Automatic auditing method for housing accumulation fund mobile terminal extraction application

Abstract

The invention discloses an automatic auditing method for a housing accumulation fund mobile terminal extraction application, and the method relates to the field of automatic auditing, and comprises the steps: obtaining the personal information of a user according to the housing accumulation fund extraction application submitted by the user, and verifying the identity of the user through employing a dynamic biological information verification technology; a cross-institution federated learning framework is introduced, a verification model is constructed in combination with multi-party data sources, and multi-dimensional cross verification is performed on user identity information; feature information in the verified user personal information is compared with preset standard feature information; deploying an auditing decision engine, performing risk behavior auditing on the user feature information, and returning auditing suggestions; auditing suggestions are pushed to the user in real time through a mobile terminal service page, a user personal information supplement list is automatically generated for applications needing material supplement or rejection, and the auditing progress is monitored in real time; a block chain technology is applied, and key data of each time of housing public accumulation fund mobile terminal extraction application and user data verification is subjected to uplink storage.

Description

Technical Field

[0001] The present invention relates to the field of automated auditing, and particularly to an automated auditing method for housing provident fund mobile extraction applications. Background Art

[0002] Housing provident fund refers to the long-term housing savings deposited by work units and their on-the-job employees. The housing provident fund is paid monthly by on-the-job employees and their employers in a certain proportion of the employees' personal wages and the total wage bill of the employees, and belongs to individuals. It is used for purchasing, building, or major repairs of self-occupied housing. When employees retire, the principal and interest balance is settled at once and returned to the employees themselves. In the current housing provident fund loan review, it mainly includes two categories: housing consumption and non-housing consumption. Materials need to be prepared according to specific situations and processed through online or offline channels. Among them, offline processing requires bringing ID cards, bank cards, and relevant supporting materials (such as purchase contracts, repayment statements, etc.) to the provident fund management center or cooperative bank branches for processing. Online processing mainly submits applications and relevant materials through the provident fund official website or official APP and then undergoes manual review. Most operations can be credited within 3 working days. There is also automatic review, but mainly uses algorithms to compare and verify the data submitted by users, and the security has not been verified. Regarding the above situation of housing provident fund online review, the following problems exist: 1. The review time is long, and it is impossible to apply for an immediate credit. If there are problems with the materials, a second review may be required; 2. For manual review, there may be missed reviews or incorrect reviews. Automatic review mainly uses face verification, and currently, instant videos generated by AI may bypass the review process; 3. Limited within the provident fund system, it is impossible to cross-verify multiple parties' information; 4. It can only simply generate review opinions of passing or rejection and cannot give targeted suggestions. Summary of the Invention

[0003] In view of the problems shown above, the present invention provides an automated auditing method for housing provident fund mobile extraction applications to solve the above problems.

[0004] An automated auditing method for housing provident fund mobile extraction applications includes the following steps: Obtain the user's personal information based on the housing provident fund extraction application submitted by the user on the mobile service page, and verify the user's identity using dynamic biometric information verification technology; By introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources, and perform multi-dimensional cross-verification on the user identity information without exposing the original data; Compare the characteristic information for housing provident fund applications in the verified user personal information with the preset standard characteristic information; Deploy a machine learning-based audit decision engine to conduct risk behavior audits on user feature information and return audit opinions. Push the audit opinions to users immediately through the mobile service page. For approved applications, the system connects to the financial payment system for provident fund transfer. For applications that require supplementary materials or are rejected, automatically generate a list of user personal information supplements and monitor the audit progress in real time. Apply blockchain technology to store the key data of each mobile housing provident fund withdrawal application and user data verification on the blockchain to form an immutable verification evidence chain.

[0005] Preferably, obtain user personal information based on the housing provident fund withdrawal application submitted by the user on the mobile service page, and use dynamic biometric information verification technology to verify the user's identity, including: After the user submits a housing provident fund withdrawal application through the mobile service page, identify the user personal information in the housing provident fund withdrawal application, and obtain the pre-stored voiceprint and face recognition information of the corresponding user in the housing provident fund database. Use the camera to obtain the user's current facial status and compare it with the pre-stored face recognition information. If it matches, it is determined that the face recognition is passed. Randomly generate a user's real-time verification action, and when the user makes the corresponding action, it is regarded as the real-time verification passed. Randomly generate a string of numbers. When the voiceprint data of the generated numbers read by the user matches the pre-stored voiceprint data, it is regarded as the voiceprint verification passed. Conduct micro-expression detection on the user's real-time video data. If it is detected that the micro-expression does not match the human real expression pattern, the real-time video data is identified as AI-generated video. When the face recognition is passed, the real-time verification is passed, the voiceprint verification is passed, and it is determined that the real-time video data is not AI-generated video, it is determined that the user identity verification is passed; otherwise, it is regarded as the user identity verification failed and needs to be verified again.

[0006] Preferably, by introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources to conduct multi-dimensional cross-verification on the user identity information without exposing the original data, including: Obtain multi-source data based on multiple official databases, construct a unified feature space mapping table, and process the multi-source data into unified data. Use the algorithm under the federated learning framework to construct a vertical federated data matrix according to the unified data, and use encryption algorithms to perform available but invisible processing on the vertical federated data matrix. Determine the standard threshold for dynamic verification settings based on the processed vertical federated data matrix, and supplement and set the standard threshold for spatio-temporal consistency verification of multi-data. Construct the joint verification model based on the standard threshold set by the dynamic verification and the spatio-temporal consistency verification standard threshold; Input the user identity information into the joint verification model. If the user identity information meets the verification standard, it is displayed as verified. If the user identity information does not meet the verification standard, it is displayed as abnormal identity information.

[0007] Preferably, compare the characteristic information used for the housing provident fund application in the verified user personal information with the preset standard characteristic information, including: Perform standardized format conversion on the verified user personal information and extract key characteristic items, and convert them into a structured comparison template; Compare the structured comparison template with the preset standard characteristic information, and determine whether it matches the preset standard characteristic information through a semantic analysis engine. If it matches, it is determined as user characteristic information. If it does not match, it is determined as invalid information; Output the user characteristic information obtained by the match.

[0008] Preferably, deploy an audit decision engine based on machine learning to conduct a risk behavior audit on the user characteristic information and return an audit opinion, including: Use a pre-trained risk identification model to perform intelligent analysis on the obtained user characteristic information. If the risk identification model identifies it as passing the audit, return an audit opinion of passing; Deploy a machine learning framework. When the divergence degree of the pre-trained risk identification model exceeds the threshold, trigger an arbitration function based on machine learning to generate an audit opinion of rejection or supplementary materials; Package the pre-trained risk identification model and the machine learning framework to obtain an audit decision engine; Input the user characteristic information into the audit decision engine, automatically traverse the decision nodes according to the matching degree of each user characteristic information, and respectively conduct a determination on the material integrity and condition compliance to generate an audit opinion for this user.

[0009] Preferably, push the audit opinion to the user immediately through the mobile service page. For applications that pass, the system connects to the fiscal payment system for provident fund appropriation. For applications that require supplementary materials or are rejected, automatically generate a user personal information supplement list and monitor the audit progress in real time, including: Perform result annotation on the information in the housing provident fund withdrawal application according to the audit opinion output by the audit decision engine and push the information on the mobile service page; Classify the result annotation to generate a first classification result with an audit opinion of passing, a second classification result with an audit opinion of rejection, and a third classification result with an audit opinion of supplementary materials; When the result is the first category result, the financial payment system of the provident fund management center is connected according to the housing provident fund withdrawal application content and user identity information provided by the user to transfer the provident fund according to the specific amount of the housing provident fund withdrawal application; When the result is the second classification result, the rejection reason of the housing provident fund withdrawal application and the user's personal information supplement list are generated and pushed to the user on the mobile service page; When the result is the third category result, the specific items of the housing provident fund withdrawal application that are missing materials are marked, and a user personal information supplement list that needs to be supplemented is generated and pushed to the user on the mobile service page; The review opinion is pushed on the mobile service end, and the status is updated in real time.

[0010] Preferably, blockchain technology is used to store key data of each housing provident fund mobile withdrawal application and user data verification on the chain, forming an unalterable verification evidence chain, including: Determine the user's personal information, the process information of the multi-dimensional cross-validation process of the user's identity information, and the review opinion as key data in the review process of the housing provident fund withdrawal application; Packing the key data into blocks, and generating a first hash value according to the blocks using a hash algorithm; The operation log generated by each operation of the mobile terminal is used to generate a second hash value using a hash algorithm; Recording the first hash value and the second hash value together with relevant timestamp information on a blockchain; A multi-level permission query channel is established, users can view personal business evidence through private keys, regulatory agencies can trace the complete chain of evidence with regulatory keys, and each participating department can access relevant chain segment data according to permissions.

[0011] Preferably, the method further comprises generating a user profile based on the user's financial behavior information and interpersonal relationships, and identifying risky behaviors of the user: By accessing financial platforms such as housing provident fund management systems, banking systems, and third-party payment platforms, the user's financial behavior data within the financial platforms is collected based on the user's personal information; Preprocessing the financial behavior data, removing scattered consumption records below a preset limit, and aligning and sorting the financial behavior data according to time series; Conduct time series analysis on the pre-processed financial behavior data, identify abnormal fluctuations in funds in the financial behavior data, establish a dynamic repayment ability prediction model, determine the user's repayment ability, and determine the analysis results as the user's financial profile; Identify contacts who have large capital flows and frequent transactions in the user's financial behavior data, and identify the contacts as key contacts; Based on the key contacts, use graph computing technology to construct a user social connection graph, determine the user's risk transmission path, establish a relationship network influence evaluation model, determine high-risk contacts in the social network, and determine the evaluation result as the user's social portrait; Use the user's financial portrait and the user's social portrait to jointly generate the user's portrait; Determine the possible risk tags for each item in the user portrait, and use an ensemble learning algorithm to construct a user portrait risk detection model based on the risk tags; Use the user portrait risk detection model to identify risk behaviors of the user portrait to be detected, and output the risk behavior detection result of the user.

[0012] Preferably, the method further includes that when a user withdraws housing provident fund for a housing purchase loan, conduct credit investigation verification on the user through a third-party institution, conduct user risk assessment based on the credit investigation verification result, and issue a risk assessment report to the housing provident fund loan institution: Conduct keyword detection on the user's housing provident fund withdrawal application, and determine that credit investigation verification is required for the user when it is detected that the purpose of the user's housing provident fund withdrawal is for a housing purchase loan; Determine the specific credit investigation items that the user needs to conduct credit investigation verification based on the user's personal information; Obtain the user's credit information from the People's Bank of China credit investigation institution, and extract the specific credit investigation items that need to be verified by a third party in the People's Bank of China credit information for the first validity verification; Obtain the user's private credit information from a private credit investigation institution, and extract the specific credit investigation items that need to be verified by a third party in the private credit information for the second validity verification; Based on the first validity verification result and the second validity verification result, confirm the corresponding risk assessment result of each specific credit investigation item; Generate a housing provident fund loan risk assessment report verified by a third party based on the risk assessment result; Obtain the risk assessment abnormal items in the housing provident fund loan risk assessment report verified by a third party, determine the corresponding abnormal reasons for each risk assessment abnormal item and the specific score value of the risk assessment result of the risk assessment abnormal item; Determine the corresponding solution based on the abnormal reason corresponding to each risk assessment abnormal item, and generate a specific risk decision according to the corresponding solution; Evaluate the risk control threshold of each specific risk decision, and determine the risk control threshold of each specific risk decision; Use the risk control threshold corresponding to each specific risk decision for abnormal items in the risk assessment to judge the specific score value corresponding to each risk assessment abnormal item in the housing provident fund loan risk assessment report verified by a third party, and confirm whether the specific risk decision corresponding to the risk assessment abnormal item can control the risk of the risk assessment abnormal item; Output the judgment result as the feasibility analysis result of risk control for each risk assessment abnormal item; Generate and output the reference judgment result of the housing provident fund loan credit for this customer based on the feasibility analysis result of risk control for each risk assessment abnormal item.

[0013] Through the above technical means, the present invention has the following beneficial effects: 1) Through this method, real-time review and completion of the housing provident fund can be achieved. If the review fails, opinions can also be given immediately, which is convenient for users and enhances the practicability of the invention; 2) Using the dynamic biological information verification technology to verify the user's identity solves the problem of user identity verification on the mobile side, and micro-expression detection can be performed, which can prevent AI from generating verification videos, improve user security, and reduce the risk of the provident fund being withdrawn by others; 3) Using the cross-institutional federated learning framework and jointly constructing a joint verification model with multiple data sources can perform multi-dimensional cross-verification on the user's identity information, enhancing the accuracy of the review; 4) The review opinions can be immediately pushed to the user through the mobile service page. For applications that need to be rejected for supplementary materials, review opinions can be automatically generated, and the review progress can be monitored in real time, enhancing the pertinence of the review opinions and improving the user experience.

[0014] Other features and advantages of the present invention will be described in the following specification, and part of them will become obvious from the specification or be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification and the drawings.

[0015] The technical solution of the present invention will be further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention.

[0017] Figure 1 It is a working flowchart of an automated review method for housing provident fund mobile extraction applications provided by the present invention; Figure 2Another flowchart of the automated review method for mobile extraction applications of housing provident funds provided by the present invention; Figure 3 Another flowchart of the automated review method for mobile extraction applications of housing provident funds provided by the present invention. Detailed implementation manners

[0018] Here, exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0019] Housing provident fund refers to the long-term housing savings deposited by work units and their on-the-job employees. The housing provident fund is paid monthly by on-the-job employees and their work units in a certain proportion of the employees' personal wages and the total wages of the employees, and belongs to individuals. It is used for purchasing, building, and major repairs of self-occupied housing. When employees retire, the principal and interest balance is settled at one time and returned to the employees themselves. In the current housing provident fund loan review, it mainly includes two categories: housing consumption and non-housing consumption. Materials need to be prepared according to specific situations and processed through online or offline channels. Among them, offline processing requires bringing an ID card, bank card, and relevant supporting materials (such as a housing purchase contract, repayment statement, etc.) to the provident fund management center or the cooperative bank network for processing. Online processing mainly submits applications and relevant materials through the provident fund official website or the official APP for manual review. Most operations can be credited within 3 working days. There is also automatic review, but mainly uses algorithms to compare and verify the data submitted by users, and the security has not been verified. Regarding the above situation of housing provident fund online review, the following problems exist: 1. The review time is long, and it is impossible to apply for an immediate credit. If there are problems with the materials, a second review may be required; 2. Manual review may result in missed reviews or incorrect reviews. Automatic review mainly uses face verification, and currently, AI-generated instant videos may bypass the review process; 3. Limited within the provident fund system, it is impossible to cross-verify multiple-party information 4. It can only simply generate review opinions of approval or rejection and cannot give targeted suggestions.

[0020] In view of the problems shown above, the present invention provides an automated review method for mobile extraction applications of housing provident funds to solve the above problems.

[0021] An automated review method for mobile extraction applications of housing provident funds, as Figure 1 shown, includes the following steps: Step S101: Obtain the user's personal information based on the housing provident fund withdrawal application submitted by the user on the mobile service page, and verify the user's identity using dynamic biometric information verification technology; In some embodiments, the user's personal information mainly includes the personal information used by the user for housing provident fund withdrawal, mainly including the user's name, ID number, housing purchase or rental information, bank card number, provident fund account, etc.; the dynamic biometric information verification technology mainly uses the user's biometric information, including voiceprint information, face recognition information, etc. for dynamic recognition to ensure that the withdrawing user matches the input user identity information; Step S102: By introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources, and perform multi-dimensional cross-verification on the user identity information without exposing the original data; In some embodiments, the cross-institutional federated learning framework allows multiple participants to collaboratively train a model without sharing data. Each participant trains the model locally and only exchanges model parameters instead of the original data, which can protect data privacy and comply with privacy protection regulations; the multiple data sources include data that can be used for housing provident fund review, such as identity information in the public security department database, financial data in the bank database, payment records in the social security department database, etc.; the multi-dimensional cross-verification is to perform joint calculations on multiple features in the encrypted space to ensure that the three-factor spatio-temporal logic is consistent and there are no abnormalities in the fund flow; Step S103: Compare the feature information used for housing provident fund application in the verified user personal information with the preset standard feature information; In some embodiments, the verified user personal information is the user personal information verified for authenticity to ensure the accuracy of the comparison. The comparison with the preset standard feature information mainly needs to map the user data to the preset standard through feature mapping, such as the user's marital status, payment duration, etc.; Step S104: Deploy an audit decision engine based on machine learning to conduct a risk behavior audit on the user feature information and return an audit opinion; In some embodiments, the machine learning is an entire model construction and encapsulation process including data preprocessing, feature engineering, and model training. The audit decision engine is a decision engine that can audit the user feature information after encapsulation. The audit opinions include approval, rejection, and the need to supplement materials; Step S105: Instantly push the audit opinion to the user through the mobile service page. For approved applications, the system connects to the fiscal payment system for provident fund transfer. For applications that require supplementary materials or are rejected, an automatic user personal information supplementary list is generated, and the audit progress is monitored in real time; In some embodiments, the financial payment system is the payment system of the housing provident fund management department, the review opinion is a passing prompt for each feature item of the user, the user personal information supplement list is each item for which the user needs to supplement materials, and the real-time monitoring of the review progress is to monitor the application review progress of the user in real time and make corresponding displays on the mobile terminal page; Step S106: Apply blockchain technology to store the key data of each housing provident fund mobile extraction application and user data verification on the blockchain, forming an immutable verification evidence chain.

[0022] In some embodiments, the blockchain technology is mainly used to make the verification evidence chain immutable and easy to verify. The key data of the secondary housing provident fund mobile extraction application and user data verification include the user's personal information, provident fund application information, work flow information of the review system, etc. The on-chain storage means packing the data into blocks and uploading them to the blockchain.

[0023] The working principle of the above technical solution is as follows: Obtain the user's personal information according to the housing provident fund extraction application submitted by the user on the mobile terminal service page, and use dynamic biometric information verification technology to verify the user's identity; Introduce a cross-institutional federated learning framework to conduct multi-dimensional cross-verification of the user's identity information; Compare the characteristic information used for housing provident fund application in the user's personal information with the preset standard characteristic information; Use an audit decision engine based on machine learning to conduct risk behavior audits on the user's characteristic information and return audit opinions; Push the audit opinions to the user immediately through the mobile terminal service page and monitor the audit progress in real time; Apply blockchain technology to store the key data of each housing provident fund mobile extraction application and user data verification on the blockchain, forming an immutable verification evidence chain.

[0024] The beneficial effects of the above technical solution are as follows: Obtain the user's personal information according to the housing provident fund withdrawal application submitted by the user on the mobile service page, and use dynamic biometric information verification technology to verify the user's identity, which enhances the accuracy and effectiveness of identity verification in this method; Introduce a cross-institutional federated learning framework to conduct multi-dimensional cross-verification on the user's identity information, verify the user's identity information through multi-institutional data, and enhance security; Compare the characteristic information used for housing provident fund applications in the user's personal information with the preset standard characteristic information, and use an audit decision engine based on machine learning to conduct risk behavior audits on the user's characteristic information and return audit opinions, which can quickly and accurately generate audit opinions, enhance the user experience, and increase the practicality of the invention; Push the audit opinion to the user immediately through the mobile service page and monitor the audit progress in real time, which enhances the usage experience of the mobile page; Apply blockchain technology to store the key data of each housing provident fund mobile withdrawal application and user data verification on the chain to form an immutable verification evidence chain, which enhances the security and verifiability of the data.

[0025] In one embodiment, as Figure 2 shown, obtain the user's personal information according to the housing provident fund withdrawal application submitted by the user on the mobile service page, and use dynamic biometric information verification technology to verify the user's identity, including: Step S201: After the user submits a housing provident fund withdrawal application through the mobile service page, identify the user's personal information in the housing provident fund withdrawal application, and obtain the pre-stored voiceprint and face recognition information of the corresponding user in the housing provident fund database; In some embodiments, when the user submits a withdrawal application through the housing provident fund mobile service page, the system first automatically parses the personal identity information in the application form (including keyword fields such as name, ID number, and mobile phone number), and then calls the biometric verification interface in the housing provident fund database to obtain the reference voiceprint data (including specific frequency band acoustic characteristic parameters) and high-precision face recognition template (including three-dimensional feature point data) pre-stored when the user opened the account; Step S202: Use the camera to obtain the user's current facial status and compare it with the pre-stored face recognition information. If it matches, it is determined that the face recognition is passed; In some embodiments, the system calls the mobile camera to collect the user's facial image in real time. After verifying that it is not forged by a photo or video through live detection technology, extract the key face feature points and conduct intelligent comparison with the pre-stored biometric template in the database. When the similarity reaches the security threshold, it is determined that the face recognition is passed; Step S203: Randomly generate a user's real-time verification action, and when the user makes the corresponding action, it is considered that the real-time verification is passed; In some embodiments, the system randomly generates dynamic verification instructions (such as blinking, nodding). The user needs to complete the specified actions in front of the camera. The authenticity of the live body is verified through real-time action trajectory analysis. If the biological characteristics are met, it is determined that the real-time verification is passed; Step S204: Randomly generate a string of numbers. When the voiceprint data read by the user matches the pre-stored voiceprint data, it is regarded as the voiceprint verification passed; In some embodiments, the system dynamically generates a 4- to 6-digit random number. The user needs to clearly read aloud this string of numbers. The voice spectrum features are extracted through a voiceprint analysis algorithm and compared with the pre-stored voiceprint model for acoustic parameter comparison. When the confidence level reaches the security threshold, it is determined that the voiceprint verification is passed; Step S205: Perform micro-expression detection on the user's real-time video data. If it is detected that the micro-expression does not match the human real expression pattern, the real-time video data is identified as an AI-generated video; In some embodiments, the system uses the front camera to collect the user's facial video stream in real time, and uses a micro-expression analysis engine to monitor subtle features such as eye muscle tremors, slight mouth movements, and facial blood flow changes. By comparing the physiological response patterns of human natural expressions (such as the blink frequency of 0.2 - 0.4 seconds per blink, the micro-expression duration of 1 / 25 to 1 / 5 seconds and other reference parameters), when features such as mechanical repetition, abnormal timing, or lack of bioelectrical signals are detected in the facial micro-movements, AI detection is automatically triggered. The pixel change rules between video frames, light and shadow consistency, and micro-expression dynamics features are comprehensively evaluated. If the matching degree with the known deepfake video synthesis mode (such as abnormal frame transition smoothness, lack of micro-expression, etc.) exceeds the risk threshold, it is determined as an AI-generated video and the verification process is immediately terminated; Step S206: When the face recognition is passed, the real-time verification is passed, the voiceprint verification is passed, and it is determined that the real-time video data is not an AI-generated video, it is determined that the user identity verification is passed; otherwise, it is regarded as the user identity verification not passed and needs to be verified again.

[0026] In some embodiments, a multi-modal biometric fusion verification mechanism is adopted. It is determined that the user identity verification is passed if and only if the following conditions are simultaneously met: 1) The face recognition confirms that the matching degree between the real-time collected facial features and the pre-stored template reaches the security threshold; 2) The dynamic verification action analysis verifies that the user correctly completes the randomly generated live body detection instruction; 3) The voiceprint recognition engine determines that the acoustic features of the read numbers are consistent with the registered voiceprint; 4) The AI forgery detection confirms that the real-time video data conforms to the human natural expression features and has no deepfake traces. If any link verification fails, the security fuse mechanism is triggered, and the current session data is automatically cleared and the user is forced to re-verify from the initial link.

[0027] The beneficial effects of the above technical solution are as follows: Identify the user's personal information in the housing provident fund withdrawal application, obtain the pre-stored voiceprint and face recognition information of the corresponding user in the housing provident fund database, use the camera to obtain the user's current facial state, compare it with the pre-stored face recognition information, randomly generate the user's real-time verification action, randomly generate a string of numbers for voiceprint verification, and verify the user's identity through multiple verification means, ensuring the security of funds; perform micro-expression detection on the user's real-time video data. If it is detected that the micro-expression does not match the human real expression pattern, the real-time video data is identified as an AI-generated video, adding AI-generated video verification to prevent current new technical means and improve system security. When face recognition passes, real-time verification passes, voiceprint verification passes, and it is determined that the real-time video data is not an AI-generated video, it is determined that the user identity verification is passed, enhancing the accuracy and effectiveness of the identity verification method of this method.

[0028] In one embodiment, as Figure 3 shown, by introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources, and perform multi-dimensional cross-verification on the user identity information without exposing the original data, including: Step S301: Obtain multi-source data based on multiple official databases, construct a unified feature space mapping table, and process the multi-source data into unified data; In some embodiments, the system retrieves information from authoritative databases such as public security, social security, and taxation in real time through cross-departmental data interfaces, uses federated learning technology to construct standardized feature vectors, converts heterogeneous data into a feature matrix with consistent dimensions and unified formats, and forms a digital identity map that can be cross-verified; Step S302: Use the algorithm under the federated learning framework to construct a vertical federated data matrix according to the unified data, and perform an available but invisible process on the vertical federated data matrix using an encryption algorithm; In some embodiments, based on the secure multi-party computing protocol, cross-departmental feature data is constructed into a distributed vertical federated matrix under the federated learning framework, and ciphertext calculation of data elements is realized through homomorphic encryption and differential privacy technologies, ensuring that all participating parties complete joint modeling and collaborative analysis on the premise that the original data is invisible; Step S303: Determine the standard threshold for dynamic verification settings based on the processed vertical federated data matrix, and supplement and set the standard threshold for spatio-temporal consistency verification of multiple data; In some embodiments, the content analyzes the cross-departmental data feature distribution through a federated learning model, dynamically generates verification thresholds including parameters such as confidence intervals and behavior pattern similarities, and supplements and sets multi-dimensional verification criteria such as geographical location rationality and operation time sequence continuity based on spatio-temporal association rules; Step S304: Construct the joint verification model based on the standard threshold set by the dynamic verification and the spatio-temporal consistency verification standard threshold; In some embodiments, the joint verification model is expressed as constructing a multi-dimensional joint verification model by fusing the dynamic verification threshold and the spatio-temporal consistency rule, integrating the federated learning technology, and realizing the automatic real-time cross-verification of cross-department data; Step S305: Input the user identity information into the joint verification model. If the user identity information meets the verification standard, it is displayed as verified passed. If the user identity information does not meet the verification standard, it is displayed as abnormal identity information.

[0029] In some embodiments, when a user needs to perform identity verification, the joint verification model automatically performs multi-dimensional verification using the collected multi-modal biometric features (including face, voiceprint, real-time behavior data) and the cross-department federated data matrix: First, compare the matching degree of the biometric features with the reference data. Second, analyze the rationality of the operation behavior and the spatio-temporal trajectory. Finally, evaluate the overall risk. If all indicators meet the preset dynamic threshold standard, the system generates a digital verification certificate with a timestamp and returns a passed result. If any verification dimension exceeds the tolerance range, an abnormal warning is immediately triggered.

[0030] The beneficial effects of the above technical solutions are as follows: Obtain multi-source data and process it into unified data, which can standardize the data, enhance the effectiveness of the data. Use the federated learning framework to construct a vertical federated data matrix and perform encryption processing using encryption algorithms, which enhances the security of the data. Perform normalization operations on multi-source data; Determine the standard threshold set by the dynamic verification based on the vertical federated data matrix, construct a joint verification model, which enhances the accuracy of the verification model; Input the user identity information into the joint verification model. If the user identity information meets the verification standard, it is displayed as verified passed. If the user identity information does not meet the verification standard, it is displayed as abnormal identity information, which enhances the review efficiency and improves the practicability of this method.

[0031] In one embodiment, compare the feature information for housing provident fund application in the verified user personal information with the preset standard feature information, including: Perform standardized format conversion on the verified user personal information and extract key feature items, and convert them into a structured comparison template; Compare the structured comparison template with the preset standard feature information, and determine whether it matches the preset standard feature information through the semantic parsing engine. If it matches, it is determined as user feature information. If it does not match, it is determined as invalid information; Output the user feature information obtained by the matching.

[0032] The beneficial effects of the above technical solution are as follows: Standardize the format of the user's personal information that has passed the verification and extract key feature items, which can standardize the data, enhance the versatility of this method, compare with the preset standard feature information, determine whether it matches the preset standard feature information through the semantic analysis engine, output the user feature information obtained by the matching, and use the semantic analysis engine to compare feature items, enabling fuzzy matching and enhancing the user experience.

[0033] In one embodiment, deploy an audit decision engine based on machine learning to conduct a risk behavior audit on the user feature information and return an audit opinion, including: Use a pre-trained risk identification model to intelligently analyze the obtained user feature information. If the risk identification model identifies that the audit is passed, return an audit opinion of passed; Deploy a machine learning framework. When the divergence degree of the pre-trained risk identification model exceeds the threshold, trigger the arbitration function based on machine learning to generate an audit opinion of rejection or supplementary materials; Package the pre-trained risk identification model and the machine learning framework to obtain an audit decision engine; Input the user feature information into the audit decision engine, automatically traverse the decision nodes according to the matching degree of each user feature information, and respectively determine the material integrity and condition compliance to generate the audit opinion of this user.

[0034] The beneficial effects of the above technical solution are as follows: Use a pre-trained risk identification model to intelligently analyze the obtained user feature information, deploy a machine learning framework, generate an audit opinion, and package the model and the framework to obtain an audit decision engine, which enhances the intelligence level and response speed of the audit opinion judgment, can quickly obtain targeted audit opinions, input the user feature information into the audit decision engine to generate the audit opinion of this user, enabling the user to obtain an audit opinion tailored to their own needs, facilitating the user to provide effective materials targeted, and reducing the computing power burden.

[0035] In one embodiment, push the audit opinion to the user immediately through the mobile service page. For applications that pass, the system connects to the financial payment system for provident fund allocation. For applications that require supplementary materials or are rejected, automatically generate a user personal information supplement list and monitor the audit progress in real time, including: Perform result annotation on the information in the housing provident fund withdrawal application according to the audit opinion output by the audit decision engine and push the information on the mobile service page; Classify the result annotation to generate a first classification result with an audit opinion of passed, a second classification result with an audit opinion of rejected, and a third classification result with an audit opinion of supplementary materials; When the result is the first classification result, connect to the financial payment system of the housing provident fund management center according to the content of the housing provident fund withdrawal application provided by the user and the user identity information, and allocate the housing provident fund according to the specific amount of the housing provident fund withdrawal application; When the result is the second classification result, generate the reasons for rejecting the housing provident fund withdrawal application and a supplementary list of the user's personal information, and push it to the user on the mobile service page; When the result is the third classification result, mark the specific items of the housing provident fund withdrawal application lacking materials, and generate a supplementary list of the user's personal information that needs to be supplemented with materials, and push it to the user on the mobile service page; Push the review opinion on the mobile server side and update the status in real time.

[0036] The beneficial effects of the above technical solution are as follows: By marking the information in the housing provident fund withdrawal application according to the review opinion output by the review decision engine and pushing the information on the mobile service page, the review results of each item can be pushed. Further, classifying the result marking can determine the type of the user's review opinion, improving the conciseness of the review opinion.

[0037] In one embodiment, applying blockchain technology, the key data of each housing provident fund mobile withdrawal application and user data verification is stored on the chain to form an immutable verification evidence chain, including: Determine the user's personal information, the process information of the multi-dimensional cross-verification process of the user identity information, and the review opinion as the key data in the housing provident fund withdrawal application review process; Package the key data into a block, and use the hash algorithm to generate the first hash value according to the block; Use the hash algorithm to generate the second hash value for the operation log generated by each operation on the mobile side; Record the first hash value, the second hash value and the relevant timestamp information on the blockchain; Establish a multi-level permission query channel. Users can view their personal business records with a private key. The regulatory agency can trace the complete evidence chain with a regulatory key, and each participating department can access the relevant chain segment data according to its permissions.

[0038] In some embodiments, a mobile terminal verification and evidence storage system for housing provident funds is constructed through blockchain technology. First, user identity information, multi-dimensional cross-verification process and audit opinions are defined as key data, which are packaged into blocks and generate a first hash value; at the same time, the mobile terminal operation log generates a second hash value, and the two types of hash values ​​and timestamps are jointly stored on the chain. The system adopts a distributed accounting mechanism to ensure that data cannot be tampered with, and establishes a hierarchical authority management system: users query personal business evidence through private keys, and regulatory agencies can trace the complete chain of evidence with regulatory keys. Each business participating department accesses the corresponding chain segment data according to the authority configuration, thereby achieving a balance between the verifiability and privacy protection of the entire verification process. This design not only retains the traceability advantage of blockchain, but also meets the data usage needs of different subjects through intelligent authority division.

[0039] The beneficial effects of the above technical solution are: by packaging key data into blocks, and using a hash algorithm to generate a hash value for the operation log generated by each operation on the mobile terminal, and storing it on the chain together with the timestamp, the security and tamper-proofness of the method are definitely improved, and a multi-level permission query channel is established. Users can view personal business evidence through private keys, and regulatory agencies can trace the complete chain of evidence with regulatory keys. Each participating department accesses relevant chain segment data according to authority, which improves the traceability of data and reduces the difficulty of supervision.

[0040] In one embodiment, the method further includes generating a user profile based on the user's financial behavior information and interpersonal relationships, and identifying risky behaviors of the user: By accessing financial platforms such as housing provident fund management systems, banking systems, and third-party payment platforms, the user's financial behavior data within the financial platforms is collected based on the user's personal information; In some embodiments, collecting the user's financial behavior data in the financial platform mainly includes connecting to the provident fund center, commercial banks and third-party payment institutions through a secure data interface to obtain the user's deposit records, account transaction flow, consumption behavior and other financial data in real time; Preprocessing the financial behavior data, removing scattered consumption records below a preset limit, and aligning and sorting the financial behavior data according to time series; In some embodiments, the data cleaning engine is mainly used to pre-process the original financial behavior data, first filtering out small scattered transaction records with a single amount below a set threshold (such as 100 yuan), and then using timestamp analysis and event sequence reorganization technology to standardize all valid transaction data in a time sequence accurate to the minute, forming a continuous and complete capital track time series chain; Conduct time series analysis on the pre-processed financial behavior data, identify abnormal fluctuations in funds in the financial behavior data, establish a dynamic repayment ability prediction model, determine the user's repayment ability, and determine the analysis results as the user's financial profile; In some embodiments, the method uses a time series analysis algorithm to mine the preprocessed financial data, identifies abnormal fluctuation patterns by detecting indicators such as the frequency of capital flow, periodic income and expenditure characteristics, and sudden large transactions; constructs a dynamic repayment ability evaluation model based on sliding window statistics and machine learning techniques, comprehensively analyzes dimensions such as the user's income stability, debt ratio, and cash flow health, quantitatively calculates the repayment ability coefficient, and finally generates a three-dimensional financial portrait containing elements such as credit ratings and risk warning labels; Identifies the contacts who generate large capital flows and frequent transactions in the user's financial behavior data, and determines the contacts as key contacts; In some embodiments, the above method mainly analyzes the user's fund transaction data, identifies the transaction accounts with a single transfer exceeding the set threshold or abnormal monthly average transaction frequency, and combines transaction feature analysis to label these high-frequency and high-amount fund interaction objects as key contacts; According to the key contacts, uses graph computing technology to construct a user social connection graph, determines the user's risk conduction path, establishes a relationship network influence evaluation model, determines the high-risk contacts in the social network, and determines the evaluation result as the user's social portrait; In some embodiments, the above method constructs a multi-dimensional social and financial network graph based on the identified key contact data, divides the associated clusters, and quantifies the intensity of fund transactions according to weight analysis, thereby depicting the risk conduction path with the user as the core. Combines a dynamic contagion model to evaluate the risk diffusion probability, and marks the high-influence nodes in the network (such as contacts who have both high transaction frequencies, large capital flows, and are at the intersection of multiple communities) with risks, and finally generates a user social portrait containing dimensions such as risk contagion coefficients, vulnerability indexes, and lists of potential risk sources; Uses the user's financial portrait and the user's social portrait to jointly generate the user's portrait; Determines the possible risk labels for each item in the user portrait, and uses an ensemble learning algorithm to construct a user portrait risk detection model based on the risk labels; In some embodiments, the above method extracts risk indicators (such as abnormal transaction frequencies, high-risk associated nodes, etc.) from the user's financial portrait and social portrait through feature engineering, and calculates the user's comprehensive risk score through weighted fusion of multi-dimensional risk labels; Uses the user portrait risk detection model to identify the risk behaviors of the user portrait to be detected, and outputs the risk behavior detection results of the user.

[0041] In some embodiments, the above method inputs the financial behavior feature vector and social relationship graph data of the user to be detected into a trained integrated learning model, and through multi-dimensional risk feature cross-analysis, outputs a detection report including a risk level score, an abnormal behavior type label, and a risk conduction path.

[0042] The beneficial effects of the above technical solution are as follows: By accessing financial platforms such as the housing provident fund management system, bank system, and third-party payment platform, collecting the financial behavior data of users within the financial platform, determining the user's financial portrait, and prompting the user's high-risk financial behaviors, the risk of property loss caused by financial fraud to the user is reduced; By using the users with whom the user has financial transactions to form a user social portrait, it is convenient to identify potential high-risk factors in the user's social environment, and the security of the user's account is improved.

[0043] In one embodiment, the method further includes that when the user withdraws the housing provident fund for a housing loan, the user is subject to credit verification by a third-party institution, and based on the credit verification result, a user risk assessment is carried out, and a risk assessment report is issued to the housing provident fund loan institution: Perform keyword detection on the user's housing provident fund withdrawal application, and determine that credit verification of the user is required when it is detected that the purpose of the user's housing provident fund withdrawal is for a housing loan; In some embodiments, the above method parses and extracts the purpose description field in the withdrawal application through natural language processing technology, and when keywords such as "purchase of a house", "housing loan", and "down payment" are recognized, the credit verification process is automatically triggered; Determine the specific credit items for which the user needs to conduct credit verification based on the user's personal information; In some embodiments, the above method intelligently matches the credit verification items according to the user's identity information (such as occupation, income, and housing situation), including core credit indicators such as loan records, credit card repayments, and guarantee situations; Obtain the user's credit information from the People's Bank of China credit institution, extract the specific credit items that need to be verified by a third party in the credit information of the People's Bank of China, and conduct the first validity verification; In some embodiments, the above method connects to the People's Bank of China credit system through a financial private network, retrieves core credit data such as the user's credit records and overdue information, and screens keyword fields (such as large loan records) for key verification to complete the first round of authenticity verification; Obtain the user's private credit information from a private credit institution, extract the specific credit items that need to be verified by a third party in the private credit information, and conduct the second validity verification; In some embodiments, the above method accesses the databases of market-oriented credit investigation institutions to obtain supplementary credit investigation data such as users' consumer credit and performance behaviors, initiates data verification for specific items to be verified (such as small loan records), and completes the second-round authenticity verification; Based on the first validity verification result and the second validity verification result, confirm the corresponding risk assessment result of each specific credit investigation item; Generate a housing provident fund loan risk assessment report verified by a third party based on the risk assessment result; In some embodiments, the above method integrates the central bank credit investigation, private credit investigation, and multi-dimensional verification results, and uses a risk quantification model to generate an assessment report including credit scores, risk levels, and abnormal index prompts; Obtain the risk assessment abnormal items in the housing provident fund loan risk assessment report verified by a third party, and determine the corresponding abnormal reasons for each risk assessment abnormal item and the specific score value of the risk assessment result of the risk assessment abnormal item; In some embodiments, the above method analyzes the abnormal identification fields in the risk assessment report, intelligently associates multi-source data to trace the abnormal causes (such as credit investigation overdue records, income-debt ratio imbalance, etc.), and quantitatively extracts the risk deduction value and weight coefficient corresponding to each abnormal item; Determine the corresponding solution based on the abnormal reason corresponding to each risk assessment abnormal item, and generate a specific risk decision according to the corresponding solution; In some embodiments, after the above method performs attribution analysis on each risk abnormal item through an intelligent decision-making engine, it automatically matches a preset disposal strategy library: for abnormal credit investigation overdue records, trigger a supplementary guarantee requirement; for income-debt ratio exceeding the standard items, generate an income verification or loan amount reduction plan; if associated risks are involved, initiate a social network review process; based on the disposal plan, automatically generate a final decision including conditional approval, quota adjustment, risk control additional terms, etc.; Evaluate the risk control threshold of each specific risk decision, and determine the risk control threshold of each specific risk decision; In some embodiments, the above method dynamically evaluates each decision parameter through a risk quantification model, and calculates the risk boundary values of each decision dimension based on the historical default data distribution, market environment variables, and user characteristic matrix. For example: for the supplementary guarantee decision, set the guarantee coverage threshold to not be lower than 150%, for the loan amount adjustment plan, determine the maximum debt-to-income ratio threshold according to the income volatility (such as 55%), and for the social network review, set the upper limit of the influence coefficient of associated risk nodes; Use the risk control threshold corresponding to each specific risk decision for each risk assessment exception item to judge the specific score value corresponding to each risk assessment exception item in the housing provident fund loan risk assessment report verified by a third party, and confirm whether the specific risk decision corresponding to this risk assessment exception item can control the risk of this risk assessment exception item; In some embodiments, the above method is mainly used to perform quantitative analysis on each exception item in the assessment report, and perform multi-dimensional comparison and verification on the actual score value of the exception item and the preset risk control threshold: for credit overdue exceptions, compare the number of overdue times with the allowable threshold range; for income-liability exceptions, check whether the debt ratio exceeds the upper limit of the dynamic threshold; for those involving associated risks, evaluate whether the social network risk contagion coefficient is within the controllable range. Automatically determine the effectiveness of each disposal plan through the decision tree algorithm. When the score value of the exception item exceeds the corresponding threshold tolerance, automatically upgrade the disposal level or trigger the manual review process. All judgment processes generate a verification report containing elements such as threshold comparison data, disposal effectiveness coefficient, and confidence score; Output the judgment result as the feasibility analysis result of the risk control for each risk assessment exception item; Generate and output the reference judgment result for the housing provident fund loan credit of this customer based on the feasibility analysis result of the risk control for each risk assessment exception item.

[0044] The beneficial effects of the above technical solution are as follows: Through the credit verification of users by a third-party institution, the credit records of users are made true and comprehensive, enhancing the accuracy of the assessment results. Based on the credit verification results, user risk assessment is carried out, and a risk assessment report is issued to the housing provident fund loan institution. By providing a credit reference for housing loans through housing provident fund data and credit data, this method has a wider application scenario and improves the utilization efficiency of user data.

[0045] After considering the specification and the disclosure here in practice, those skilled in the art will readily think of other embodiments of the present disclosure. This application aims to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include well-known common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.

[0046] It should be understood that the present disclosure is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. An automated review method for housing provident fund mobile extraction applications, characterized in that, It includes the following steps: Obtain the user's personal information according to the housing provident fund withdrawal application submitted by the user on the mobile service page, and verify the user's identity using dynamic biometric information verification technology; By introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources to conduct multi-dimensional cross-verification of the user's identity information without exposing the original data; Compare the feature information used for housing provident fund applications in the verified user's personal information with the preset standard feature information; Deploy an audit decision engine based on machine learning to conduct risk behavior audits on the user's feature information and return audit opinions; Instantly push the audit opinions to the user through the mobile service page. For approved applications, the system connects to the financial payment system for provident fund allocation. For applications that require supplementary materials or are rejected, automatically generate a user personal information supplement list and monitor the audit progress in real time; Apply blockchain technology to store the key data of each housing provident fund mobile withdrawal application and user data verification on the blockchain to form an immutable verification evidence chain.

2. The automated review method for housing provident fund mobile extraction application according to claim 1, wherein, Obtain the user's personal information according to the housing provident fund withdrawal application submitted by the user on the mobile service page, and verify the user's identity using dynamic biometric information verification technology, including: After the user submits a housing provident fund withdrawal application through the mobile service page, identify the user's personal information in the housing provident fund withdrawal application, and obtain the pre-stored voiceprint and face recognition information of the corresponding user in the housing provident fund database; Use the camera to obtain the user's current facial status and compare it with the pre-stored face recognition information. If it matches, it is determined that the face recognition is passed; Randomly generate a user's real-time verification action, and when the user makes the corresponding action, it is regarded as the real-time verification passed; Randomly generate a string of numbers. When the voiceprint data of the generated numbers read by the user matches the pre-stored voiceprint data, it is regarded as the voiceprint verification passed; Conduct micro-expression detection on the user's real-time video data. If it is detected that the micro-expression does not match the human real expression pattern, the real-time video data is identified as AI-generated video; When the face recognition is passed, the real-time verification is passed, the voiceprint verification is passed, and it is determined that the real-time video data is not AI-generated video, it is determined that the user identity verification is passed; otherwise, it is regarded as the user identity verification failed and needs to be verified again.

3. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that, By introducing a cross-institutional federated learning framework, jointly construct a joint verification model with multiple data sources to conduct multi-dimensional cross-verification of the user's identity information without exposing the original data, including: Obtain multi-source data based on multiple official databases, construct a unified feature space mapping table, and process the multi-source data into unified data; Use the algorithm under the federated learning framework to construct a vertical federated data matrix according to the unified data, and use an encryption algorithm to perform an available but invisible process on the vertical federated data matrix; Determine the standard threshold for dynamic verification settings based on the processed vertical federated data matrix, and supplement and set the standard threshold for spatio-temporal consistency verification of multiple data; Construct the joint verification model based on the standard threshold for dynamic verification settings and the standard threshold for spatio-temporal consistency verification; Input the user identity information into the joint verification model. If the user identity information meets the verification standard, it is displayed as verified passed. If the user identity information does not meet the verification standard, it is displayed as abnormal identity information.

4. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that, Compare the feature information for housing provident fund application in the verified user personal information with the preset standard feature information, including: Perform standardized format conversion on the verified user personal information and extract key feature items, and convert them into a structured comparison template; Compare the structured comparison template with the preset standard feature information, and determine whether it matches the preset standard feature information through a semantic parsing engine. If it matches, it is determined as user feature information. If it does not match, it is determined as invalid information; Output the user feature information obtained by the match.

5. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that Deploy an audit decision engine based on machine learning to conduct a risk behavior audit on the user feature information and return an audit opinion, including: Use a pre-trained risk identification model to perform intelligent parsing on the obtained user feature information. If the risk identification model identifies it as verified passed, return a passed audit opinion; Deploy a machine learning framework. When the divergence degree of the pre-trained risk identification model exceeds the threshold, trigger the arbitration function based on machine learning to generate an audit opinion of rejection or supplementary materials; Package the pre-trained risk identification model and the machine learning framework to obtain an audit decision engine; Input the user feature information into the audit decision engine, automatically traverse the decision nodes according to the matching degree of each user feature information, and respectively conduct judgments on the material integrity and condition compliance to generate the audit opinion of this user.

6. The automated review method for housing provident fund mobile extraction application according to claim 1, wherein Push the audit opinion to the user immediately through the mobile service page. For the passed application, the system docks with the financial payment system to conduct provident fund allocation. For the application that requires supplementary materials or rejection, automatically generate a user personal information supplementary list and monitor the audit progress in real time, including: Perform result annotation on the information in the housing provident fund withdrawal application according to the audit opinion output by the audit decision engine and push the information on the mobile service page; Classify the result annotation to generate a first classification result with an audit opinion of passed, a second classification result with an audit opinion of rejection, and a third classification result with an audit opinion of supplementary materials; When the result is the first classification result, dock with the financial payment system of the provident fund management center according to the housing provident fund withdrawal application content and user identity information provided by the user, and conduct provident fund allocation according to the specific amount of the housing provident fund withdrawal application; When the result is the second classification result, generate the rejection reason of the housing provident fund withdrawal application and a user personal information supplementary list, and push it to the user on the mobile service page; When the result is the third classification result, mark the specific items of the housing provident fund withdrawal application lacking materials, and generate a user personal information supplementary list that needs to supplement materials, and push it to the user on the mobile service page; Push the audit opinion on the mobile server and update the status in real time.

7. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that Apply blockchain technology to store the key data of each housing provident fund mobile withdrawal application and user data verification on the chain to form an immutable verification evidence chain, including: Determine the user's personal information, the process information of the multi-dimensional cross-validation process of the user's identity information, and the review opinion as key data in the review process of the housing provident fund withdrawal application; Packing the key data into blocks, and generating a first hash value according to the blocks using a hash algorithm; The operation log generated by each operation of the mobile terminal is used to generate a second hash value using a hash algorithm; Recording the first hash value and the second hash value together with relevant timestamp information on a blockchain; A multi-level permission query channel is established, users can view personal business evidence through private keys, regulatory agencies can trace the complete chain of evidence with regulatory keys, and each participating department can access relevant chain segment data according to permissions.

8. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that, The method further includes generating a user profile based on the user's financial behavior information and interpersonal relationships, and identifying risky behaviors of the user: By accessing financial platforms such as housing provident fund management systems, banking systems, and third-party payment platforms, the user's financial behavior data within the financial platforms is collected based on the user's personal information; Preprocessing the financial behavior data, removing scattered consumption records below a preset limit, and aligning and sorting the financial behavior data according to time series; Conduct time series analysis on the pre-processed financial behavior data, identify abnormal fluctuations in funds in the financial behavior data, establish a dynamic repayment ability prediction model, determine the user's repayment ability, and determine the analysis results as the user's financial profile; Identify contacts who have large capital flows and frequent transactions in the user's financial behavior data, and identify the contacts as key contacts; Based on the key contacts, a user social connection map is constructed using graph computing technology to determine the risk transmission path of the user, a relationship network influence assessment model is established, high-risk contacts in the social network are determined, and the assessment results are determined as the user's social profile; Generate a user profile of the user using the user financial profile and the user social profile; Determine the possible risk labels for each item in the user profile, and use an ensemble learning algorithm to build a user profile risk detection model based on the risk labels; Use the user profile risk detection model to identify risky behaviors of user profiles that need to be detected, and output the user's risk behavior detection results.

9. The automated review method for housing provident fund mobile extraction application according to claim 1, characterized in that, The method further includes, when a user withdraws housing provident fund for housing loan, conducting credit verification on the user through a third-party institution, conducting risk assessment on the user based on the credit verification result, and issuing a risk assessment report to the housing provident fund lending institution: Perform keyword detection on the user's housing provident fund withdrawal application. If it is detected that the user's housing provident fund withdrawal is for housing loan, it is determined that the user needs to be verified for credit; Determine the specific credit items that the user needs to verify based on the user's personal information; Obtain the user's People's Bank credit information through the People's Bank credit agency, extract the specific credit items in the People's Bank credit information that need to be verified by a third party for the first validity verification; Obtain the user's private credit information through a private credit agency, extract specific credit items in the private credit information that require third-party verification for a second validity verification; Confirm the corresponding risk assessment results for each specific credit investigation item based on the first validity verification result and the second validity verification result; Generate a housing provident fund loan risk assessment report verified by a third party based on the risk assessment results; Obtain the risk assessment abnormal items in the housing provident fund loan risk assessment report verified by a third party, and determine the corresponding abnormal reasons for each risk assessment abnormal item and the specific score value of the risk assessment result of the risk assessment abnormal item; Determine the corresponding solution based on the abnormal reason corresponding to each risk assessment abnormal item, and generate a specific risk decision according to the corresponding solution; Evaluate the risk control threshold of each specific risk decision, and determine the risk control threshold of each specific risk decision; Use the risk control threshold of the specific risk decision corresponding to each risk assessment abnormal item to judge the specific score value corresponding to each risk assessment abnormal item in the housing provident fund loan risk assessment report verified by a third party, and confirm whether the specific risk decision corresponding to the risk assessment abnormal item can control the risk of the risk assessment abnormal item; Output the judgment result as the feasibility analysis result of the risk control of each risk assessment abnormal item; Generate and output the reference judgment result of the housing provident fund loan credit granting for the user based on the feasibility analysis result of the risk control of each risk assessment abnormal item.

Images