Idle article intelligent matching and cross-region publishing interception method

Abstract

The application discloses an idle article intelligent matching and cross-region publishing interception method, and belongs to the technical field of community electronic commerce and location service. The application generates a 3-kilometer electronic geographic fence with the user authentication cell coordinates as the center, and establishes a four-level publishing permission system from L0 to L3. The publishing link calculates the spherical distance in real time through the Haversine formula, and automatically intercepts the cross-region listing. The matching link performs the same cell priority four-level distance sorting, and fuses the distance, time efficiency, price and credit multi-factor calculation comprehensive sorting score. The risk control link detects the cross-cell batch publishing behavior through the sliding time window multi-dimensional detection, and identifies the same goods through the text semantic similarity and image perception hash dual judgment, and calculates the merchant suspicion score through the weighted calculation, and links the dynamic ban and blacklist mechanism. The application effectively solves the problem of cross-city goods spreading of idle transaction platform vendors, and guarantees the localization and authenticity of community idle transaction.

Description

Technical Field

[0001] This invention belongs to the field of community e-commerce and location-based service (LBS) technology, specifically involving a method for intelligent matching and cross-regional posting interception of idle goods. It is applicable to community neighborhood idle goods trading platforms, community group buying platforms, local O2O service platforms, and other Internet application scenarios that require posting control and intelligent matching based on geographical location. Background Technology

[0002] With the popularization of the sharing economy and circular economy concepts, online trading platforms for idle goods (such as Xianyu and Zhuanzhuan) have developed rapidly. However, existing idle goods trading platforms have the following prominent problems in actual operation: (1) The proliferation of cross-city sales by vendors has overwhelmed genuine individual sellers. Existing platforms do not impose any restrictions on posting locations, allowing users to post goods anywhere. This has led to a large number of professional vendors mass-producing goods across the country, with the same vendor posting the same product in dozens of cities simultaneously, severely crowding out the display space for genuine individual sellers. When buyers search for "nearby secondhand goods," the results list is filled with cross-city advertisements from vendors, making it impossible to find truly nearby individual sellers. Research shows that on mainstream secondhand trading platforms, approximately 65% ​​of search results are for goods posted by professional vendors, while the actual exposure rate of genuine neighborhood transactions is less than 15%.

[0003] (2) Lack of community-level geographical location constraints. Although existing platforms offer a "sort by distance" function, this distance is calculated based on the user's current Global Positioning System (GPS) location, rather than the user's verified community / neighborhood location. Users' GPS locations can be changed at will (e.g., by using virtual location tools or simulators to modify GPS data), failing to form effective community-based geographical constraints. At the same time, buyers cannot prioritize viewing products posted by neighbors in the same community, and the trust basis of neighborhood transactions—the convenience of face-to-face transactions brought by geographical proximity and the sense of community belonging—is not fully utilized.

[0004] (3) Lack of effective merchant identification and interception mechanisms. Existing platforms mainly rely on user reports and manual review to identify professional merchants, which has problems such as poor timeliness (average report processing cycle exceeds 48 hours), limited coverage (only about 12% of merchants are successfully identified), and high costs. There is a lack of automated anomaly detection algorithms based on publishing behavior patterns, which makes it impossible to identify and intercept merchants' behavior of mass distribution in multiple communities in a short period of time, resulting in merchants having already completed a large number of distributions before being identified.

[0005] (4) Inefficient management of publishing permissions. Existing platforms typically manage publishing permissions using a binary model of "all allowed" or "all prohibited," lacking a hierarchical permission system based on user authentication status, credit rating, and behavioral history. All authenticated users enjoy equal publishing permissions, making it impossible to achieve differentiated management of high-risk users or provide differentiated incentives for high-credit users.

[0006] (5) The matching algorithm ignores geographical proximity. The existing platform's search ranking algorithm mainly relies on keyword matching degree and posting time, and lacks a mechanism to use geographical distance as a core ranking factor. Even if users intend to find nearby idle items, the system will not actively prioritize displaying goods from the same or neighboring communities, resulting in low efficiency of neighborhood transaction matching.

[0007] In summary, there is an urgent need for a localized intelligent matching and cross-regional posting management method for idle goods that can effectively constrain the geographical location of postings, prioritize matching with neighboring products in the same community, automatically identify and block professional merchants, and implement hierarchical permission control. Summary of the Invention

[0008] This invention aims to solve the following technical problems: (1) How to generate electronic geofences based on user-authenticated communities to restrict the geographical scope of product postings, block cross-regional / cross-city posting behavior from the source of posting, and ensure the authenticity of community idle transactions; (2) How to achieve multi-level distance sorting from nearest to farthest within the same community, and combine multiple factors such as distance, time, price and credit to improve the accuracy and efficiency of neighborhood idle matching; (3) How to automatically identify professional merchants through multidimensional behavior analysis, and use text semantic similarity and image perceptual hashing to identify the same products, so as to realize real-time detection and tiered automatic blocking; (4) How to construct a four-level hierarchical publishing permission system from L0 to L3 to achieve refined publishing control over users with different authentication levels and credit status; (5) How to use full-link geofencing constraints to integrate geolocation verification into all stages of publishing, matching, and risk control, forming a closed-loop community territorial management system.

[0009] Given the current shortcomings of online trading platforms for idle goods, such as rampant cross-city distribution by vendors, lack of community-level geographical constraints, lack of effective automatic merchant identification mechanisms, crude management of posting permissions, and matching algorithms that ignore geographical proximity, this invention provides a method for intelligent matching and cross-regional posting interception of idle goods. This method can curb cross-regional distribution by vendors from the source of posting, improve the authenticity and efficiency of neighborhood matching, automatically identify and intercept professional merchants, and achieve refined hierarchical permission control.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent matching and cross-regional posting interception of idle items, characterized by the following steps: Step S1, Geofence Generation: Obtain the geographic coordinates of the user's authenticated cell, and generate an electronic geofence with the coordinates as the center and a preset radius. The geofence is defined as a geographic range constraint area centered on the authenticated cell. When a user has multiple authenticated cells, generate a corresponding independent geofence with each authenticated cell as the center. The user can publish products within any geofence. Step S2, Location Verification and Cross-Regional Interception: Receive user's product posting request, extract product posting location coordinates, calculate the spherical distance between the location and the certified cell coordinates, and determine whether the spherical distance is less than or equal to a preset radius; if it is greater than the preset radius, automatically intercept the posting request and return an interception prompt message; if it is less than or equal to the preset radius, proceed to step S3. Step S3, Hierarchical Publishing Permission Verification: Query the user's current publishing permission level, which includes four levels: L0 Guest Level, L1 Authenticated Resident Level, L2 Credit Resident Level, and L3 Restricted User Level. If the user is at level L0 or L3, the publishing request is blocked. If the user is at level L1 or L2, the daily publishing quantity is further verified to see if it exceeds the publishing limit of the corresponding level. If it does not exceed the limit, proceed to step S4. Step S4, Abnormal Merchant Detection: Within a preset sliding time window T_window (default 24 hours), extract multi-dimensional features of user posting behavior. These features include four dimensions: abnormal posting frequency, homogeneity of products, address clustering degree, and location consistency. The scores of each dimension are normalized and then weighted and summed according to preset weight coefficients. The total score is calculated on a scale of 0-100. A merchant suspicion score is calculated based on the multi-dimensional features and their preset weight coefficients. When the merchant suspicion score exceeds a preset threshold, the merchant is identified as a professional merchant, and posting privileges are temporarily banned. All posted products are also removed from the platform and added to the blacklist candidate database. Step S5, localized intelligent matching: receiving a buyer's product search request, obtaining the coordinates of the buyer's authenticated community, and executing a same-community priority matching strategy, which divides priorities into four levels in ascending order of distance: the first priority is the same community (d=0km), marked with a "Neighbor" tag; the second priority is adjacent communities (0km<d≤1km); the third priority is surrounding communities (1km<d≤2km); the fourth priority is peripheral communities (2km<d≤3km). Products beyond the fence range (d>3km) will not be included in the sorting results, or will be demoted to the bottom of the city-wide list. The comprehensive sorting score is calculated according to the formula Score=α×(1-d / R)+β×Tnorm+γ×Pnorm+δ×Cnorm by integrating the distance weight α, release time freshness β, price rationality γ and seller credit score δ, and a sorted result list is generated and returned.

[0011] Further, the generation of the electronic geofence in step S1 specifically includes: The preset radius is a configurable parameter with a default value of 3000 meters; the geofence supports two geometric representations: a circular area and a polygonal area. The circular area is directly defined with the coordinates of the authenticated community as the center of the circle and the preset radius, and the polygonal area is an approximately circular polygon generated based on the coordinates of the authenticated community according to a preset number of vertices; When the user completes location authentication, the system automatically records the latitude and longitude coordinates of the authenticated community, and persistently stores the binding relationship between the user and the community in the data storage module, which serves as the reference coordinate for all subsequent geofence verifications; In the scenario of multiple authenticated communities, the user corresponds to an independent fence area for each authenticated community. When the spherical distance from the product release location to any authenticated community is less than or equal to the preset radius, the product is allowed to be released within that fence.

[0012] Further, the release location verification and cross-zone interception in step S2 specifically includes: The system preferentially obtains the current device positioning coordinates of the user as the product release location; if the user does not authorize positioning permission or the positioning information is unavailable, the system automatically uses the coordinates of the user's authenticated community as the default release location; calculates the spherical distance between the product release location and the coordinates of the authenticated community, and judges whether the spherical distance is less than or equal to the preset radius; if the spherical distance is greater than the preset radius, it is determined as cross-zone release, and the system automatically intercepts it and returns a cross-zone interception prompt; if the spherical distance is less than or equal to the preset radius, the product is allowed to enter the subsequent release process; for the scenario of multiple authenticated communities, the system traverses the fences corresponding to each authenticated community, and as long as the product release location falls within the range of any fence, it is determined to be within the fence and the release is allowed to proceed.

[0013] Further, the specific configuration of the hierarchical release permission system in step S3 is as follows: L0 Visitor Level: Users who have not completed local verification cannot post products, can only browse products throughout the city, have no community priority in sorting, and do not enjoy priority matching within the same community; L1 Certified Resident Level: Users who have completed local verification for at least one community can post products within the verified community's boundaries, with a daily posting limit of the first preset number, enjoy priority matching within the same community, and can view neighbor tags within the same community; L2 Credit Resident Level: Based on L1 level, users whose credit score reaches the preset credit threshold and have no violations in the past 30 days will automatically upgrade to L2 level; the daily posting limit is increased to the second preset number, and they can initiate neighborhood chats. Credit guarantee: Both parties receive credit score bonuses upon successful credit approval. L3 Restricted User Level: Users are temporarily banned from posting due to triggering risk control rules; all listed products are removed from the platform. Users automatically revert to L1 level after the ban period ends. Users who repeatedly trigger L3 level restrictions are placed on a watchlist and their subsequent detection thresholds are lowered. The upgrade / downgrade rules between these permission levels are as follows: Upgrading from L0 to L1 requires completing local authentication; upgrading from L1 to L2 requires a credit score reaching the threshold and no violations; downgrading from L1 or L2 to L3 requires triggering risk control rules; reverting from L3 to L1 requires the ban period to expire or a manual appeal to be approved.

[0014] Furthermore, the identification of identical or similar products in step S4 employs a dual determination mechanism of textual semantic similarity and image perceptual hashing, specifically including: Text similarity calculation: Extract product title and description text, generate semantic feature vectors through pre-trained language model or word frequency inverse document frequency vectorization, calculate the cosine similarity between the feature vectors of the two product texts, and determine text similarity when the cosine similarity is greater than or equal to a first preset threshold. Image similarity calculation: Extract product images, generate binary fingerprints using a perceptual hash algorithm, calculate the Hamming distance between the fingerprints of two images, and determine that the images are similar when the Hamming distance is less than or equal to a second preset threshold; Comprehensive judgment rule: When the text and images of two products are similar, the two products are judged to be the same or similar products; The identification results of the same or similar products are used simultaneously for calculating the merchant suspicion score in step S4 and sorting and deduplication in step S5.

[0015] Furthermore, the merchant suspicion score mentioned in step S4 is obtained by weighting and summing the scores of four dimensions (denoted as f1, f2, f3, and f4 respectively) based on the abnormality of posting frequency, the degree of homogeneity of products, the degree of address clustering, and the consistency of location. The scores are calculated using preset weighting coefficients w1=0.30, w2=0.25, w3=0.25, and w4=0.20. The suspicion score S = round(w1×f1 + w2×f2 + w3×f3 + w4×f4) is calculated using a score scale of 0-100. The default duration of the sliding time window T_window is 24 hours. The window moves forward 24 hours from the current time and is dynamically updated as real time progresses.

[0016] Furthermore, the duration of the temporary ban mentioned in step S4 is dynamically determined based on the merchant's suspicion score. A segmented, tiered ban strategy is adopted, dividing the suspicion score S into four risk ranges: (1) S<40 is the normal level, and no action is taken; (2) 40≤S<60 is the warning level, and a warning notification is pushed to the user for 24 hours; (3) 60≤S<80 is the traffic restriction level, and the posting limit is reduced to 1 item per day for 24 hours; (4) S≥80 is the ban level, which determines the merchant as a professional, revokes the certification and temporarily bans the posting permission, and all posted products are removed from the shelves. The preset threshold, the first preset duration, the second preset duration and the third preset duration are all configurable parameters. During the ban period, users can submit an appeal. If the appeal is approved by a human reviewer, the ban will be lifted and the user will be restored to L1 level. If the appeal is rejected, the ban period will be recalculated. Users identified as professional merchants are automatically added to the blacklist candidate database. Users identified two or more times are added to the official blacklist. If a newly registered account is associated with a blacklist user's device fingerprint, mobile phone number, or payment account, its initial credit score will be automatically lowered to a preset warning value.

[0017] Furthermore, the comprehensive ranking score in step S5 is obtained by weighting and summing the distance weight factor, the posting time freshness factor, the price reasonableness factor, and the seller credit score factor according to preset weight coefficients; the distance weight factor is calculated based on the spherical distance between the buyer's authentication area and the product posting area and the fence radius; the posting time freshness factor is calculated based on the product posting timestamp, the current timestamp, and the earliest posting timestamp within the fence; the price reasonableness factor is calculated based on the product price and the historical average transaction price of similar products; and the seller credit score factor is calculated based on the seller's current credit score and the maximum credit score.

[0018] On the other hand, the present invention provides an electronic device, characterized in that it includes: 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 aforementioned method for intelligent matching and cross-regional posting interception of idle items.

[0019] On the other hand, the present invention provides a computer-readable storage medium, characterized in that it stores executable instructions thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for intelligent matching and cross-regional posting interception of idle items.

[0020] Compared with the prior art, the beneficial effects of the present invention are reflected in the following five aspects: (1) Curbing cross-city distribution by vendors from the source. By generating a 3km electronic geofence centered on the certified community, the location of goods is verified in real time during the posting process, and cross-regional postings are automatically blocked, thus technically preventing professional vendors from distributing goods in bulk nationwide. The coordinates of the certified community, as an immutable "digital anchor point," effectively prevent GPS spoofing and virtual location bypassing. Experimental data shows that this mechanism can reduce cross-regional posting behavior by about 92%, and increase the exposure rate of real neighborhood products from 15% to 78%.

[0021] (2) Significantly improve the authenticity and efficiency of neighborhood matching. The four-level distance sorting strategy, prioritizing products from the same community and from nearest to farthest, ensures that buyers see products posted by their neighbors first. The distance weight factor α=0.50 makes it dominant in the overall ranking. The distance factor for products from the same community is set to the maximum value of 1.0. Combined with the "neighbor" tag and the self-pickup bonus mechanism, it makes full use of the community's trust base and the convenience of face-to-face transactions to promote neighborhood transactions. Compared with the traditional city-wide sorting, the proportion of products from the same community in the top 10 search results has increased from less than 5% to about 60%.

[0022] (3) Automated and multi-dimensional merchant identification capabilities. The system uses a sliding time window (T_window) to monitor cross-cell posting behavior in real time, employs dual judgment of text semantics and image perceptual hashing to identify identical products, and detects abnormal posting frequency. These three dimensions are used to comprehensively determine professional merchants. Compared to the traditional method relying on user reports, the merchant identification rate has increased from approximately 12% to approximately 95%, the average identification time has been reduced from 48 hours to less than 30 seconds, and the entire blocking process is automated.

[0023] (4) Refined hierarchical access control. The L0-L3 four-level access control system combines local authentication status, credit score, and risk control triggering to achieve differentiated control over different user groups. The L0 level (no posting for visitors), L1 level (limited posting), L2 level (extended posting), and L3 level (prohibited posting) not only ensure the posting experience for normal users but also effectively restrict high-risk users. The credit-driven upgrade and downgrade mechanism incentivizes users to maintain good transaction behavior, forming a positive cycle.

[0024] (5) Full-link, configurable, and scalable algorithm architecture. Geofencing constraints are integrated throughout the entire process of publishing, matching, and risk control, forming a closed-loop community territorial management system. Key parameters such as fence radius, weight coefficient, detection threshold, and blocking duration can be flexibly configured according to scenario requirements such as community density and city size, without requiring code modification. The algorithm has good adaptability and scalability, and can be applied to more scenarios based on geographical location control, such as community group buying and local O2O services. Attached Figure Description

[0025] Figure 1 This is the overall flowchart of the algorithm.

[0026] Figure 2 This is a flowchart of the product release process and the cross-regional interception algorithm.

[0027] Figure 3 This document presents a diagram illustrating the permission levels and upgrade / downgrade paths within a hierarchical permission system.

[0028] Figure 4 This is a flowchart of an anomaly detection algorithm for professional merchants.

[0029] Figure 5 Flowchart of the localized intelligent matching and sorting algorithm. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one. Example

[0032] Step S1: Geofencing Generation. The geofence is the fundamental constraint unit of this invention. The system generates an electronic geofence with the user-authenticated cell coordinates as the center and a preset radius R (default 3000 meters). The geofence supports two geometric representations: circular area representation: a circular area with the authenticated cell coordinates (P_centerLat, P_centerLng) as the center and R as the radius. The condition for determining if any point (P_lat, P_lng) is within the geofence is HaversineDistance(P_centerLat, P_centerLng, P_lat, P_lng) ≤ R. Polygonal region representation: Centered on the coordinates of the certified cell, an approximately circular regular polygonal region is generated with a preset number of vertices N (default N=32). The coordinates of each vertex i are: vertex_i_lat = P_centerLat + (R / R_earth) × cos(2πi / N) × (180 / π), vertex_i_lng = P_centerLng +(R / R_earth) × sin(2πi / N) × (180 / π) / cos(P_centerLat×π / 180), where R_earth=6371000 meters is the average radius of the Earth. Polygonal representation is suitable for scenarios requiring precise boundary control. The determination of whether a point falls within the polygon uses the ray casting algorithm: A ray is emitted from the point P to be determined to the horizontal right. The number of intersections between this ray and each side of the polygon is counted. If the number of intersections is odd, P is inside the polygon; if it is even, P is outside. For a polygon with N vertices, traverse all edges (i, i+1) and determine whether the ray intersects the edge. The time complexity is O(N). The specific determination steps are as follows: (1) Store the polygon vertices in clockwise or counterclockwise order as a vertex array V[0..N-1]; (2) For each edge (V[i], V[i+1]), check whether the y coordinate of P is within the y coordinate interval of the edge; (3) If it is within the interval, calculate the x coordinate of the intersection point of the ray and the edge: x_intersect = V[i].x + (Py - V[i].y) ×(V[i+1].x - V[i].x) / (V[i+1].y - V[i].y); (4) If x_intersect > Px, then increment the intersection point count by 1; (5) After traversal, if the intersection point count is odd, determine that P is inside the polygon and issue a request to pass the fence verification. When a user has multiple authenticated cells, the system generates an independent fence for each cell, storing it as a fence set G={G1, G2,…, Gn}. When a user posts a product, the geographical constraint is satisfied as long as the product's location falls within any of these fences.The fence radius R is a configurable parameter that supports differentiated settings based on scenario type: Residential community scenarios have a default R=3000 meters, office building / commercial area scenarios have a default R=5000 meters, and campus scenarios have a default R=5000 meters. The configuration parameter is stored in the data storage module and supports dynamic adjustment from the operations backend, adapting to different community densities and urban planning characteristics without code modification.

[0033] Step S2: Location Verification and Cross-Regional Interception. Receive the user's product posting request and extract the product posting location coordinates (P_pubLat, P_pubLng) and the user's authentication cell coordinates (P_centerLat, P_centerLng). Calculate the spherical distance d between the posting location and the authentication cell coordinates: using the Haversine formula: a = sin²(Δlat / 2) + cos(lat1)×cos(lat2)×sin²(Δlng / 2), c = 2×atan²(√a, √(1-a)), d = R_earth × c. Where Δlat = P_pubLat - P_centerLat, Δlng = P_pubLng - P_centerLng, and R_earth is the Earth's radius (taken as 6,371,000 meters). Determining whether the spherical distance d is less than or equal to the preset radius R specifically includes: S2.1 Fence-in-the-area verification—when d ≤ R, it is determined that the publishing location is within the fence range, the publishing request passes the cross-region verification, and proceeds to step S3; S2.2 Cross-region interception—when d > R, it is determined that the publishing location exceeds the fence range, the publishing request is automatically intercepted and an interception prompt message is returned, the interception prompt message includes the interception reason identifier, the distance exceeded Δd = d - R, and the publishing range prompt; S2.3 Historical cross-region behavior record—the interception event is written to the cross-region behavior log, recording the user identifier, request timestamp, publishing location coordinates, certified cell coordinates, actual distance, excess distance, and interception result, providing a data basis for abnormal merchant detection in step S4.

[0034] Step S3: Hierarchical Posting Permission Verification. Query the user's current posting permission level, which is divided into four levels: L0, L1, L2, and L3. S3.1 Permission Level Determination – Read the user's posting permission level data. If the user is newly registered and has not yet completed local authentication, the default permission level is L0 (unauthenticated resident). If the user triggers the abnormal merchant detection in Step S4 and the suspicion score S≥80, the level is downgraded to L3 (restricted resident), and a behavior observation period and recovery conditions are set according to the severity of the violation. S3.2 Posting Quota Verification – Obtain the corresponding daily posting quota based on the current permission level: L0 level has a daily quota of 0 (i.e., posting is prohibited), L1 level has a daily quota of 10, L2 level has a daily quota of 30, and L3 level has a daily quota of 0 (i.e., posting is prohibited). Query the number of products a user has published that day. If the number of products published that day has reached the limit for the corresponding level, intercept the publishing request and return a "limit reached" message. If the user is at level L0 or L3, intercept the publishing request and return a "level mismatch" message. If the user is at level L1 or L2, further verify the number of products published that day: if the number of products published that day has reached the limit for the corresponding level, intercept the publishing request and return a "limit reached" message. If the number of products published that day has not reached the limit, the verification passes and proceeds to step S4.

[0035] Step S4: Abnormal Merchant Detection. Within a preset sliding time window T_window (default 24 hours), extract multidimensional feature data of user posting behavior and calculate the merchant suspicion score. S4.1 Content Similarity Analysis—Calculate the text similarity between the user's current post and the previous K (default K=5) posts, using Jaccard similarity J(A,B)=|A∩B| / |A∪B|, and simultaneously calculate the visual similarity based on deep features cosine_sim(F_new, F_old)=(F_new·F_old) / (‖F_new‖×‖F_old‖); if the Jaccard similarity exceeds the preset threshold τ1 (default τ1=0.55) or the visual similarity exceeds the preset threshold τ2 (default τ2=0.80), the content is judged to be highly repetitive. S4.2 Suspicion Score and Judgment—Calculate the suspicion score S based on the four dimensions of features, using a 0-100 score scale: S = w1×f1 + w2×f2 + w3×f3 + w4×f4, where f1 is the posting frequency anomaly (detecting short-term high-frequency and long-term high-frequency posting behavior: ≥4 posts within 1 hour get 100 points, ≥8 posts within 24 hours get 80 points), f2 is the product homogeneity (the proportion of the same / similar products within the sliding window: Jaccard similarity ≥0.55 and ≥3 similar products get 95 points, ≥5 similar products get 60 points), f3 is the address clustering degree (cross-cell posting behavior within the sliding window: posting similar products across ≥3 cells gets 90 points, across 2 cells gets 40 points), f4 is the location consistency (the consistency between the user's posted location GPS trajectory and the certified cell: across ≥3 cities or regions gets 85 points, deviating from the certified cell ≥5km and ≥30% of trajectory points gets 60 points), w1, w2, w3, and w4 are the weights of each feature, with default values ​​of w1=0.30, w2=0.25, w3=0.25, and w4=0.20. Suspicion score S is triggered in different ranges, resulting in tiered actions: S < 40 indicates a normal user, with no action taken; 40 ≤ S < 60 is a warning level, with a warning notification sent to the user for 24 hours; 60 ≤ S < 80 is a traffic restriction level, with the posting limit reduced to 1 item per day for 24 hours; S ≥ 80 is a ban level, indicating the user is a professional merchant, their certification is revoked, posting privileges are temporarily banned, all posted products are removed from the platform, and the user is added to the blacklist candidate database.

[0036] Step S5: Localized Intelligent Matching. Receive buyer product search requests, obtain the buyer's verified cell coordinates, and execute a cell-priority matching strategy. Prioritize products within the same cell (same verified cell) according to distance, with four priority levels from closest to furthest: First priority is same cell (same verified cell); second priority is adjacent cell (distance ≤ 1km); third priority is surrounding cell (1km < distance ≤ 2km); fourth priority is outer cell (2km < distance ≤ 3km). Products beyond 3km are not included in the ranking results. For each priority level of candidate products, a comprehensive ranking score is calculated: Score = α × (1 - d / R) + β × Tnorm + γ × Pnorm + δ × Cnorm, where d is the spherical distance between the buyer's certified area and the product's posting location, R is the geofence radius (default R = 3000m), and when d = 0 (same area), d / R = 0, the distance score reaches its maximum value of 1.0; Tnorm = (Tnow - Tpublish) / (Tnow - Toldest) is the posting time freshness factor, with a value range of [0,1], where Tnow is the current timestamp, Tpublish is the product posting timestamp, and Toldest is the earliest posting timestamp within the geofence; Pnorm = 1 - |Pgoods - Pavg| / max(Pavg, Pgoods) is the price reasonableness factor, where Pgoods is the product price, and Pavg is the average transaction price of similar products in the past 30 days; Cnorm = Cseller / Cmax is the seller credit score factor, Cseller is the seller credit rating, and Cmax=100 is the maximum credit score; α, β, γ, and δ are the weight coefficients for each dimension, with default values ​​of α=0.50, β=0.20, γ=0.15, and δ=0.15, with distance weighting being dominant. The results are sorted in descending order by the overall score, generating a list of sorted results and returning it. Example 2

[0037] Figure 2 The flowchart of the product release process and cross-region interception algorithm is as follows: Figure 2 As shown in the figure, this embodiment demonstrates how the system performs geofence verification and cross-regional interception when the user is in different geographical locations.

[0038] When a user initiates a product posting request, the system executes the verification process in the following six steps: Step 1: Location Authentication Status Determination. The system reads the authentication status field from the user's profile. If the user has completed at least one community location authentication, the system proceeds to Step 2, Geofencing Generation. If the user has not yet completed location authentication (i.e., is at L0 guest level), the system immediately rejects the posting request, returns the interception message "You have not completed community location authentication and cannot post products. Please complete the authentication first," and pushes a location authentication guide card to the client, terminating the posting process. Location authentication is a prerequisite for posting products; users who have not passed authentication cannot trigger any subsequent verification steps.

[0039] Step 2: Geofencing Generation and Coordinate Acquisition for Product Posting. The system extracts the set of center coordinates of all authenticated cells in the user's profile, {(P_cLat_j, P_cLng_j) | j=1..n}, where n is the number of authenticated cells. For each authenticated cell, an electronic geofence G_j is generated with its coordinates as the center and a preset radius R (default R=3000 meters), forming a geofence set G={G_1, G_2, …, G_n}. Simultaneously, the system prioritizes obtaining the user's current device GPS location (P_pubLat, P_pubLng) as the product posting coordinates; if GPS positioning is unavailable or the user has not authorized location permissions, the coordinates of the user's preferred authenticated cell are used as the default posting coordinates.

[0040] Step 3: Geofencing Determination. The system calculates the spherical distance d between the published location and the center of the certified cell fence, using the Haversine formula: a = sin²(Δlat / 2) + cos(lat1)×cos(lat2)×sin²(Δlng / 2) c = 2 × atan2(√a, √(1-a)) d = R_earth × c = 6371000 × c Where lat1 and lng1 are the coordinates of the certified cell (converted to radians), lat2 and lng2 are the coordinates of the publishing location (converted to radians), Δlat = lat2 - lat1, Δlng = lng2 - lng1, and R_earth = 6371000 meters is the average radius of the Earth. This formula maintains high accuracy even when calculating long distances, with an error of less than 0.3%.

[0041] The system iterates through all fences G_j in the fence set G. If any fence exists that allows the posting location to fall within its range (i.e., HaversineDistance(P_cLat_j, P_cLng_j, P_pubLat, P_pubLng) ≤ R), it is determined that the posting location is within the fence, and the fourth step of posting permission status verification is initiated. If none of the fences contain the posting location, it is determined to be a cross-region posting, and the posting request is automatically intercepted, returning the interception message "You are currently not within the certified cell range and cannot post products." The message includes the distance exceeded Δd = min(d_j) - R and the direction of the nearest fence, and the posting process terminates. Cross-region interception events are synchronously written to the cross-region behavior log, recording the user identifier, request timestamp, posting location coordinates, certified cell coordinates, actual distance, excess distance, and interception result, providing a data foundation for subsequent S4 abnormal merchant detection. The time complexity of this traversal process is O(n), where n is the number of user certified cells, which in practice usually does not exceed 3.

[0042] Step 4: Publishing Permission Status Verification. The system reads the user's current publishing permission level L (value range {L0, L1, L2, L3}). If the user's level is L0 (unverified resident) or L3 (restricted user level, temporarily banned due to triggering risk control rules), the publishing request is immediately intercepted, and the interception message "Your current publishing permission is restricted, and you cannot publish products" is returned, along with the reason for the restriction and the estimated recovery time. The publishing process is terminated. If the user's level is L1 (verified resident level) or L2 (credit resident level), the process proceeds to Step 5: Publishing Frequency Check.

[0043] Step 5: Posting Frequency Check. The system obtains the corresponding daily posting limit based on the user's current permission level: L1 level daily limit N1 = 10 items, L2 level daily limit N2 = 30 items. The system queries the number of products n_pub that the user has successfully posted on the current day (from 00:00:00 on the calendar day to the current time). If n_pub has reached the limit for the corresponding level, the posting request is intercepted and a message "You have reached the limit for posting today, please try again tomorrow" is returned; if n_pub has not reached the limit, the verification passes and proceeds to Step 6: Batch Same-Item Risk Control Judgment.

[0044] Step 6: Batch Same-Item Risk Control Judgment. To identify suspected professional merchant behavior, the system performs batch same-item judgment within a sliding time window T_window (default 24 hours): extracts all products posted by users within the window, calculates the text Jaccard similarity (threshold τ1=0.55) and visual cosine similarity (threshold τ2=0.80) for any two products. If the text and visual similarity are similar, they are judged as the same product. When the number of the same product posted across different communities within the window is ≥2, the batch same-item risk control judgment is triggered as "yes", entering the suspected professional merchant branch; otherwise, it is judged as "no", entering the normal release branch.

[0045] Suspect Branch Handling: When a user is identified as a suspected professional merchant, the system will immediately perform the following actions: (1) Mark the user as a suspected professional merchant and add them to the risk control candidate database; (2) Trigger a temporary ban, reduce the user's permission level to L3 (restricted user level), and freeze the posting permission; (3) Remove all listed products from the platform; (4) Push a violation notification through the client: "The system has detected that you have posted the same items in multiple communities in a short period of time. Your posting permission has been temporarily banned. Please wait patiently for the manual review result"; (5) Record the risk control log, including user identification, trigger time, posting records in the window, similarity judgment details and handling results, to facilitate subsequent auditing and manual appeal review.

[0046] Normal branch handling: When it is determined that the behavior is not a batch of the same product, the system executes the product listing and release process: (1) Write the product into the main product table and set the product status to "on sale"; (2) Record the release log, including product identifier, user identifier, release timestamp, release location coordinates, the certified community identifier and fence distance; (3) Update the product index, including geospatial index and full-text search index; (4) Return a release success response through the client, with a link to the product details page and a sharing entry, and the release process ends. Example 3

[0047] Figure 3 To illustrate the hierarchical release of the permission system, including permission levels and upgrade / downgrade paths, as follows: Figure 3 As shown, the system queries the user's current publishing permission level. The level system is defined as follows: Level 0 (Guest): Users who have not completed local verification. Their permission status is "Cannot Post," and they can only browse products across the city. There is no priority for sorting by neighborhood; the default is to sort by posting time in descending order. When Level 0 users search for products, all city-wide results are displayed, with a message at the top of the page stating, "After verifying your neighborhood, you can prioritize matching neighboring products."

[0048] Level 1 Resident Certification: Users who have completed at least one community-based certification. Their permission status is "Limited Posting," allowing them to post products within the certified community's boundaries, with a daily posting limit of N1=10 items. They enjoy priority matching within the same community, and products from the same community are labeled "Neighbor" in search results. L1 is the default initial level after a user completes certification.

[0049] Level 2 Credit Resident: Users whose credit score reaches a preset threshold (default 80 points, maximum 100 points) based on Level 1 and who have no violations in the past 30 days. Their permission status is "Extended Posting," and the daily posting limit is increased to N2=30 posts. They can initiate neighborhood guarantee credit granting; upon successful granting, both the guarantor and guarantor receive credit points (+2 points per instance, maximum +6 points per day). Level 2 users enjoy the right to have their credit tag displayed.

[0050] The specific implementation process for neighborhood guarantee credit is as follows: ① Credit Initiation Conditions—L2 credit resident users can initiate guarantee credit requests to certified L2 users in the same community on the product details page, with a daily limit of 3 requests. The validity period of a single guarantee is 30 days by default; ② Credit Confirmation Mechanism—After receiving the credit request, the guaranteed party must confirm acceptance through facial recognition verification within 48 hours. It will automatically expire after this time. The system will push a confirmation notification to both parties, and the guarantee relationship will take effect upon confirmation; ③ Credit Score Rules—After successful credit granting, both the guarantor and guarantor will each receive +2 credit points. The daily maximum score for credit granting through guarantee credit is +6 points (i.e., a maximum of 3 valid guarantees), with a maximum credit score of 100 points. If the guaranteed party triggers risk control rules (suspicion assessment) within the guarantee validity period... ④ Guarantee termination mechanism – The guarantee will be automatically terminated upon expiration of the guarantee period. The guarantor can actively cancel the guarantee relationship at any time through the "My Guarantees" page. Credit points already obtained by both parties will not be recovered after cancellation. If the guaranteed party triggers a ban penalty (S≥80), the guarantee relationship will be forcibly terminated and the guarantor's points will be deducted immediately. The maximum number of guarantee relationships that can be valid at the same time for a single user is 5. ⑤ Guarantee data management – ​​Guarantee records are persistently stored in the data storage module, including the user identifiers of both parties, the initiation timestamp, the confirmation timestamp, the guarantee expiration time, the termination time, and the type of termination reason. These serve as reference feature data for long-term training of the credit scoring model.

[0051] Level 3 (Restricted User): Users whose posting privileges have been temporarily suspended due to triggering risk control rules. Their privilege status is "Posting Prohibited," and all listed products will be removed. During the suspension period, users cannot post products, but can browse and purchase normally. Users will automatically revert to Level 1 after the suspension period ends.

[0052] Privilege escalation / demotion rules: Upgrade path: L0→L1: Automatic upgrade to local certification completed; L1→L2: Automatic upgrade for credit scores ≥80 and no violations in the past 30 days, with batch upgrade tasks executed every day at midnight.

[0053] Downgrade path: L1 / L2→L3: Triggering risk control rules results in immediate downgrade; L2→L1: Credit score drops below 60 points, and is assessed every 30 days.

[0054] Recovery path: L3→L1: Reinstatement occurs after the ban period expires or after a manual appeal is approved; Users who trigger Level 3 three times consecutively will enter a 90-day observation period after recovery.

[0055] The specific rules for the observation period are as follows: (1) During the observation period, the suspect score detection threshold was reduced by 20%, that is, the original blocking threshold S≥80 was reduced to S≥64 to trigger blocking, the original warning threshold 40 was reduced to 32, and the original traffic limiting threshold 60 was reduced to 48, so as to achieve stricter monitoring of high-risk users; (2) During the observation period, the daily release limit will be reduced to the L1 level standard (10 items per day), and even if it was originally L2 level, it will not enjoy the extended limit; (3) Conditions for lifting the observation period: If no risk control rules are triggered during the observation period (suspicion score S<32), the observation period will be automatically lifted and the standard detection threshold and publishing permissions will be restored. If risk control rules are triggered again during the observation period, the observation period will be recalculated and extended by 30 days each time. Users who have been reset three times will have their publishing permissions permanently restricted (up to L1 level, not up to L2). Example 4

[0056] Figure 4 A flowchart of the algorithm for detecting anomalies in professional merchants, such as Figure 4 As shown, the system monitors user posting behavior in real time through a sliding time window T_window (default 24 hours) to identify professional merchants in a multi-dimensional feature space. The specific process is as follows: Step 1: Real-time Data Stream Publishing and Event Listening. The system continuously receives real-time data streams from the platform, and the event listener captures each publishing event. Each publishing event includes at least the user ID, community ID, product category, product image characteristics, publishing timestamp, publishing location, and product title, description, and image information. Upon receiving a publishing event, the listener writes the event to the risk control detection queue and establishes a traceable detection context using the user identifier and product identifier.

[0057] Step 2: Sliding Time Window Detection. The system uses a sliding time window T_window with a default value of 24 hours. The right end of the window represents the current detection time, and the left end represents the current detection time minus 24 hours. Upon receiving a new published event, the system adds it to the corresponding user's window data set and deletes expired events from the left end of the window to ensure that the statistical results only reflect real-time behavior within the most recent 24 hours.

[0058] Step 3: In-window metric statistics and trigger judgment for any dimension. The system reads cross-regional behavior log data within the sliding time window T_window, extracting the user's cross-regional interception records within the past 24 hours as the input data source for address clustering feature f3. The cross-regional behavior log is written in real time by step S2.3, including user identifier, request timestamp, posting location coordinates, certified cell coordinates, actual distance, and interception result. Based on this log, the system statistically analyzes at least three detection metrics: cross-cell posting frequency or posting breadth N_cross (extracting the number of posting attempts across different cells from the cross-regional log), the proportion of the same product R_identical, and the total posting frequency deviation F_ratio. Among them, N_cross reflects the user's posting range in different certified cells or different cell boundaries; R_identical reflects the proportion of the number of the same product in the window to the total number of posts; and F_ratio reflects the degree of deviation of the user's total posting frequency in the window from the platform's average posting frequency. The system compares the above metrics with corresponding thresholds.

[0059] When N_cross reaches the cross-cell posting threshold, R_identical reaches the same product percentage threshold, or F_ratio reaches the total posting frequency threshold, it is determined that any one of the detection dimensions has been triggered; if none of the three dimensions have reached the corresponding threshold, it is determined to be a normal user, and the normal user monitoring process continues, the window data is retained and waiting for subsequent posting events, and no ban is executed.

[0060] The third step, expansion—cross-region behavior log feature extraction: Before the multi-dimensional feature extraction described above, the system first reads the cross-region behavior log database (written by the publishing interception module each time a cross-region publishing attempt is detected) and extracts the following cross-region behavior features to assist merchants in detection: ① Cross-region frequency feature—statistically counts the number of cross-region publishing attempts C_cross within the sliding window T_window, serving as supplementary input for address clustering features. A higher C_cross indicates more frequent cross-region publishing behavior and a greater degree of suspicion; ② Cross-region address distribution feature—extracts the set of all target address coordinates attempted by the user from the cross-region logs and calculates the dispersion of the address distribution (measured by the standard deviation σ_cross of the coordinate set). A larger σ_cross indicates a wider cross-region range and a greater degree of suspicion; ③ Location drift feature—compares the device's GPS location coordinates in the cross-region logs with the user's authentication and registration address coordinates, calculates the average location deviation Δpos, and considers a location anomaly when Δpos exceeds a preset threshold (default 500m). The above cross-region log features and address clustering features are input into the address clustering degree score f3 calculation module, realizing a complete closed loop between cross-region behavior data and merchant risk control detection.

[0061] Step 4: Merchant Suspicion Score Calculation. When at least one detection dimension is triggered, the system calculates the merchant suspicion score S. The merchant suspicion score is calculated according to the weighted formula: S = w1×f1 + w2×f2 + w3×f3 + w4×f4 The metrics are as follows: f1 is the frequency of posting anomalies (detecting short-term and long-term high-frequency posting behavior: ≥4 posts within 1 hour earns 100 points, ≥8 posts within 24 hours earns 80 points); f2 is the degree of product homogeneity (the proportion of identical / similar products within the sliding window: similarity ≥0.55 and ≥3 similar products earns 95 points, ≥5 similar products earns 60 points); f3 is the degree of address clustering (cross-cell posting behavior within the sliding window: posting similar products across ≥3 cells earns 90 points, across 2 cells earns 40 points); f4 is the consistency of location (the consistency between the user's posted location's GPS trajectory and the certified cell: across ≥3 cities or regions earns 85 points, deviation from the certified cell ≥5km and ≥30% of trajectory points earns 60 points). The default weighting coefficients are w1=0.30, w2=0.25, w3=0.25, w4=0.20. The posting frequency weight w1 is the highest (0.30) because high-frequency posting is the most significant behavioral characteristic of professional merchants; the product homogenization weight w2 (0.25) and the address clustering weight w3 (0.25) are tied for second place, reflecting product characteristics and geographical behavior patterns, respectively; the location consistency weight w4 (0.20) is used as a supplementary dimension to identify GPS spoofing and virtual location bypassing behaviors.

[0062] Step 5: Threshold determination. The suspicion score S adopts a 0-100 point system. According to the score range, the graded handling is carried out. There are four risk levels: (1) S < 40 is the normal level, no action is taken; (2) 40 ≤ S < 60 is the warning level, a warning notification is pushed to the user, and the duration is 24 hours; (3) 60 ≤ S < 80 is the traffic restriction level, the posting limit is reduced to 1 item per day, lasting for 24 hours; (4) S ≥ 80 is the ban level, it is determined to be a professional merchant, the certification is revoked and the posting permission is temporarily banned.

[0063] Step 6: Banning Professional Merchants and Handling Published Products. Once a user is identified as a professional merchant, the system will temporarily ban them: freezing their posting privileges and removing all published products; sending a violation notification to the user; and recording a risk control log, including user ID, trigger time, posting history within the window, similarity assessment details, and the handling result, to facilitate subsequent auditing and manual appeal review.

[0064] Step 7: User Appeal Triage and Review. After a temporary ban, the system determines whether the user has submitted an appeal. If the user does not appeal, the ban enters a downgrade observation process upon expiration. The ban will be lifted or adjusted after the ban period expires, and subsequent posting behavior will be continuously monitored. If the user submits an appeal, the appeal request will be transferred to manual review to verify the user's identity, product source, posting history, and triggering indicators. If the manual review is successful, the ban will be lifted and posting privileges will be restored; if the manual review is rejected, the ban will be upgraded to a permanent ban.

[0065] Step 8: Blacklist Accumulation and Upgrade Mechanism. The system maintains a user violation accumulation counter, ViolationCount. Each time a user is identified as a professional merchant (suspicion score S≥80) and a temporary ban is executed, ViolationCount increments by 1. When ViolationCount accumulates to 2 violations, the system automatically upgrades the user from the blacklist candidate pool to the official blacklist pool, and performs the following associated risk control measures: (a) The initial credit score of all associated accounts registered by the user (identified through device fingerprint, mobile phone number hash value, and payment account binding relationship) is lowered to 40 points (the default initial score is 60 points). The score can only be restored to 60 points after completing full identity verification and a 30-day observation period; (b) The posting permission of officially blacklisted users is permanently restricted to L0 visitor level (browsing only), and cannot be upgraded through credit accumulation; (c) Products posted by associated accounts within 30 days are automatically entered into the high-risk control review queue, triggering a manual review process. ViolationCount is reset once a year (reset to zero on January 1st each year), but the official blacklist status is not automatically removed. Users must submit appeal materials and pass the platform's manual review before being removed. This mechanism ensures an effective deterrent against professional merchants and prevents them from circumventing risk control by changing their accounts.

[0066] Step 9: Accumulation of blacklist and risk control of associated accounts. The system maintains a two-level risk control list of blacklist candidate library and official blacklist. When a user is first identified as a professional merchant (S≥80), he / she is written into the blacklist candidate library and the number of violations count=1 is recorded. When the same user is identified as a professional merchant again (accumulated number of violations count≥2), he / she is officially included in the blacklist and the following associated account risk control measures are implemented: (1) Extract the device fingerprint (IMEI / IDFA / device model combination hash), mobile phone number and payment account of the blacklist user and establish associated feature vector; (2) Perform associated feature matching on newly registered accounts in the platform. If any feature is successfully matched, the initial credit score of the new account is reduced to the preset warning value C_warn=30 points; (3) The posting permission of the associated account is reduced to L1 level and the daily limit is halved to 5 items, and a 60-day observation period is entered; (4) If the associated account has no violations during the observation period, the credit score is gradually restored to the normal level (≥60 points) by +1 point per day. After the observation period, the standard posting permission is restored. Blacklist data is persistently stored in the data storage module, serving as the basic data source for risk control across the entire platform. Example 5

[0067] Figure 5 Here is a flowchart of the localized intelligent matching and sorting algorithm, such as Figure 5 As shown, when a buyer searches for or browses secondhand goods, the system executes the following intelligent matching process: Step 1: Obtain the coordinates of the buyer's verified neighborhood. If the buyer has completed local verification, obtain their verified neighborhood coordinates as the sorting reference point; if the buyer has multiple verified neighborhoods, sort independently using the coordinates of each neighborhood as the center, and take the optimal position of each sorting result as the final display position. If the buyer is not verified, the search results will be downgraded to the entire city sorted in descending order by posting time, and a prompt will be displayed at the top of the page indicating that verification will give priority to matching neighboring products.

[0068] Step 2: Four-level distance priority classification. Using the buyer's verified community coordinates as the center, all products within a 3000-meter radius (R=3000 meters) are divided into four priority levels based on distance (d): • First priority (same neighborhood): Goods within d = 0km are labeled "neighbor"; • Second priority (neighboring cells): Goods within a 0km < d ≤ 1km radius; • Third priority (surrounding communities): Goods within 1km < d ≤ 2km; • Fourth priority (outer perimeter): Goods within 2km < d ≤ 3km.

[0069] Products that are outside the fenced area (d > 3km) will not be included in the sorting results, or will be relegated to the bottom of the city-wide list.

[0070] Step 3: Calculate the overall ranking score. Calculate the ranking score for each item within the designated area using the following formula: Score = α × (1 - d / R) + β × Tnorm + γ × Pnorm + δ × Cnorm The calculation methods for each factor are as follows: DistWeight = 1 - d / R: When d=0 (same cell), it takes the maximum value of 1.0, which fully reflects the priority of the same cell; when d=R, it takes the minimum value of 0.

[0071] The publication time freshness factor Tnorm = (Tnow - Tpublish) / (Tnow - Toldest): the value range is [0,1]. Tnow is the current timestamp, Tpublish is the product publication timestamp, and Toldest is the earliest publication timestamp of the product displayed within the fence. The closer the value is to 1, the newer the product is.

[0072] Price rationality factor Pnorm = 1 - |Pgoods - Pavg| / max(Pavg, Pgoods): Pgoods is the price of the product, and Pavg is the historical average transaction price of similar products within the fence in the past 30 days. The value range is [0,1]. The closer the price is to the market average price, the higher the score.

[0073] Seller credit score factor Cnorm = Cseller / Cmax: Cseller is the seller's current credit score, Cmax=100 is the full credit score, and the value range is [0,1].

[0074] The default weight configuration is α=0.50, β=0.20, γ=0.15, δ=0.15, with distance having the largest weight ratio.

[0075] Step 4: Self-Pickup Bonus Mechanism. For products with "Self-Pickup Priority" in the product label or those where the seller supports self-pickup, an additional self-pickup bonus ε=0.05 is added to the overall ranking score, i.e., Score_final = Score + ε. This mechanism aims to promote face-to-face transactions within the neighborhood and reduce logistics costs and transaction risks.

[0076] Step 5: Output the sorting results. Sort the results in descending order by Score_final to generate the final sorted list, which is then displayed to the buyer. After sorting the products with coordinates, append the products without coordinates or with invalid coordinates to the end of the list. Example 6

[0077] Comparative experiment with existing secondhand trading platforms. To verify the technical effectiveness of this invention, mainstream secondhand trading platforms (Xianyu, Zhuanzhuan) were selected as the benchmark for comparison. Comparative tests were conducted on the same test dataset (10,000 product listing records, covering real geographical location data from 50 neighborhoods). Evaluation metrics included: ① Professional Merchant Interception Rate – This invention achieves a professional merchant interception rate of 92.3% through multi-dimensional feature weighted scoring and geofencing verification, compared to the platform's average interception rate of 61.7% (which relies on keyword filtering and user reports). ② False interception rate - The false interception rate of this invention is only 2.1% (mainly due to GPS positioning deviation), while the false interception rate of the comparison platform is 8.5% (due to the lack of accurate geofence verification). ③ Matching efficiency—This invention improves the intra-cell matching hit rate to 78.6% through a four-level distance hierarchical and localized priority matching strategy, compared to a platform that only sorts by distance with an intra-cell hit rate of 42.3%. ④ User Trust Level – After introducing neighborhood guarantee credit and credit rating mechanism, the user transaction trust level score of the platform of this invention (calculated based on transaction completion rate and dispute rate) reached 87.4 points, while the comparison platform was 71.2 points.

[0078] The comparative data above demonstrates that the present invention has significant technical advantages in professional merchant identification, localized matching efficiency, and community trust building.

[0079] 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, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent matching and cross-regional posting interception of idle items, characterized in that, comprising the following steps: Step S1, geofence generation: acquiring the geographic coordinates of the user's authenticated cell, and generating an electronic geofence with the coordinates as the center and a preset radius, wherein the fence is defined as a geographic range constraint area centered on the authenticated cell; When a user has multiple authenticated cells, a corresponding independent geofence is generated with each authenticated cell as the center, and the user can publish commodities within any of the fences; Step S2, publishing position verification and cross-area interception: receiving a user's commodity publishing request, extracting the position coordinates of commodity publishing, calculating the spherical distance between the position and the coordinates of the authenticated cell, and judging whether the spherical distance is less than or equal to the preset radius; if the spherical distance is greater than the preset radius, the publishing request is automatically intercepted and interception prompt information is returned, and if the spherical distance is less than or equal to the preset radius, the process proceeds to step S3; Step S3, graded publishing permission verification: querying the user's current publishing permission level, wherein the permission levels include a total of four levels: L0 visitor level, L1 authenticated resident level, L2 credit resident level, and L3 restricted user level; If the user is at the L0 or L3 level, the publishing request is intercepted; if the user is at the L1 or L2 level, it is further verified whether the number of publications on the current day exceeds the publishing quota of the corresponding level, and if the quota is not exceeded, the process proceeds to step S4; Step S4, abnormal merchant detection: within a preset sliding time window T_window (default 24 hours), multi-dimensional features of the user's publishing behavior are extracted, wherein the multi-dimensional features include four dimensions: publishing frequency abnormality degree, commodity homogenization degree, address clustering degree, and positioning consistency, and the scores of each dimension are normalized to the [0,100] interval and then weighted and summed according to a preset weight coefficient; a merchant suspicion score is calculated according to the multi-dimensional features and the preset weight coefficients, and when the merchant suspicion score exceeds a preset threshold, the user is determined as a professional merchant, the publishing permission is temporarily blocked, all published commodities are taken off the shelves uniformly, and the user is written into a blacklist candidate library at the same time; Step S5, localized intelligent matching: receiving a buyer's commodity search request, acquiring the coordinates of the buyer's authenticated cell, executing a same-cell priority matching strategy, and dividing into four levels of priority by distance from near to far: the first priority is the same cell (d=0km), labeled with a "neighbor" tag; The second priority is adjacent cells (0km<d≤1km); the third priority is peripheral cells (1km<d≤2km); the fourth priority is outer cells (2km<d≤3km). Commodities beyond the fence range (d>3km) will not be included in the sorting results or will be downgraded to the bottom of the city-wide list. By integrating the distance weight α, publishing time freshness β, price rationality γ and seller credit score δ, the comprehensive sorting score is calculated according to the formula Score=α×(1-d / R)+β×Tnorm+γ×Pnorm+δ×Cnorm, and a sorting result list is generated and returned.

2. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The generation of the electronic geofence in step S1 further includes: the preset radius is a configurable parameter with a default value of 3000 meters; the geofence supports two geometric representations: circular areas and polygonal areas. Circular areas are directly defined with the coordinates of the certified community as the center and the preset radius, while polygonal areas are generated as approximate circular polygons based on the coordinates of the certified community and the preset number of vertices; when a user completes local authentication, the system automatically records the latitude and longitude coordinates of the certified community and persistently stores the binding relationship between the user and the community in the data storage module as the reference coordinates for all subsequent geofence verifications; in the multi-certified community scenario, the user corresponds to an independent fence area in each certified community, and when the spherical distance from the product posting location to any certified community is less than or equal to the preset radius, posting is allowed within that fence.

3. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The publishing location verification and cross-regional interception in step S2 further includes: the system prioritizes obtaining the user's current device location coordinates as the product publishing location; if the user has not authorized location permissions or the location information is unavailable, the system automatically uses the user's certified cell coordinates as the default publishing location; the system calculates the spherical distance between the product publishing location and the certified cell coordinates, and determines whether the spherical distance is less than or equal to a preset radius; if it is greater than the preset radius, it is determined to be a cross-regional publishing, automatically intercepted, and a cross-regional interception prompt is returned; if it is less than or equal to the preset radius, the system allows the subsequent publishing process; for multi-certified cell scenarios, the system traverses the fences corresponding to each certified cell, and as long as the product publishing location falls within any fence range, it is determined to be within the fence, and publishing is allowed to continue.

4. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The specific configuration of the hierarchical publishing permission system mentioned in step S3 is as follows: L0 Guest level: Users who have not completed local authentication cannot publish products, can only browse products in the city, have no community priority in sorting, and do not enjoy the right to priority matching within the same community; L1 Certified Resident Level: Users who have completed at least one community-based certification can post products within the certified community's fence. The daily posting limit is the first preset quantity. They enjoy priority matching within the same community and can view the tags of neighbors within the same community. L2 Credit Resident Level: Based on the L1 level, users whose credit scores reach the preset credit threshold and have no violations in the past 30 days will be automatically upgraded to the L2 level; the daily posting limit will be increased to the second preset number, and they can initiate neighborhood guarantee credit granting. After the credit granting is successful, both the guarantor and the guarantor will receive credit points; L3 Restricted User Level: Users whose posting permissions are temporarily suspended due to triggering risk control rules will have their posting permissions removed. After the suspension period ends, they will be automatically restored to the L1 level; If the L3 level is triggered consecutively, the user will be placed on a watch list and the subsequent detection threshold will be lowered. The rules for upgrading and downgrading between the permission levels are as follows: the condition for upgrading from L0 to L1 is to complete the local authentication; the condition for upgrading from L1 to L2 is to reach the credit score threshold and have no violation records; the condition for downgrading from L1 or L2 to L3 is to trigger the risk control rules; the condition for restoring from L3 to L1 is that the ban period expires or the manual appeal is approved.

5. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The identification of identical or similar products in step S4 adopts a dual judgment mechanism of text semantic similarity and image perceptual hashing, specifically including: text similarity calculation: extracting product title and description text, generating semantic feature vectors through pre-trained language models or word frequency inverse document frequency vectorization, calculating the cosine similarity between the text feature vectors of the two products, and determining text similarity when the cosine similarity is greater than or equal to a first preset threshold; image similarity calculation: extracting product images, generating binary fingerprints through a perceptual hashing algorithm, calculating the Hamming distance between the fingerprints of the two images, and determining image similarity when the Hamming distance is less than or equal to a second preset threshold; comprehensive judgment rule: when the text of two products is similar and the images are similar, the two products are determined to be identical or similar products; the identification results of identical or similar products are also used for the calculation of merchant suspicion score in step S4 and sorting and deduplication in step S5.

6. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The merchant suspicion score mentioned in step S4 is obtained by weighting and summing the scores of four dimensions: abnormal posting frequency, homogeneity of products, address clustering degree, and location consistency. The default duration of the sliding time window T_window is 24 hours. The window moves forward 24 hours from the current time and is dynamically updated as real time progresses.

7. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The duration of the temporary ban mentioned in step S4 is dynamically determined based on the merchant's suspicion score. A segmented, tiered ban strategy is adopted, dividing the suspicion score S into four risk ranges: (1) S<40 is the normal level, and no action is taken; (2) 40≤S<60 is the warning level, and a warning notification is pushed to the user for 24 hours; (3) 60≤S<80 is the traffic restriction level, and the posting limit is reduced to 1 item per day for 24 hours; (4) S≥80 is the ban level, which determines that the merchant is a professional, and the certification is revoked and the posting permission is temporarily banned. All posted products are removed from the platform. The preset threshold, the first preset duration, the second preset duration, and the third preset duration are all configurable parameters.

8. The method for intelligent matching and cross-regional posting interception of idle items according to claim 1, characterized in that, The comprehensive ranking score in step S5 is obtained by weighting and summing the distance weight factor, the freshness factor of the posting time, the reasonableness factor of the price, and the seller credit score factor according to preset weight coefficients; the distance weight factor is calculated based on the spherical distance between the buyer's certified community and the product posting community and the fence radius; The publication time freshness factor is calculated based on the product publication timestamp, the current timestamp, and the earliest publication timestamp within the fence. The price reasonableness factor is calculated based on the product's pricing and the historical average transaction price of similar products; the seller credit score factor is calculated based on the seller's current credit score and the maximum credit score.

9. 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 method for intelligent matching and cross-regional posting interception of idle items as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions, which, when executed by a processor, enable the processor to implement the method for intelligent matching and cross-regional posting interception of idle items as described in any one of claims 1-8.

Images