A multi-path integrated community resident location authentication risk control method and system

By integrating multiple pathways such as LBS continuous multi-day location verification, neighborhood guarantee credit granting, and OCR intelligent de-identification verification of invoices, the system solves the problems of insufficient accuracy and privacy protection in community resident identity authentication, and achieves high-precision, dynamic authentication-behavior linkage risk control, providing differentiated credit evaluation and continuous management.

CN123118250APending Publication Date: 2026-11-17HEFEI MATRIX DIGITAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611219428.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-12
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing community resident identity authentication schemes suffer from problems such as insufficient authentication accuracy, inadequate privacy protection, lack of dynamic scoring and anti-fraud mechanisms that integrate multiple channels, lack of dynamic risk control loops that link authentication and behavior, and lack of differentiated authentication design.

Method used

The system employs a weighted fusion of three parallel authentication channels: LBS continuous multi-day location verification, neighborhood guarantee credit granting, and OCR intelligent de-identification verification of invoices. Combined with anomaly channel detection and dynamic weight adjustment, a dynamic risk control closed loop is formed. Through a tiered credit risk control mechanism and multimodal behavior monitoring, it achieves accurate identification and continuous management of the local identity of community residents.

Benefits of technology

It achieves high-precision local authentication, protects user privacy, prevents authentication fraud, has dynamic risk control capabilities, provides differentiated credit evaluation, and forms a sustainable governance system that links authentication and behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118250A_ABST
    Figure CN123118250A_ABST
Patent Text Reader

Abstract

This invention discloses a multi-path fusion-based community resident location authentication risk control method and system, belonging to the field of internet identity authentication and risk control technology. This invention utilizes three parallel channels: LBS continuous location verification, neighborhood guarantee credit granting, and local de-identification verification via OCR for invoices, outputting location credit scores, guarantee credit scores, and invoice verification scores respectively. An abnormal channel detection and dynamic weight adjustment are performed by a weighted fusion scoring engine to calculate a comprehensive location credit score. Based on this score, resident permissions are granted and four credit levels (L0 to L3) are assigned. Combined with device fingerprint anti-fraud and multimodal behavior monitoring, a dynamic risk control closed loop linking authentication and behavior is formed. This invention can efficiently verify location identity without obtaining complete privacy documents, effectively identifying itinerant vendors and authentication fraud gangs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of Internet identity authentication and risk control technology, specifically involving a multi-path fusion community resident location authentication risk control method and system, and particularly involving a community resident location identity verification and dynamic hierarchical risk control method based on the weighted fusion of three parallel authentication channels: LBS continuous multi-day location verification, neighborhood guarantee credit linkage deduction, and local privacy desensitization of invoice OCR terminal. Background Technology

[0002] With the rapid development of the community O2O economy, community social platforms, such as those for trading secondhand goods and providing neighborhood mutual assistance, are becoming increasingly popular. The core value of these platforms lies in "neighborhood trust," where users establish a trust relationship for transactions based on their shared residential community identity. However, existing technological solutions face the following technical challenges in authenticating the resident identity of community members: First, the lack of accuracy due to the reliance on a single authentication channel. Existing solutions mostly rely on a single authentication method—either based solely on a single location snapshot using LBS, or solely on facial recognition combined with access control hardware, or solely on QR code authentication—which cannot cross-verify a user's location identity from multiple dimensions, including time, physical credentials, and social guarantees. This makes it difficult to effectively distinguish between long-term residents and temporary passersby, itinerant vendors, and organized authentication fraud groups.

[0003] Second, privacy protection mechanisms are lacking or severely inadequate. Some solutions require users to upload complete photos of their property ownership certificates or ID cards, involving highly sensitive privacy information such as names, ID numbers, and detailed apartment numbers, resulting in extremely low user retention rates. Even when de-identification solutions exist, the de-identified information is not used for the specific scenario of local identity verification, and the de-identified data still needs to be uploaded to a third-party platform, posing a risk of indirect privacy leaks.

[0004] Third, there is a lack of dynamic scoring and anti-cheating mechanisms that integrate multiple channels. Although some existing solutions involve multi-dimensional weighted authentication, they use fixed preset weights and cannot dynamically adjust the weights according to the scoring deviations between channels. They also cannot detect abnormal cheating channels and automatically reduce or clear their weights, nor do they include a channel integrity reward mechanism or targeted device fingerprint anti-fraud measures.

[0005] Fourth, there is no dynamic risk control loop linking authentication and behavior. Existing solutions grant fixed permissions after authentication, without linking the guarantor's credit to the behavior of the guaranteed party for penalty points. They do not establish a complete closed loop of "authentication → authorization → behavior monitoring → credit adjustment → permission revocation", and lack the ability to automatically downgrade and revoke permissions for users whose behavior continues to deteriorate after authentication.

[0006] Fifth, there is a lack of differentiated authentication designs for various types of residents in the community. Risk control solutions in the financial lending field are geared towards loan credit assessment, and smart community solutions are geared towards access control. Neither has been specifically designed for the local identity authentication needs in community second-hand transactions and neighborhood mutual assistance scenarios—that is, how to provide fair and reasonable local credit evaluations for users with different living situations, such as owners, tenants, and temporary residents, while protecting user privacy and utilizing the unique neighborhood relationships and various types of residential trace information in the community.

[0007] Therefore, there is an urgent need for a method and system that can efficiently verify the local identity of community residents while protecting user privacy, effectively distinguish between real residents and itinerant vendors and authentication fraud gangs, and has dynamic risk control capabilities that link authentication and behavior. Summary of the Invention

[0008] To address the aforementioned technical challenges, this invention provides a multi-path fusion-based community resident location authentication risk control method and system. It achieves comprehensive and accurate identification of community residents' location through the weighted fusion of three parallel authentication channels: LBS continuous multi-day location verification (time dimension), neighborhood guarantee credit granting (social guarantee dimension), and OCR intelligent de-identification verification of invoices (physical voucher dimension). Furthermore, it enhances the anti-fraud capabilities of the fusion score through anomaly channel detection and dynamic weight adjustment mechanisms. Finally, it dynamically binds authentication scores to functional permissions and transaction limits through a tiered credit risk control mechanism, continuously monitoring multimodal behavioral data to achieve dynamic upgrades and downgrades, forming a complete risk control closed loop of authentication-authorization-behavior monitoring.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A multi-path integrated community resident location authentication risk control method includes: Step S1: Collect LBS positioning data from user terminals. Based on the preset observation time window and spatial threshold, calculate the percentage of effective stay days and the location distribution entropy value of users. Perform trajectory smoothness detection to identify GPS simulation cheating behavior and output the location credit score S_LBS. Step S2: Establish neighborhood guarantee credit relationship, allocate guarantee credit limit and guarantee number limit to certified users, bind the risk control status of the guaranteed person's account to the guarantor's credit limit, deduct the guarantor's credit limit proportionally when the guaranteed person violates the rules, maintain the guarantee relationship graph to detect group cheating behavior, and output guarantee credit score S_GUARANTEE; Step S3: Receive the ticket image containing community information uploaded by the user, perform OCR text recognition and automatic privacy information desensitization processing on the local terminal, extract only the community name field and upload it to the server for database matching and verification, and verify the authenticity of the ticket image, and output the ticket verification score S_OCR; Step S4: Input S_LBS, S_GUARANTEE, and S_OCR into the multi-dimensional weighted fusion scoring engine, perform abnormal channel detection and dynamic weight adjustment, and calculate the comprehensive local credit score S_total = W1'×S_LBS + W2'×S_GUARANTEE + W3'×S_OCR, where Wi' is the dynamically adjusted weight coefficient. Step S5: Compare the comprehensive local credit score S_total with the preset authorization threshold. When S_total reaches the threshold, automatically grant the user community resident privileges and classify the user into different credit levels according to the value range of S_total. Each level corresponds to different functional permissions and transaction limits. Step S6: Continuously monitor user behavior data. When behavior indicators deteriorate, automatically lower the credit rating and restrict functional permissions to form a dynamic risk control closed loop.

[0010] Furthermore, in step S1, the formula for calculating the location distribution entropy value H_loc is H_loc = -∑(p_i×log(p_i)), where p_i is the proportion of the user's location points in the i-th geographical region; when the proportion of effective stay days is ≥ the first threshold T1=0.6 and H_loc≤ the second threshold T2=1.2, it is determined to be a fixed-point permanent user; when H_loc> the third threshold T3=2.0 and a large-scale cross-city displacement is detected, it is determined to be a cross-city drifting account and marked as a suspected itinerant vendor; the trajectory smoothness detection includes calculating the displacement distance and movement speed between adjacent location points, and when the movement speed is detected to exceed the preset upper limit of reasonable human movement speed of 350 km / h or an instantaneous displacement jump that does not conform to the laws of physical motion occurs, it is determined to be GPS simulation cheating behavior.

[0011] Furthermore, in step S2, the guarantee relationship graph analysis includes: when a closed loop of guarantee relationships is detected (i.e., guarantor A guarantees guaranteed party B, and guaranteed party B directly or indirectly guarantees guarantor A, forming a circular guarantee link), or multiple guaranteed parties of the same guarantor exhibit a batch of violations within a preset time window (default 30 days) (the number of violating guaranteed parties exceeds 50%), a group fraud warning mechanism is triggered, and the following downgrade measures are implemented: (a) suspending all guarantee permissions of guarantors within the closed loop or related guarantors; (b) downgrading the guarantor's credit rating by one level; (c) relevant users enter a 90-day observation period; (d) during the observation period, no violations must be maintained before applying for restoration of guarantee permissions; (e) after restoration, the upper limit of the number of guarantors is reduced to 70% of the original upper limit, and the guarantee credit limit is recalculated by multiplying by the restoration discount factor η_recovery (default 0.6).

[0012] Furthermore, in step S3, the automatic desensitization of privacy information is achieved through a Named Entity Recognition (NER) model deployed on the local terminal. The identified privacy entity categories include name, ID card number, detailed room number, and phone number. After replacing the detected privacy entities with mask characters or deleting them directly, only the non-sensitive community name field is uploaded to the server. The authenticity verification of the invoice includes texture feature analysis of the uploaded image, font edge jaggedness consistency detection, seal color distribution analysis, and Exchangeable Image File Format (EXIF) metadata verification to identify fake invoices synthesized by Photoshop or downloaded from the network. In step S3, a multimodal large language visual model is used as the core audit engine for invoices. The model automatically determines the invoice type and adapts the extracted template through layout analysis. Supported invoice types include at least real estate ownership certificates, lease contracts, property fee vouchers, water, electricity, and gas bills, broadband installation slips, and community registration forms. The model performs semantic consistency checks on the extracted community name, address, and homeowner name. When the community name and address do not semantically match, it is marked as a contradiction in the invoice information and the verification confidence is reduced. The model outputs an auxiliary score for authenticity by analyzing the clarity of the seal edge, color uniformity, continuity of the anti-counterfeiting background pattern, and paper texture features. The model inference is executed in a secure isolation sandbox on the server side. After the invoice image inference is completed, the image data in the GPU memory and system memory is immediately cleared, and the original invoice image file is not retained. It also includes a differentiated confidence allocation mechanism for multi-level certification materials: the system automatically assigns basic confidence weights based on the type of certification material, with real estate ownership certificates as w_max, registered lease contracts as w_mid, and residential trace materials such as property fees or utility bills as w_low; when a user submits multiple materials of different types, a material type diversity reward is introduced, with the comprehensive confidence S_OCR_final = min(100, S_OCR×(1+γ_type×n_types)), where n_types is the number of unique material types and γ_type is the material type diversity reward coefficient; the system also implements validity period management for certification materials, with expired materials having their contribution weight reduced by a decay function W_valid(t)=W_initial×max(0,1-(t-t_issue) / T_valid), where t_issue is the material issuance date and T_valid is the material validity period, and a recertification reminder is automatically pushed after the material expires. The invoice verification score S_OCR is the comprehensive confidence score S_OCR_final after the above multimodal review and material differential confidence allocation.

[0013] Further, in the step S4, the abnormal channel detection is implemented by calculating the pairwise deviation d_ij = |S_i - S_j| of the three scores (where S_i and S_j are respectively the credit scores of the i-th authentication channel and the j-th authentication channel). The dynamic weight adjustment rules include: (a) the system presets initial weights for the three channels; (b) when the deviation between the score of a single channel and the scores of the other two channels both exceeds the preset threshold Δ, a decay factor is applied to the weight of this channel, and the reduced weight is proportionally distributed to the other normal channels; (c) when a certain channel is detected to have a suspicion of cheating, the weight of this channel is set to zero; (d) when the user completes all three authentication channels (all three scores are greater than zero), the result is multiplied by an integrity reward coefficient β (β>1.0).

[0014] Further, in the step S5, the credit levels include four levels from L0 to L3: the L0 level is an unauthenticated state, where the user can only browse but cannot publish or communicate; the L1 level is basic authentication, where the user can publish items and communicate, with a single-day transaction limit set; the L2 level is standard authentication, where the user can publish items, initiate purchase requests and act as a guarantor, with a limited upper limit of the number of guarantors; the L3 level is advanced authentication, where the user enjoys all functional permissions, the highest transaction limit and priority recommendation weight. It also includes community function dynamic authorization driven by authentication completeness: the system defines the authentication completeness index A_complete = Σ(w_channel_i×flag_i), where w_channel_i is 0.35 for LBS location verification, 0.40 for bill OCR verification, and 0.25 for neighborhood guarantee credit granting, and flag_i is the completion status of each channel; the system gradually opens community functions according to A_complete — when A_complete=0, only the browsing function is opened; when 0<A_complete<θ_low (default θ_low=0.4), the purchase request demand and basic private message functions are additionally opened; when θ_low≤A_complete<θ_mid (default θ_mid=0.7), the idle item publishing and neighborhood circle interaction functions are additionally opened; when θ_mid≤A_complete<1.0, the transaction reservation and evaluation functions are additionally opened; when A_complete=1.0, the guarantee credit granting and all community functions are opened, and the user enjoys the priority recommendation weight in search sorting. The default value of the single guarantee credit granting upper limit is 30 points.

[0015] Furthermore, it also includes an authentication anti-fraud mechanism that associates device fingerprints with multiple accounts: The system collects multi-dimensional attributes of terminal devices (including device model, system version, screen resolution, sensor list, network standard, startup timestamp, and application installation list hash value), and generates a unique device identifier F_dev through multi-feature fusion hashing; The system maintains a device-account association graph. When the same device fingerprint is associated with more than N_max accounts performing authentication in the same cell within a preset time window, a discount factor is applied to reduce the LBS location credit score of the relevant accounts and they are forced to undergo manual review; When multiple accounts associated with the same device fingerprint submit authentication materials to different cells within a time window, it is marked as cross-cell authentication fraud and automatic approval is suspended; If the similarity coefficient between the device fingerprint of a newly registered account and the fingerprint in the blacklist exceeds a preset threshold, the initial credit score is reduced and a behavior observation period of a preset duration is placed.

[0016] Furthermore, in step S6, the system continuously collects user transaction behavior data (fulfillment rate, number of defaults, dispute rate), social behavior data (message response rate, response time, number of times blocked), content behavior data (publication compliance rate, number of reports, report verification rate), and authentication behavior data (frequency of material submission, number of rejections, frequency of information changes). The system then calculates a behavioral health score B_health = w_trade × S_trade + w_social × S_social + w_content × S_content + w_auth × S_auth according to a preset period. S_trade is based on performance rate and dispute rate, S_social is based on message response rate and number of times blocked, S_content is based on publication compliance rate and number of reports, and S_auth is based on material submission frequency and number of rejections. w_trade, w_social, w_content, and w_auth are preset weights for each dimension, and the sum of the four weights is 1. When B_health is below the warning line for several consecutive periods, the credit rating is automatically downgraded. After downgrading, the user enters a cooling-off period and needs to maintain behavioral improvement before applying for level restoration. When restored, the user can be restored to the next level below the previous level.

[0017] Furthermore, the present invention also provides a multi-path fusion community resident location authentication risk control system, including: The LBS continuous multi-day location verification module is used to collect user terminal location data, statistically analyze the proportion of effective dwell days and the location distribution entropy value based on preset time windows and spatial thresholds, perform trajectory smoothness detection to identify GPS simulation cheating behavior, and output location credit score. The neighborhood guarantee credit module is used to manage the credit relationship between guarantors and guaranteed parties, allocate guarantee credit limits and guarantee frequency limits, realize the linkage between the guaranteed party's violation and the guarantor's credit limit deduction, maintain the guarantee relationship graph and detect group cheating behavior, and output guarantee credit scores. The invoice OCR intelligent desensitization verification module is used to perform OCR text recognition and automatic privacy information desensitization processing on the invoice images uploaded by users on the local terminal. It extracts only the community name field for database matching and verification, and identifies the authenticity of the invoice image and outputs the invoice verification score. The multi-dimensional weighted fusion scoring engine module is used to receive the scores output by the three modules, perform abnormal channel detection and dynamic weight adjustment, and calculate the comprehensive local credit score. The tiered credit risk control module is used to classify users' credit levels based on comprehensive scores, allocate corresponding functional permissions and transaction limits, and continuously monitor user behavior data to achieve dynamic upgrades and downgrades.

[0018] The invoice OCR intelligent desensitization verification module is deployed locally on the user terminal device. The complete privacy invoice image does not leave the user terminal, and only the desensitized community name field is transmitted to the server through the network for matching and verification.

[0019] On the other hand, the present invention also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors enable the aforementioned multi-path fusion community resident location authentication risk control method.

[0020] The beneficial effects of this invention are as follows: (1) High-precision local authentication: By parallel operation and weighted fusion of three authentication channels—LBS continuous multi-day location verification (time dimension), neighborhood guarantee credit granting (social guarantee dimension), and invoice OCR intelligent de-identification verification (physical voucher dimension)—the user's community local identity is cross-verified from three independent directions, significantly improving authentication accuracy. Built-in trajectory smoothness detection (identifying GPS simulation cheating), guarantee map analysis (identifying gang cheating), and invoice authenticity identification (identifying PS forgery) make it difficult for any single cheating method to pass the comprehensive scoring threshold.

[0021] (2) Local Privacy Desensitization and Data Security on the Terminal: The invoice OCR module automatically identifies and desensitizes privacy information locally on the user terminal using a Named Entity Recognition (NER) model, extracting only the non-sensitive cell name field and uploading it to the server for location matching. The complete privacy invoice image does not leave the user terminal, fundamentally avoiding the risk of privacy leakage. Compared with existing solutions that require complete uploading or rely on crowdsourcing platforms for manual recognition, this invention achieves the optimal balance between privacy protection and authentication effectiveness.

[0022] (3) Differentiated confidence level allocation for various types of certification materials: By classifying certification materials into a three-tiered system based on their evidentiary strength (real estate ownership certificates, registered lease contracts, and residential evidence materials such as utility bills), and combining this with a diverse reward mechanism, differentiated processing of materials with different evidentiary strengths can be achieved. Users can choose a suitable certification path based on their residential status (owner / tenant / borrower) to obtain a fair and reasonable local credit score.

[0023] (4) Dynamic risk control closed loop and tiered credit granting: The tiered credit granting risk control mechanism dynamically binds authentication scores with functional permissions and transaction limits, and links the guarantor's credit with the behavior of the guaranteed party to deduct points, forming a complete risk transmission and constraint chain. Through functional gradient authorization driven by authentication completion and continuous monitoring of multimodal behavioral data, the concept of progressive trust building of "the more authentication, the more trustworthy, the more free" is realized.

[0024] (5) Four-fold anti-fraud defense: device fingerprint and account association graph analysis + LBS GPS simulation cheating detection + invoice OCR authenticity identification + guarantee relationship graph gang cheating detection. The four mechanisms work together to reduce the success rate of organized authentication fraud attacks to an extremely low level.

[0025] (6) A sustainable governance system linking authentication and behavior: authentication grants initial permissions, behavior determines the upgrade or downgrade of permissions, and violations lead to credit damage. These three elements work together to form a complete closed loop. This achieves continuous measurement of authentication completion and continuous assessment of behavioral health, providing the community platform with an adaptive and sustainable trust governance system.

[0026] For ease of review and understanding, the key parameters involved in this invention, their default values, and value ranges are summarized as follows: Parameter symbol meaning Default value / range Ownership Rights T_window LBS observation time window 14 days Power 2 R_geo Target cell spatial threshold 3000 meters Power 2 T1 First threshold for percentage of effective stay days 0.6 Power 2 T2 Location distribution entropy value second threshold 1.2 Power 2 T3 Third threshold of cross-city drift distribution entropy 2.0 Power 2 V_max Upper limit of reasonable human movement speed 350 km / h Power 2 Δ Abnormal channel deviation threshold 40 points Power 5 γ Deviation channel attenuation factor 0.3 Power 5 β Completeness reward coefficient 1.1(β>1.0) Power 5 W1 / W2 / W3 Three-channel initial weights 0.40 / 0.30 / 0.30 Power 5 S_threshold Comprehensive scoring authorization threshold 60 points Power 6 w_max Confidence weight of real estate ownership certificate 1.0 Power 4 w_mid Confidence weight of registered lease contracts 0.80 Power 4 w_low Confidence weight of residential trace materials 0.55 Power 4 γ_type Material type diversity reward coefficient n_types=2:0.08, n_types≥3:0.12 Power 4 T_valid Validity period of certification materials Set according to material type Power 4 w_channel(LBS) Authentication completion rate LBS channel weight 0.35 Power 6 w_channel(OCR) Authentication completion rate OCR channel weight 0.40 Power 6 w_channel (guarantee) Certification completion guarantee channel weight 0.25 Power 6 θ_low Functional gradient authorization low threshold 0.4 Power 6 θ_mid Threshold in Functional Gradient Authorization 0.7 Power 6 N_max Maximum number of authenticated accounts on the same device 2 Power 7 η_group Same equipment rating discount factor 0.6 Power 7 θ_dev Device fingerprint Jaccard similarity threshold 0.85 Power 7 M_threshold Community name matching threshold 0.85 Power 4 η_recovery Guarantee rights restoration discount factor 0.6 Power 3 Single guarantee credit limit Maximum credit score for a single guarantee relationship 30 points Power 3 Attached Figure Description

[0027] Figure 1 This is a block diagram of the overall architecture of the community resident location authentication risk control system with multi-path fusion according to the present invention; Figure 2 Flowchart for LBS location verification over multiple consecutive days; Figure 3 Flowchart for intelligent de-identification verification of invoices using OCR; Figure 4 A flowchart for multi-dimensional weighted fusion scoring. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] like Figure 1 As shown, the multi-path fusion community resident location authentication risk control system provided by this invention consists of four logical layers: user terminal layer, authentication channel layer, fusion scoring engine layer, and risk control authorization layer. These layers are connected from bottom to top to form a complete data processing and decision-making chain. Information is transmitted between each layer through three key data flows: (a) The user terminal layer distributes the raw authentication data to the three parallel channels of the authentication channel layer; (b) The authentication channel layer transmits the credit scores S_LBS, S_GUARANTEE, and S_OCR calculated independently by each channel to the fusion scoring engine layer; (c) The fusion scoring engine layer submits the comprehensive location credit score S_total to the risk control authorization layer for the final permission decision.

[0030] Furthermore, the user terminal layer, which runs on the user's smartphone, is the front-end entry point for system-user interaction and contains four core functional modules. (1) GPS positioning data acquisition module: After the user authorizes location permissions, it continuously collects device geographic coordinate data at a preset frequency and uploads it to the server-side location database as the input data source for the LBS location verification channel. (2) Ticket image shooting / selection module: It provides two image acquisition methods: system camera call and local album selection, and supports users to shoot or select certified ticket images containing community information. (3) Guarantee request initiation module: Users can send a guarantee relationship establishment request to neighboring users who have passed full authentication. This module encapsulates the guarantee application information in a structured manner and submits it to the neighbor guarantee credit channel on the server side. (4) Local privacy desensitization module: This module is a lightweight data processing component built into the client. Before the ticket image is uploaded to the server, it performs privacy-sensitive entity detection and automatic mask replacement on the complete text identified by OCR through a local named entity recognition model, and only retains the community name field in non-sensitive information for network transmission. This "local desensitization, only transmitting the desensitization result" architecture design ensures that the complete privacy ticket image and all the personal information it contains never leave the user's terminal device.

[0031] Furthermore, the authentication channel layer is the core authentication engine of the system, containing three parallel authentication channels that are independent in terms of technical implementation and data source. These three channels can be started simultaneously or completed step-by-step by the user, each independently calculating and outputting a credit score. Channel one is an LBS continuous multi-day location verification channel. Through long-term statistical analysis of the user's mobile phone GPS location data, it comprehensively judges whether the user is continuously and regularly located within the target community space using three indicators: the percentage of effective dwell days, the location distribution entropy value, and trajectory smoothness detection. Channel two is a neighborhood guarantee credit channel, which uses the community's familiar relationship network to build a distributed trust system. Residents who have completed full authentication can act as guarantors to provide credit endorsement for new users or users who have not completed authentication. When calculating the guarantee score, factors such as the guarantor's own credit rating, historical guarantee performance rate, guarantee circle density, and anomaly detection in the guarantee relationship graph are comprehensively considered. Channel three is the intelligent de-identified verification channel for invoices. It takes a user-submitted image of an invoice containing community information as input. After local OCR recognition and privacy de-identification preprocessing on the client side, only the de-identified community name is uploaded to the server for fuzzy matching verification. Simultaneously, a sub-model for invoice authenticity authentication is invoked to perform anti-counterfeiting analysis on the image. The design of the three channels follows the principle of "mutual independence, mutual complementarity, and cross-validation"—the LBS channel verifies the physical location dimension, the invoice channel verifies the document evidence dimension, and the guarantee channel verifies the social relationship dimension. Failure in any one dimension does not directly lead to overall authentication failure; instead, a comprehensive balance is achieved through a fusion scoring engine layer.

[0032] Furthermore, the multi-dimensional weighted fusion scoring engine layer is located between the authentication channel layer and the risk control authorization layer, undertaking the functions of aggregation, verification, and fusion calculation of multi-dimensional scoring data. This layer adopts a four-stage pipeline processing architecture: the first stage is abnormal channel detection, which identifies abnormal deviation channels that significantly deviate from other channels by calculating the pairwise deviations of the three input scores; the second stage is dynamic weight adjustment, which attenuates and compresses the weights of abnormal deviation channels and distributes the reduced weights proportionally to normal channels, sets the weights of cheating channels to zero, and does not participate in weight allocation for incomplete channels; the third stage is integrity reward, which applies an integrity reward coefficient β (β>1.0) to the final score when the user completes all three authentication channels; the fourth stage is comprehensive score calculation, which calculates the comprehensive local credit score S_total according to the adjusted dynamic weights and integrity reward coefficient, and the result is included in a preset scoring range.

[0033] Furthermore, the risk control authorization layer receives the comprehensive local credit score S_total output by the fusion scoring engine layer and executes tiered credit granting decisions and continuous behavior monitoring. This layer adopts a five-in-one risk control closed-loop mechanism of "access decision-level classification-behavior monitoring-dynamic upgrade / downgrade-guarantee linkage". The access decision node compares S_total with the preset authorization threshold S_threshold to form a binary decision branch of "pass-restricted". Authenticated users are divided into multiple progressively increasing credit levels, and differentiated platform function permissions are granted according to the L0-L3 level gradient, realizing the tiered authorization concept of "the more authenticated, the more trustworthy, the more free". After obtaining authorization, the system continuously collects and analyzes four types of user behavior data: transaction behavior, social behavior, content behavior, and authentication behavior, and calculates a comprehensive behavior health score. When the behavior health score continues to decline or triggers the violation threshold rule, the system automatically performs a credit level downgrade operation. Downgraded users can apply for credit restoration by supplementing authentication or entering a repair observation period. Regarding the linkage of guarantees, when a guaranteed party commits a violation, the system will deduct credit limits from the guarantor's account according to the severity of the violation. When the accumulated deductions reach the upper limit, the system will automatically suspend the guarantor's guarantee privileges and lower their credit rating. Through this closed-loop mechanism, the system achieves dynamic and continuous management of the authentication status of community residents, rather than a one-time static judgment.

[0034] Example 1: Implementation of LBS Continuous Multi-Day Location Verification

[0035] Reference Figure 2 (LBS Verification Flowchart) This embodiment demonstrates the complete implementation process of the LBS location verification channel. The specific parameter settings are as follows:

[0036] The observation time window T_window = 14 days, the spatial threshold R_geo = 3000 meters (centered on the target area), and the GPS positioning data acquisition frequency is once every 30 minutes. The positioning distribution entropy value H_loc is calculated as H_loc = -∑(p_i×log(p_i)), where p_i is the proportion of the user's positioning points in the i-th geographical region (divided by administrative region or custom grid).

[0037] Step S101: After the user authorizes location permissions, the client collects GPS positioning data every 30 minutes and uploads it to the server-side location database.

[0038] Step S102: The server counts the number of valid location points N_valid that the user has been within a 3000-meter radius of the target cell center in the past 14 days, and calculates the percentage of valid stay days R_stay = D_valid / D_total, where D_valid is the number of days with at least one valid location point.

[0039] Step S103: Calculate the entropy value H_loc of the user location distribution. The higher the entropy value, the wider the user's movement range and the more discrete the distribution.

[0040] Step S104: Perform trajectory smoothness detection: Calculate the displacement distance Δd and the moving speed v = Δd / Δt between adjacent positioning points. When the speed exceeds the upper limit of reasonable human moving speed V_max (e.g., 350km / h) or there is an instantaneous displacement jump that does not conform to the laws of physics, it is marked as GPS simulation cheating.

[0041] Step S105: Comprehensive Judgment: If R_stay≥0.6 and H_loc≤T2 and no GPS cheating is detected, it is judged as "fixed-point permanent residence", S_LBS=80~100 points; if R_stay≥0.6 but H_loc>T2, it is judged as "permanent residence but wide activity range", S_LBS=60~79 points; if R_stay<0.6, it is judged as "non-permanent residence", S_LBS=0~59 points; if GPS cheating or cross-city drift characteristics are detected (H_loc>T3), S_LBS=0 points and a risk control alarm is marked.

[0042] Example 2: Implementation Method of Intelligent De-identification Verification of Invoices via OCR

[0043] Reference Figure 3 (Flowchart of Invoice OCR Verification) This example demonstrates the complete implementation process of the invoice OCR verification channel.

[0044] Step S201: The user selects or takes a picture of a ticket containing community information on the client (supported real estate ownership certificates, housing rental contracts, property fee payment vouchers, water, electricity, gas and other utility bills, etc.). The client preprocesses the image (compresses it to an appropriate resolution, corrects perspective, and enhances contrast).

[0045] Step S202: The client calls the local AI model to perform text recognition on the image and outputs the complete recognized text.

[0046] Step S203: Client-side local privacy masking: The following privacy entity categories are detected in the text using a named entity recognition model: PERSON (name), ID_NUM (ID card number), ROOM_NUM (detailed room number), and PHONE (phone number). The detected privacy entities are replaced with mask characters (such as "\*\*\*"). The masked text retains only non-sensitive information, especially the community name (COMMUNITY_NAME entity). Throughout the process, the complete privacy ticket image remains on the user's terminal.

[0047] Step S204: The client uploads only the anonymized cell name field to the server via the network. The server performs a double fuzzy match between the extracted cell name and the target cell database (containing a standard cell name and alias mapping table). It calculates string similarity using the Levenshtein Distance algorithm and combines a semantic similarity model (such as BERT-wwm, Bidirectional Encoder Representations from Transformers with Whole Word Masking) to handle different representations of the same cell (e.g., "Sunshine Garden" and "Sunshine Flower Garden"). If the matching score exceeds the threshold M_threshold=0.85, the verification is considered successful; otherwise, a re-upload is prompted.

[0048] Step S205: The server calls the document authenticity verification sub-model to perform authenticity checks on the uploaded image. This includes: image texture consistency analysis (detecting splicing / composite boundaries), font edge jaggedness feature analysis (detecting inconsistent font rendering when copied and pasted in Photoshop), seal red area color gradation uniformity analysis (detecting seal cutout replacement), and Exchangeable Image File Format (EXIF) metadata verification (detecting the reasonableness of the shooting device, time, and geographical location). If the detection is genuine, it is marked as passed, S_OCR=100; if the detection is questionable, it is marked as requiring manual review, S_OCR=50; if the detection is counterfeit, S_OCR=0 and a risk control alarm is triggered.

[0049] Example 3: Implementation of Multi-Dimensional Weighted Fusion Scoring and Tiered Credit Risk Control

[0050] Reference Figure 4 (Flowchart of Fusion Scoring) This embodiment demonstrates the implementation process of the weighted fusion scoring engine and the tiered credit risk control mechanism. Key parameters are set as follows: initial weights W1=0.40 (LBS), W2=0.30 (guarantee), W3=0.30 (OCR); abnormal deviation threshold Δ=40; attenuation factor γ=0.3; integrity reward coefficient β=1.1; authorization threshold S_threshold=60.

[0051] Step S301: Collect the score outputs S_LBS, S_GUARANTEE, and S_OCR from the three authentication channels, and normalize each score to the range [0, 100]. For authentication channels that the user has not completed, the corresponding score is recorded as 0.

[0052] Step S302: Perform abnormal channel detection—calculate the pairwise deviation of the three scores d_ij = |S_i - S_j|. If the deviation of a certain channel from the other two channels exceeds the preset threshold Δ=40, mark the channel as an "abnormal deviation channel".

[0053] Step S303: Perform dynamic weight adjustment: (a) If channel i is marked as abnormal deviation, its weight Wi is multiplied by the decay factor γ=0.3, and the reduced weight is distributed proportionally to other normal channels; (b) If channel i is detected as cheating, Wi is directly set to 0; (c) If the user completes all three channels (all three scores are greater than 0), it is multiplied by the integrity reward coefficient β=1.1.

[0054] Step S304: Calculate the comprehensive local credit score S_total = (W1'×S_LBS + W2'×S_GUARANTEE+ W3'×S_OCR) × β.

[0055] Step S305: Compare S_total with the preset authorization threshold S_threshold=60. If S_total≥60, automatically grant permissions to residents of the community and classify credit levels according to Table 1; if S_total<60, enter restricted mode.

[0056] Table 1. Credit Rating and Authority Comparison Table grade S_total range Items that can be published (daily limit) You can initiate a purchase request. Daily transaction limit Guarantee upper limit Priority recommendation L0 <60 0 no ¥0 0 people no L1 60~74 3 items no ¥500 0 people no L2 75~89 10 items yes ¥2,000 3 people no L3 ≥90 30 items yes ¥10,000 10 people yes Guarantee-linked risk control rules: When the guaranteed party violates regulations, the system deducts credit points from both the guaranteed party's and the guarantor's credit limit according to the severity of the violation—minor violations deduct 10 points from the guaranteed party and 5 points from the guarantor; general violations deduct 25 and 15 points respectively; serious violations deduct 50 and 30 points respectively; and extremely serious violations deduct 100 and 60 points respectively. When a single guarantor's accumulated deductions reach the maximum credit limit, their guarantee privileges will be automatically suspended, and their own credit rating will be lowered by one level.

[0057] Example 4: Implementation of Intelligent Invoice Review Using a Multimodal Large Language Visual Model

[0058] This embodiment demonstrates the use of a multimodal large language visual model with image and text understanding capabilities, referring to the aforementioned... Figure 1 System architecture and Figure 3 The invoice OCR verification process is a complete technical process for intelligently reviewing different types of certified invoices.

[0059] Taking renter Zhao as an example, he submitted two documents through the community APP for local verification: a registered rental contract and the utility bill for the most recent month. Step S401: After preprocessing and locally desensitizing the ticket image, the client uploads it to the server-side AI review module.

[0060] Step S402: The multimodal large language visual model performs layout analysis on the lease contract image, identifying structured sections such as the contract title area, the information area of ​​Party A and Party B, the location area of ​​the property, the lease term area, and the signature area.

[0061] Step S403: The model automatically determines the document type as "House Rental Contract" based on the layout features, calls the corresponding key field extraction prompts, and extracts structured information: lessor's name (already anonymized as "Qian*"), lessee's name (already anonymized as "Zhao*"), location of the leased property (XX City XX District XX Road XX Road No. XX Community XX Building XX Room XX), lease term (January 1, 2026 to January 1, 2027), and registration number.

[0062] Step S404: Perform semantic cross-validation on the model: (a) Verify whether the name "XX Community" matches the community name in the address - matches; (b) Verify whether the contract validity period covers the current date - valid; (c) Verify whether the filing number format conforms to the coding standard of the local housing rental management department - conforms.

[0063] Step S405: The model performs the same process on the water and electricity bill images, extracting the community name field. The system verifies whether the community names in the two documents are consistent—both point to "XX Community"—consistency verification passes. n_types=2, triggering the type diversity reward mechanism, γ_type=0.08.

[0064] Step S406: The model performs independent anti-counterfeiting auxiliary analysis on the stamp and signature areas of the two documents—the lessor's signature and stamp position on the lease contract are normal, and the red official stamp on the utility bill has clear edges and uniform color distribution. All anti-counterfeiting auxiliary scores are normal.

[0065] Step S407: The server performs fuzzy matching of the extracted cell name "XX cell" with the target cell database using the edit distance method. The matching degree of 0.96 exceeds the preset threshold of 0.85, and the match is successful.

[0066] Step S408: Comprehensive score calculation: Lease contract verification score S_contract=100, water and electricity bill verification score S_bill=100; material comprehensive score S_OCR_raw = (100×0.80+100×0.55) / (0.80+0.55) = 100; superimposed type diversity bonus S_OCR_final = min(100, 100×(1+0.08×2)) = 100.

[0067] Step S409: After the inference process is complete, the server immediately clears the image data from the GPU memory and system memory. The audit log only records the anonymized structured fields (community name, invoice type, verification result, etc.), and does not retain the original invoice image file.

[0068] Example 5: Implementation of Device Fingerprint Anti-fraud Method

[0069] This embodiment refers to the foregoing. Figure 1 The system architecture demonstrates the complete process by which the device-account association graph effectively identifies organized fraudulent activities such as "authentication intermediaries."

[0070] Scenario setting: Suspected user Zhou logged into 5 different WeChat accounts using the same smartphone within 48 hours, and submitted local authentication applications to "Sunshine Garden Community" for each of the 5 accounts. Each account uploaded a picture of a real estate ownership certificate with a different name (which was later verified to be a Photoshop forgery).

[0071] Step S501: Each time a user submits an authentication request, the system automatically collects device fingerprint information from the client, including device model, operating system version, screen resolution, sensor list hash value, network type, system startup timestamp, and application installation list hash value. A unique device identifier F_dev is generated using a multi-feature fusion hash function—Secure HashAlgorithm 256-bit (SHA-256). The same device fingerprint is collected in all 5 authentication submissions. Five new records are added to the device-account association graph.

[0072] Step S502: Risk control detection rule one (multi-account authentication detection on the same device) triggered: If the fingerprint F_dev of the same device is associated with 5 different accounts within 72 hours (exceeding the threshold N_max=2) and performs local authentication in the same community "Sunshine Garden Community", the system determines that there is a risk of batch authentication on the same device and generates a risk control alarm event.

[0073] Step S503: The system performs the following automated processing measures: (a) The LBS location credit scores of the 5 related accounts are uniformly reduced by a discount factor of η_group=0.6; (b) All authentication applications of the 5 accounts on the same day are suspended from automatic approval process and forcibly transferred to the manual review queue; (c) The device fingerprint is written into the device fingerprint observation list and 72-hour key monitoring is initiated.

[0074] Step S504: Manual review revealed that the Exchangeable Image File Format (EXIF) metadata of the five invoice images all originated from the same camera (same serial number) and the same shooting time. Further analysis revealed obvious signs of Photoshop compositing in the texture features and font rendering of the invoice images—abnormally jagged font edges and an overly uniform red tone in the seal. All five documents were confirmed as forged, and the five accounts were uniformly marked as authentication fraud and permanently banned.

[0075] Step S505: The device fingerprint is moved from the observation list to the official blacklist. The system adds a device fingerprint verification step to the global user registration and login process—for any newly registered or logged-in account, if the Jaccard similarity coefficient between its device fingerprint and an existing fingerprint in the blacklist exceeds the threshold θ_dev=0.85, its initial credit score will be automatically reduced to the warning preset value of 40 points during the account creation stage, and a 90-day behavior observation period will be set.

[0076] Example 6: Implementation of Functional Gradient Authorization Driven by Authentication State

[0077] This embodiment refers to the foregoing. Figure 1 The system architecture, using four users at different authentication progress stages as examples, fully demonstrates how the authentication completion metric A_complete drives the tiered opening of platform function permissions. The authentication completion metric is defined as A_complete = Σ(w_channel_i×flag_i), where w_channel_i takes values ​​of 0.35 for LBS location verification, 0.40 for invoice OCR verification, and 0.25 for neighborhood guarantee credit granting, and flag_i represents the completion status of each channel (0 or 1).

[0078] User Wang (A_complete=0, no authentication initiated): Only homepage product browsing (100 items per day) and community activity viewing (50 items per day) are enabled; posting and communication are not allowed. The app continuously displays the authentication guide banner "Complete community authentication to unlock neighbor transactions."

[0079] User Chen (A_complete=0.35, only LBS location verification completed): Added the function to post purchase requests (3 requests per day) and basic private messaging function (5 messages per day). The system displays a prompt at the relevant function entry point: "Complete the invoice verification again to post your idle items to earn pocket money."

[0080] User Lin (A_complete=0.75, completed LBS+ticket OCR): Added the function of posting idle items (10 items per day), the collection function, and the function of posting / interacting in the neighborhood circle. He enjoys the right to be prioritized in the search ranking for the same community.

[0081] User Zhang (A_complete=1.0, all three certifications completed): Has access to all community features—posting unwanted items (30 items daily), initiating neighborhood mutual aid activities, acting as a guarantor (maximum of 10 guarantors), priority recommendation in search rankings, and the exclusive "Full Certification Expert" identity tag. Enjoys the highest level of transaction limits and all platform benefits.

[0082] The core concept of this tiered authorization mechanism is "the more certified, the more trustworthy, the more free"—by precisely linking the certification progress with functional permissions, it minimizes the negative impact of mandatory certification on the user's cold start experience while ensuring the security of platform transactions.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-path fusion community resident location authentication risk control method, characterized in that, include: Step S1: Collect LBS positioning data from user terminals. Based on the preset observation time window and spatial threshold, calculate the percentage of effective stay days and the location distribution entropy value of users. Perform trajectory smoothness detection to identify GPS simulation cheating behavior and output the location credit score S_LBS. Step S2: Establish neighborhood guarantee credit relationship, allocate guarantee credit limit and guarantee number limit to certified users, bind the risk control status of the guaranteed person's account to the guarantor's credit limit, deduct the guarantor's credit limit proportionally when the guaranteed person violates the rules, maintain the guarantee relationship graph to detect group cheating behavior, and output guarantee credit score S_GUARANTEE; Step S3: Receive the ticket image containing community information uploaded by the user, perform OCR text recognition and automatic privacy information desensitization processing on the local terminal, extract only the community name field and upload it to the server for database matching and verification, and verify the authenticity of the ticket image, and output the ticket verification score S_OCR; Step S4: Input S_LBS, S_GUARANTEE, and S_OCR into the multi-dimensional weighted fusion scoring engine, perform abnormal channel detection and dynamic weight adjustment, and calculate the comprehensive local credit score S_total = W1'×S_LBS + W2'×S_GUARANTEE+ W3'×S_OCR, where Wi' is the dynamically adjusted weight coefficient. Step S5: Compare the comprehensive local credit score S_total with the preset authorization threshold. When S_total reaches the threshold, automatically grant the user community resident privileges and classify the user into different credit levels according to the value range of S_total. Each level corresponds to different functional permissions and transaction limits. Step S6: Continuously monitor user behavior data. When behavior indicators deteriorate, automatically lower the credit rating and restrict functional permissions to form a dynamic risk control closed loop.

2. The method according to claim 1, characterized in that, The formula for calculating the location distribution entropy value H_loc in step S1 is H_loc = -∑(p_i×log(p_i)), where p_i is the proportion of the user's location points in the i-th geographical region. When the proportion of effective stay days is ≥ the first threshold T1=0.6 and H_loc≤ the second threshold T2=1.2, the user is determined to be a fixed-point permanent user. When H_loc> the third threshold T3=2.0 and a large-scale cross-city displacement is detected, the user is determined to be a cross-city drifting account and marked as a suspected itinerant vendor. The trajectory smoothness detection includes calculating the displacement distance and movement speed between adjacent location points. When the movement speed exceeds the preset upper limit of reasonable human movement speed of 350 km / h or an instantaneous displacement jump that does not conform to the laws of physical motion occurs, it is determined to be GPS simulation cheating behavior.

3. The method according to claim 1, characterized in that, The guarantee relationship graph analysis in step S2 includes: when a closed loop of guarantee relationship is detected or multiple guaranteed parties of the guarantor commit a batch of violations within a preset time window, a group cheating warning mechanism is triggered, the guarantee permissions of all users within the closed loop are suspended, the credit rating of the guarantor is reduced by one level, and an observation period is set. Only if there are no violations during the observation period can the guarantee permission be restored. After restoration, the upper limit of the number of guarantors and the credit limit are reduced according to a preset ratio.

4. The method according to claim 1, characterized in that, The automatic desensitization of privacy information in step S3 is achieved through a Named Entity Recognition (NER) model deployed on the local terminal. The identified privacy entity categories include name, ID card number, detailed room number, and phone number. After replacing the detected privacy entities with mask characters or deleting them directly, only the non-sensitive community name field is uploaded to the server. The authenticity verification of the invoice includes texture feature analysis, font edge jaggedness consistency detection, seal color distribution analysis, and Exchangeable Image File Format (EXIF) metadata verification of the uploaded image to identify fake invoices synthesized by Photoshop or downloaded from the network. A multimodal large language visual model is used as the core review engine for invoices. The model automatically determines the invoice type and adapts the extracted template through layout analysis. It supports multiple types of invoices such as real estate ownership certificates, lease contracts, property fee vouchers, and water, electricity, and gas bills. Differentiated confidence weights are assigned according to the material type. A type diversity reward is introduced for multiple different types of materials. The comprehensive confidence S_OCR_final = min(100, S_OCR×(1+γ_type×n_types)). The invoice verification score S_OCR is the comprehensive confidence score S_OCR_final after the above multimodal review and material differential confidence allocation.

5. The method according to claim 1, characterized in that, The dynamic weight adjustment rules in step S4 include: (a) when the deviation of a single channel score from the scores of the other two channels exceeds a preset threshold Δ, the weight of that channel is reduced; (b) when a channel is suspected of cheating, the weight of that channel is reset to zero; (c) when a user completes all three authentication channels, the user is multiplied by the integrity reward coefficient β, where β > 1.

0.

6. The method according to claim 1, characterized in that, The credit rating mentioned in step S5 includes at least four levels: Level L0 is an uncertified state, which can only browse but not post or communicate; Level L1 is basic certification, which can post items and communicate, and set a daily transaction limit; Level L2 is the standard certification, which allows users to post items, initiate purchase requests, and act as guarantors, but the maximum number of guarantors is limited. Level L3 is advanced certification, granting full functional permissions, maximum transaction limits, and priority recommendation weight. It also includes dynamic authorization of community functions driven by certification completeness, defining the certification completion index A_complete = Σ(w_channel_i×flag_i), where w_channel_i represents LBS location verification (0.35), invoice OCR verification (0.40), and neighborhood guarantee credit (0.25), and flag_i represents the completion status of each channel. The system unlocks community functions based on the A_complete gradient.

7. The method according to claim 1, characterized in that, It also includes an authentication anti-fraud mechanism that associates device fingerprints with multiple accounts: the system collects multi-dimensional attributes of terminal devices and generates a unique device identifier F_dev through multi-feature fusion hashing; the system maintains a device-account association graph, and when the same device fingerprint is associated with more than N_max accounts in the same cell for authentication within a preset time window, the LBS location credit score of the relevant accounts is reduced and forced to be transferred to manual review; when multiple accounts associated with the same device fingerprint submit authentication materials to different cells within the time window, it is marked as cross-cell authentication fraud and automatic approval is suspended.

8. The method according to claim 1, characterized in that, It also includes a multimodal behavioral risk control credit dynamic management mechanism: the system continuously collects users' transaction behavior data, social behavior data, content behavior data, and authentication behavior data, and calculates the behavioral health score B_health=w_trade×S_trade+w_social×S_social+w_content×S_content+w_auth×S_auth according to a preset period. Where S_trade is based on the performance rate and dispute rate, S_social is based on the message response rate and the number of times blacklisted, S_content is based on the publication compliance rate and the number of times reported, and S_auth is based on the frequency of material submission and the number of times rejected. w_trade, w_social, w_content, and w_auth are preset weights for each dimension, and the sum of the weights of the four dimensions is 1. When B_health is below the warning line for several consecutive periods, the credit level is automatically downgraded. After downgrading, the user enters a cooling-off period and needs to maintain behavioral improvement before applying for level restoration. When restored, the user can be restored to the next level below the level before downgrading, forming a complete risk control closed loop of authentication-behavior linkage.

9. A multi-path fusion community resident location authentication risk control system, characterized in that, include: The LBS continuous multi-day location verification module is used to collect user terminal location data, statistically analyze the proportion of effective dwell days and the location distribution entropy value based on preset time windows and spatial thresholds, perform trajectory smoothness detection to identify GPS simulation cheating behavior, and output location credit score. The neighborhood guarantee credit module manages the credit relationship between guarantors and guaranteed parties, allocates guarantee credit limits and guarantee frequency caps, links guaranteed party violations with guarantor credit limit deductions, maintains the guarantee relationship graph, detects group fraud, and outputs guarantee credit scores. The invoice OCR intelligent desensitization verification module, deployed locally on the user's terminal, performs OCR text recognition and automatic privacy desensitization on invoice images, extracts only the community name field for database matching verification, and authenticates invoice images, outputting invoice verification scores. The multi-dimensional weighted fusion scoring engine receives scores from the three modules, performs abnormal channel detection and dynamic weight adjustment, and calculates a comprehensive local credit score. The tiered credit risk control module classifies users' credit levels based on the comprehensive score, assigns corresponding functional permissions and transaction limits, and continuously monitors user behavior data to dynamically upgrade or downgrade levels.

10. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the multi-path fusion community resident location authentication risk control method as described in any one of claims 1-8.