological feature matching degree is lower than a preset threshold, it is determined as an identity fraud risk, triggering a first-level early warning.

3. The artificial intelligence-based data observation system of claim 2, wherein, The abnormality identification module is configured with a second identification unit for individual financial fraud behavior and false material application behavior, and the second identification unit comprises: an application frequency statistical sub-unit for real-time statistics of the number of loan applications and the amount of individual loan subjects to different financial institutions within a preset time window; a material verification sub-unit for verifying the consistency of the salary stream, work certificate and social security, and public accumulation fund payment records provided by the applicant through the government data platform; The second identification unit loads an abnormal frequency judgment model, and when the number of applications exceeds a preset number threshold and the number of application institutions is greater than an institution threshold, or the material inconsistency rate exceeds a material threshold, it is determined as a financial fraud or false material application risk, generating a second-level early warning.

4. The artificial intelligence-based data observation system of claim 3, wherein, The abnormality identification module is configured with a third identification unit for loan behavior of empty shell companies at the enterprise level, and the third identification unit comprises: an operating state checking sub-unit for collecting the water and electricity payment records, recruitment information and tax declaration data of the enterprise to determine whether the enterprise has actual business activities; a material tracing sub-unit for tracking the upstream and downstream enterprise transaction records corresponding to the trade contract submitted by the enterprise when applying for a loan to verify the contract authenticity; The third identification unit fuses the operating index and material authenticity data through a logistic regression model, and when the missing rate of operating activity indicators exceeds 80% and there is no corresponding real transaction in the contract, it is determined as an empty shell company loan risk, triggering a third-level early warning.

5. The artificial intelligence-based data observation system of claim 4, wherein, The abnormality identification module is configured with a fourth identification unit for the behavior of misappropriating loan funds at the enterprise level, and the fourth identification unit comprises: a fund flow tracking sub-unit for recording each flow node of the loan funds using blockchain technology to form a traceable fund link; a use comparison sub-unit for matching the actual fund flow with the intended use as agreed upon in the loan application; The fourth identification unit sets a fund use deviation degree index, and when the fund flows into a high-risk field or to an account irrelevant to the intended use within 24 hours after the fund is issued, it is determined that there is a fund misappropriation risk, a fourth-level warning is generated, and a fund freezing instruction is pushed.

6. The artificial intelligence-based data observation system of claim 5, wherein, The abnormality identification module is configured with a fifth identification unit for behaviors such as enterprise-level shell companies, bankruptcy fraud, and fraudulent credit insurance among related enterprises. The fifth identification unit includes: An asset transfer analysis subunit traces the transfer path of the core assets of an enterprise within the last six months through a correlation graph to identify abnormal asset transfer to related parties. A guarantee chain identification subunit analyzes the guarantee relationship among multiple enterprises, constructs a guarantee network, and calculates the network vulnerability. When the proportion of core asset transfer of an enterprise exceeds 60% of the total assets without reasonable commercial reasons, or when the guarantee chain identification subunit constructs a guarantee network and calculates the network vulnerability by analyzing the guarantee relationship among multiple enterprises, and the default transmission probability under the single enterprise default state is ≥80% and the number of predicted affected enterprises is ≥3, it is determined that there is a malicious debt evasion risk, triggering a fifth-level warning.

7. The artificial intelligence-based data observation system of claim 6, wherein, The abnormality identification module is configured with a sixth identification unit for professional debtors. The sixth identification unit includes: An associated network mining subunit calculates the risk correlation degree between nodes based on a relationship graph. The risk correlation degree calculation formula is: R = α × Nc / Nt + β × Mc / Mt, where R is the risk correlation degree, α and β are weight coefficients, Nc is the number of applicants sharing the same device / mobile number, Nt is the total number of applicants, Mc is the number of cross-reused emergency contacts, and Mt is the total number of emergency contacts. A gang feature extraction subunit identifies clusters of applicants with concentrated addresses and uniform operation modes. When R ≥ 0.7, it is determined that there is a professional debtor intermediary or gang, triggering a sixth-level warning and marking all associated applicants.

8. The artificial intelligence-based data observation system of claim 7, wherein, The data collection module also includes a behavior feature collection subunit for recording the operation behavior data of individual loan subjects during the application process in the APP, including typing speed, form modification times, and clause reading duration. The abnormality identification module is configured with a seventh identification unit for this operation behavior data. The seventh identification unit learns the behavior patterns of normal users through a convolutional neural network model, calculates the similarity between the applicant's behavior and the typical behavior of professional debtors, and the similarity calculation formula is: S = γ × Ds / Dt + δ × Ts / Tt, where S is the behavior similarity, γ and δ are weight coefficients (γ + δ = 1), Ds is the operation duration deviation value of the applicant and the typical behavior, Dt is the average duration of the typical behavior, Ts is the modification frequency deviation value of the applicant and the typical behavior, and Tt is the average modification frequency of the typical behavior. When S ≥ 0.85, it is determined that the applicant is a suspected object of professional debtors, and an artificial review instruction is pushed.

9. The artificial intelligence-based data observation system of claim 8, wherein, The warning response module includes: A trusted payment verification sub-module performs trusted payment verification on large loan funds to ensure that the funds are directly transferred to the account of the contract counterparty rather than the personal account of the applicant. A flow anomaly analysis sub-module monitors the fund transfer behavior of the account after the loan is issued in real time and identifies abnormal transfer patterns. When the trust payment avoidance behavior or abnormal transfer mode is detected, an early warning is triggered and a notice of early repayment of the loan is generated.

10. The artificial intelligence-based data observation system of claim 9, wherein, The model iteration and punishment linkage module comprises: A predictive early warning subunit constructs an LSTM prediction model based on the operating indicators and public opinion information of the enterprise to predict the risk of business deterioration three months before the enterprise officially defaults; A blacklist sharing subunit synchronizes the feature information of the identified professional debtors and maliciously escaping and escaping debt enterprises to an industry fraud risk information sharing platform; The model iteration and punishment linkage module re-trains the AI classification model according to new identified fraud case data every quarter, updates the identification features and thresholds, and simultaneously links the judicial system to push fraud clues.

Images