ETC card unbinding anti-fraud method based on relation graph and multi-dimensional risk assessment

Abstract

The invention proposes an ETC card unbinding anti-fraud method based on a relation graph and multi-dimensional risk assessment, and the method comprises the steps: responding to an ETC equipment unbinding request submitted by a new vehicle owner, and querying an original vehicle owner ETC account toll deficit state based on a vehicle license plate index and the personal information of a new vehicle owner; and analyzing vehicle transfer associated data between the new vehicle owner and the old vehicle owner through an association relationship analysis engine, and inputting a fraud risk assessment model to generate a risk level assessment result: if the original vehicle owner is arrearage and high in risk, rejecting a request and triggering an alarm, if the original vehicle owner is medium in risk, postponing processing and prompting to supplement ownership proof, and if the original vehicle owner is low in risk, executing unbinding auditing. According to the method, the identity overlapping degree, the historical behavior similarity and the transaction path characteristics are associated in real time, dynamic evaluation is combined with the arrearage state, collusion fraud behaviors of false transfer debt evading are accurately identified and intercepted, an operation chain of malicious debt account logout is blocked, convenience is maintained, the debt revenue basis of an ETC management party is guaranteed, and the method is suitable for popularization and application. And the problem of toll loss caused by fraudulent unbinding is thoroughly solved.

Description

Technical Field

[0001] This invention relates to the field of ETC services, and in particular to a method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment. Background Technology

[0002] When a new car owner acquires a vehicle and its license plate through a transfer, they cannot bind a new ETC device to the license plate because the license plate is already linked to the original owner's ETC device. The original owner may lack the willingness to actively unbind the device due to various factors. Therefore, existing technology provides a quick process for new car owners to unbind the vehicle from the original owner's ETC device. This process is not only convenient but also avoids cancellation fees, allowing new car owners to unbind the original owner's ETC device from the vehicle license plate without going through the original owner. However, some original vehicle owners with outstanding toll fees deliberately transfer their vehicles and license plates to acquaintances. They then use this quick process to unbind the original owner's ETC device from the license plate, creating a new ETC device linked to the license plate. The new owner then hands over the vehicle, license plate, and the bound ETC device to the original owner. This deprives the ETC device management authorities of the means to effectively curb the original owners' refusal to repay the outstanding toll fees, while the original owners can continue to enjoy the convenience of the ETC device, resulting in huge losses for the ETC device management authorities. Summary of the Invention

[0003] This invention proposes a method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment, including: S1. Responding to the ETC device unbinding request submitted by the new vehicle owner user through the ETC vehicle license plate occupancy quick processing system, wherein the ETC device unbinding request is for the original vehicle owner's ETC device corresponding to the vehicle license plate transferred from the original vehicle owner user. S2. Based on the vehicle license plate index and the personal information of the new vehicle owner, query the toll arrears status data of the original vehicle owner's ETC device account in the ETC settlement system, and analyze the vehicle transfer relationship data between the new vehicle owner and the original vehicle owner through the relationship analysis engine. S3. Input the vehicle transfer relationship data into a preset fraud risk assessment model, and the fraud risk assessment model generates a fraud risk level assessment result based on the vehicle transfer relationship data; S4. When the original vehicle owner's ETC device account has outstanding fees: If the fraud risk assessment result exceeds the high-risk threshold, the unbinding request will be rejected and a fraud alarm will be triggered. If the fraud risk assessment result is in the medium risk range, the process will be temporarily suspended and the new car owner will be prompted to supplement the ownership certificate. If the fraud risk assessment result is lower than the low risk threshold, the unbinding review process will be executed.

[0004] Furthermore, the vehicle transfer relationship data specifically includes: direct and indirect transaction records between the new vehicle owner and the original vehicle owner, identity overlap index, historical behavior similarity index, and economic relationship overlap score.

[0005] Furthermore, step S2, which involves parsing the vehicle transfer relationship data between the new car owner and the original car owner, specifically includes: S21. Based on the vehicle license plate index and the personal information of the new vehicle owner, locate the original vehicle owner node and the new vehicle owner node in the ETC user knowledge graph database, and extract the direct and indirect transaction records between the new vehicle owner and the original vehicle owner from the ETC user knowledge graph database based on the original vehicle owner node and the new vehicle owner node. S22. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the shared attribute fields between the new vehicle owner user node and the original vehicle owner user node, and the interpersonal relationships between the original vehicle owner user node and the new vehicle owner user from the ETC user knowledge graph database, and generate the identity overlap index between the new vehicle owner user and the original vehicle owner user based on the shared attribute fields and interpersonal relationships. S23. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the historical ETC passage behavior datasets of the original vehicle owner user and the new vehicle owner user from the ETC user knowledge graph database, and compare and analyze the historical ETC passage behavior datasets of the original vehicle owner user and the historical ETC passage behavior datasets of the new vehicle owner user to obtain the historical behavior similarity index and economic relationship overlap score between the original vehicle owner user and the new vehicle owner user.

[0006] Further, step S21 specifically includes: based on the original vehicle owner user node and the new vehicle owner user node located from the ETC user knowledge graph database, traversing the direct and indirect relationship paths between the original vehicle owner user node and the new vehicle owner user node, and identifying valid transaction records containing direct and indirect transaction records between the new vehicle owner user and the original vehicle owner user through a graph database path query algorithm. The transaction records specifically include transaction frequency statistics, total transaction amount, transaction time distribution characteristics, and relationship type weight coefficient. The valid transaction records must meet the following requirements: within a preset time window and the total transaction amount exceeds a preset amount threshold.

[0007] Furthermore, the direct relationship paths are traversed using a breadth-first search algorithm and a shortest path algorithm; the indirect relationship paths are traversed using a depth-first search algorithm; transaction records within a preset time window and whose total transaction amount exceeds a preset amount threshold are considered valid transaction records; and the relationship type weight coefficients are dynamically generated based on a preset weight mapping table.

[0008] Further, step S22 includes: extracting original shared attribute data containing the original vehicle owner's ID card number, mobile phone number, residential address, and social relationship network of the original vehicle owner and the new vehicle owner from the ETC user knowledge graph database; performing standardization processing on the original shared attribute data to obtain a standardized shared attribute dataset; performing a Boolean matching algorithm on the ID card number field in the standardized shared attribute dataset to obtain a basic score for ID card similarity between the new vehicle owner and the original vehicle owner; performing a character similarity algorithm on the mobile phone number field and the standardized residential address field in the standardized shared attribute dataset to obtain a partial matching score for communication information between the new vehicle owner and the original vehicle owner; performing a common contact ratio calculation on the social relationship network field in the standardized shared attribute dataset to obtain a social relationship network similarity score between the new vehicle owner and the original vehicle owner; inputting the basic ID card similarity score, the partial matching score for communication information, and the social relationship network similarity score into a weighted allocation model to generate an identity overlap index between the new vehicle owner and the original vehicle owner.

[0009] Further, step S22 includes: performing an address normalization algorithm on the residential address field to remove special characters and convert it into a standard format address string; applying Levenstein distance to calculate the similarity of the mobile phone number field, and applying cosine similarity to calculate the similarity of the standardized residential address field; and assigning weights to the basic score of ID card similarity, the partial matching score of communication information, and the similarity score of social relationship network based on preset weight coefficients.

[0010] Further, step S23 includes: extracting original ETC behavior record data from the ETC user knowledge graph database; performing missing value imputation and outlier removal processing on the original ETC behavior record data to generate a clean ETC behavior dataset; outputting a traffic trend similarity score based on time series correlation analysis performed on the passage timestamp sequence in the clean ETC behavior dataset; outputting a route preference overlap score based on high-frequency path clustering analysis performed on the passage path coordinates in the clean ETC behavior dataset; and outputting arrears behavior deviation score based on the difference between the frequency and amount of arrears counted from the arrears event identifiers in the clean ETC behavior dataset. Based on the calculation of the discount usage rate of the same road segment triggered by the discount strategy in the clean ETC behavior dataset, the discount mode consistency score is output; the traffic trend similarity score, route preference overlap score, arrears behavior deviation score, discount mode consistency score and economic relationship overlap score are input into the multi-dimensional evaluation model to generate the historical behavior similarity index between the original vehicle owner and the new vehicle owner.

[0011] Further, step S23 includes: filling missing travel timestamps using nearest neighbor interpolation; identifying and removing abnormal travel path coordinates using a standard deviation threshold algorithm; and calculating historical behavior similarity index based on a linear weighted summation formula, wherein travel trend similarity score, route preference overlap score, overdue payment behavior deviation score, and discount mode consistency score are weighted according to preset rules.

[0012] This application also proposes an ETC vehicle license plate occupancy rapid processing system for implementing the aforementioned ETC card unbinding and fraud prevention method based on relationship graphs and multi-dimensional risk assessment, including: The request receiving module is used to respond to the ETC device unbinding request submitted by the new vehicle owner user; The query module is used to query the toll arrears status data of the original owner's ETC device account in the ETC settlement system based on the vehicle license plate index; The relationship analysis engine is used to analyze the vehicle transfer relationship data between new car owners and original car owners. A fraud risk assessment model is used to generate fraud risk level assessment results based on vehicle transfer relationship data. The processing module is used to execute a Level 3 response based on the fraud risk assessment result when the original vehicle owner's ETC device account has outstanding toll payment records. High risk: The unbinding request will be rejected and a fraud alert will be triggered; Medium risk: Processing will be temporarily suspended and supplementary proof of ownership will be requested; Low risk: Perform the unbinding review process.

[0013] The proposed ETC card unbinding fraud prevention method based on relationship graphs and multi-dimensional risk assessment in this invention accurately identifies and intercepts collusive fraudulent behavior that uses fake vehicle transfers to evade debts. This method effectively blocks the operational chain of the original vehicle owner maliciously canceling the overdue ETC account through transactions with acquaintances, while maintaining the convenience of a fast processing method. This ensures the ETC management authority's basis for recovering outstanding debts and completely solves the problem of huge toll losses caused by fraudulent unbinding in the background technology. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the ETC card unbinding and fraud prevention method based on relationship graphs and multi-dimensional risk assessment proposed in this invention. Detailed Implementation

[0015] refer to Figure 1 This invention proposes a method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment, including: S1. Responding to the ETC device unbinding request submitted by the new vehicle owner user through the ETC vehicle license plate occupancy quick processing system, wherein the ETC device unbinding request is for the original vehicle owner's ETC device corresponding to the vehicle license plate transferred from the original vehicle owner user.

[0016] Specifically, the method proposed in this application is executed by an ETC vehicle license plate occupancy rapid processing system deployed on the ETC service provider's server. The new vehicle owner refers to an individual or legal entity that acquires the vehicle license plate and the original ETC device through vehicle transfer. The original vehicle owner refers to the registered owner of the vehicle license plate and ETC device before the vehicle transfer. When the new vehicle owner needs to bind the original ETC device to their newly established account due to vehicle transfer, they must submit an unbinding request through the system's front-end interface. This request must include at least the target vehicle license plate number, the new vehicle owner's identity information, and the original ETC device identifier. Upon receiving this request, the system initiates the subsequent anti-fraud processing procedure. The method provided in this application can receive and identify ETC device unbinding requests arising from vehicle transfers, providing a clear processing target and initiation signal for the subsequent risk assessment process.

[0017] S2. Based on the vehicle license plate index and the personal information of the new vehicle owner, query the toll arrears status data of the original vehicle owner's ETC device account in the ETC settlement system, and analyze the vehicle transfer relationship data between the new vehicle owner and the original vehicle owner through the relationship analysis engine.

[0018] Specifically, the method proposed in this application executes two data query and analysis tasks in parallel. The vehicle license plate index refers to the vehicle license plate number in the unbinding request. The ETC settlement system is the back-end system responsible for recording, calculating, and settling toll fees. First, using the vehicle license plate number as an index, the ETC settlement system's database is queried for the original owner's ETC device account bound to that license plate. The current fee settlement status of the account specifically includes: whether there are any unpaid toll bills, the amount owed, and the duration of the arrears, etc., which constitute toll arrears status data. Second, the vehicle license plate number and the new owner's personal information are input into the relationship analysis engine. This engine is a software module with a built-in algorithm for extracting and analyzing relationships between users from the ETC user knowledge graph database. Based on this input information, the engine locates relevant user nodes and parses out whether and to what extent a vehicle transfer relationship exists between the new owner and the original owner, outputting structured vehicle transfer relationship data. The purpose of this step is to simultaneously obtain preliminary evidence of financial risk and relationship authenticity risk.

[0019] S3. Input the vehicle transfer relationship data into a preset fraud risk assessment model, and the fraud risk assessment model generates a fraud risk level assessment result based on the vehicle transfer relationship data.

[0020] Specifically, the method proposed in this application quantifies the risk assessment of related data. The pre-set fraud risk assessment model is a trained gradient boosting decision tree model deployed in the model inference service of the system server. The input of this model explicitly receives the structured vehicle transfer related data output by step S2 as a feature vector. This feature vector specifically includes: the number of valid transaction records, the total transaction amount in the past year, the identity overlap index score, the historical behavior similarity index score, and the economic relationship overlap score. In the processing layer, the gradient boosting decision tree model consists of 100 decision trees, each with a maximum depth of 6 and a learning rate of 0.1. During the training phase, the model uses historical unbinding application samples and their final labels indicating whether they were fraudulent. The internal processing logic of the model is as follows: the input feature vector passes through these 100 decision trees sequentially, each tree performs a series of binary judgments based on the feature value, and finally reaches a leaf node, which outputs a basic prediction value. The basic prediction values ​​of all trees are weighted and summed, and then mapped using the Sigmoid function to obtain an original risk probability value between 0 and 1. At the output end, the original risk probability value is converted into a discrete fraud risk level assessment result based on a preset threshold: when the original risk probability value is ≥ 0.7, high risk is output; when 0.3 ≤ original risk probability value < 0.7, medium risk is output; and when the original risk probability value < 0.3, low risk is output. The technical effect of this model lies in its ability to non-linearly fuse multi-dimensional correlation features into a unified risk quantification index through ensemble learning to support subsequent decision-making.

[0021] S4. When the original vehicle owner's ETC device account has outstanding fees: If the fraud risk assessment result exceeds the high-risk threshold, the unbinding request will be rejected and a fraud alarm will be triggered; if the fraud risk assessment result is in the medium-risk range, the process will be suspended and the new vehicle owner will be prompted to supplement the ownership certificate; if the fraud risk assessment result is below the low-risk threshold, the unbinding review process will be executed.

[0022] Specifically, the method proposed in this application employs a tiered approach based on dual conditions. The processing module is a decision-making and execution component within the system. The processing module first determines whether there are any outstanding toll payment records in the queried toll payment arrears data. If so, a three-tiered response is initiated based on the generated fraud risk level assessment. In the method proposed in this application, two risk thresholds are preset: a high-risk threshold and a low-risk threshold, with the area between them representing a medium-risk range. When the risk level is high, the unbinding request is automatically rejected to prevent the use of unbinding to evade arrears. Simultaneously, a fraud alarm is sent to the risk control platform, including request details, risk score, and related evidence, triggering manual auditing or further investigation. When the risk level is medium, the request is temporarily suspended, and a prompt message is sent to the new vehicle owner through the front-end interface, requiring them to upload supplementary vehicle ownership documents for further verification. The system sets a 48-hour deadline for supplementary materials. When the risk level is low, the request is allowed to proceed to the subsequent unbinding review process, which includes steps such as automatically verifying the consistency of account information. The technical effect of this step is to achieve refined risk control, striking a balance between preventing fraud and ensuring a normal user experience.

[0023] The data collection, tag management, rule setting, and recommendation decision-making processes proposed in this application are all based on publicly available and legal ETC business data and user authorization information. Before initiating the anti-fraud processing flow, the system first obtains and verifies the explicit authorization issued by the new vehicle owner user regarding the personal information required for this method. This authorization clearly covers the data scope required to complete the unbinding and risk assessment, including their transaction records and social relationships with the original vehicle owner user. Simultaneously, the system has obtained authorization from the original vehicle owner user through legal means such as the ETC service agreement to conduct necessary analysis and processing of their account-related information to prevent fraud and ensure system security. Data collection originates from user-submitted ETC business applications, legal vehicle transfer records, and toll collection and payment records generated during ETC use. All data collection is conducted within the scope of the aforementioned user agreement authorization. Tag management refers to modeling and classifying entities and relationships such as user nodes and transaction relationships in a knowledge graph. This process is based on objective business data and does not involve subjective evaluation or discriminatory classification of users. The rule settings refer to preset risk thresholds, transaction validity judgment conditions, and weight mapping tables. These rules aim to identify abnormal transaction patterns to prevent fraud. Their settings are based on historical data analysis and business risk control experience, and their purpose is to maintain the legitimate rights of ETC debt recovery and transaction order. They do not contain any content that violates laws, social morality, or harms public interests. The recommended decision refers to the tiered handling process automatically executed by the system based on risk assessment results. This decision-making logic is transparent and aims to block the fraud chain and protect the legitimate rights and interests of normal users. The decision-making process does not involve illegal or unethical purposes. Therefore, the technical solution provided in this application complies with the provisions of Article 5, Paragraph 1 of the Patent Law.

[0024] Furthermore, the vehicle transfer relationship data specifically includes direct and indirect transaction records between the new vehicle owner and the original vehicle owner, identity overlap index, historical behavior similarity index, and economic relationship overlap score.

[0025] Specifically, the method proposed in this application defines the specific composition of vehicle transfer-related relationship data. Direct transaction records refer to fund transfer records between the new and original vehicle owners, completed through bank transfers, third-party payments, etc., directly related to the current or historical vehicle transfer. Indirect transaction records refer to fund or asset transfer records generated through intermediaries, affiliated companies, or other indirect channels. The identity overlap index is a quantitative score used to measure the degree of similarity or overlap between the two parties in static identity attributes such as ID information, contact information, residential address, and social network. The historical behavior similarity index is a quantitative score used to measure the degree of similarity in dynamic behavior patterns of the two parties during the use of ETC devices. The economic relationship overlap score is a quantitative indicator used to measure the strength of the association between the two parties in economic dimensions such as financial activities and consumption preferences. These data collectively constitute a multi-dimensional, three-dimensional user relationship profile, providing rich feature inputs for fraud risk assessment models.

[0026] Further, step S2 includes: S21. Based on the vehicle license plate index and the personal information of the new vehicle owner, locate the original vehicle owner node and the new vehicle owner node in the ETC user knowledge graph database, and extract the direct and indirect transaction records between the new vehicle owner and the original vehicle owner from the ETC user knowledge graph database based on the original vehicle owner node and the new vehicle owner node. S22. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the shared attribute fields between the new vehicle owner user node and the original vehicle owner user node, and the interpersonal relationships between the original vehicle owner user node and the new vehicle owner user from the ETC user knowledge graph database, and generate the identity overlap index between the new vehicle owner user and the original vehicle owner user based on the shared attribute fields and interpersonal relationships. S23. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the historical ETC passage behavior datasets of the original vehicle owner user and the new vehicle owner user from the ETC user knowledge graph database, and compare and analyze the historical ETC passage behavior datasets of the original vehicle owner user and the historical ETC passage behavior datasets of the new vehicle owner user to obtain the historical behavior similarity index and economic relationship overlap score between the original vehicle owner user and the new vehicle owner user.

[0027] Specifically, the method proposed in this application details the three sub-step logics of the relationship analysis engine in parsing data. The ETC user knowledge graph database is a system built on a graph database, in which users, vehicles, transaction records, passage records, etc., are modeled as nodes, and the relationships between them are modeled as edges. S21: The engine first executes a query statement to locate the original vehicle owner user node and the new vehicle owner user node. After locating the nodes, the engine performs a path query to find all transaction paths within a certain number of hops and extracts the transaction record node attributes from them. S22: The engine queries the attribute sets of the two user nodes to find the common attribute fields, and at the same time queries whether there are interpersonal relationship edges connecting the two users in the graph. Then, it calls the identity overlap calculation submodule, which generates an identity overlap index score between 0 and 100 based on the matching degree of shared attributes and the existence and strength of interpersonal relationships using a weighted calculation formula. S23: Starting from the two user nodes respectively, the engine queries all historical ETC passage behavior record nodes connected to them, forming two datasets. The behavior comparison analysis submodule performs multi-dimensional comparisons between the two datasets, such as calculating the correlation of travel time distribution, the overlap of common routes, and the differences in overdue payment patterns. Through a series of predefined similarity calculation algorithms, it outputs historical behavior similarity indices and economic relationship overlap scores. These three sub-steps together complete the extraction and calculation of structured relational data from the original map data.

[0028] Further, step S21 specifically includes: based on the original vehicle owner user node and the new vehicle owner user node extracted from the ETC user knowledge graph database, applying a graph database path query algorithm to traverse the direct and indirect relationship paths between the new vehicle owner user and the original vehicle owner user, extracting transaction frequency statistics, total transaction amount, transaction time distribution characteristics and relationship type weight coefficients from the direct and indirect relationship paths, and filtering out valid transaction records based on preset validity rules.

[0029] Specifically, the method proposed in this application refines the rules and content for extracting transaction records. A direct relationship path refers to a path directly connecting the original owner user node and the new owner user node via only one edge representing a transaction. An indirect relationship path refers to a path connecting two user nodes via one or more intermediate nodes, and containing edges representing transactions. A graph database path query algorithm is used to discover and return these paths. For each path, the engine extracts transaction information: it counts the number of transactions within a certain period to obtain transaction frequency statistics; it sums all transaction amounts to obtain the total transaction amount; it analyzes the timing patterns of transactions to obtain transaction time distribution characteristics; and it assigns different relationship type weight coefficients based on the type of intermediate nodes in the path. To filter noisy data, a validity rule is set: only transaction records whose transaction time falls within a preset time window before the time of this unbinding request and whose cumulative total transaction amount exceeds a preset threshold are considered valid transaction records and included in subsequent analysis. This ensures that the analyzed transaction relationships are timely and economically significant.

[0030] Furthermore, the direct relationship paths are traversed using a breadth-first search algorithm and a shortest path algorithm; the indirect relationship paths are traversed using a depth-first search algorithm; transaction records within a preset time window and whose total transaction amount exceeds a preset amount threshold are considered valid transaction records; and the relationship type weight coefficients are dynamically generated based on a preset weight mapping table.

[0031] Specifically, the method proposed in this application illustrates the specific technical means of path traversal algorithm and weight generation. For discovering direct relationship paths, the relationship analysis engine employs a breadth-first search algorithm, starting from the original car owner user node and prioritizing the exploration of all edges that are first-degree related. If an edge directly points to the new car owner user node and its edge type is transaction-type, it is identified as a direct relationship path. The shortest path algorithm can be used to find the most representative path when multiple direct or simple indirect paths exist. For discovering complex indirect relationship paths, the engine employs a depth-first search algorithm, searching along a transaction chain in depth for up to three hops to discover indirect transaction relationships that may be formed through multiple intermediate links. The criteria for determining a valid transaction record are two logical AND conditions: a time window constraint and a monetary threshold constraint. The relationship type weight coefficient is not a fixed value but is dynamically determined by a preset weight mapping table. This mapping table defines the weight values ​​corresponding to different intermediate node types or relationship edge types. When the engine identifies the relationship type of a path, it queries this table to obtain the corresponding weight coefficient, thereby dynamically adjusting the importance of that transaction record in the overall evaluation.

[0032] Further, step S22 includes: extracting original shared attribute data containing the original vehicle owner's ID card number, mobile phone number, residential address, and social relationship network of the original vehicle owner and the new vehicle owner from the ETC user knowledge graph database; performing standardization processing on the original shared attribute data to obtain a standardized shared attribute dataset; performing a Boolean matching algorithm on the ID card number field in the standardized shared attribute dataset to obtain a basic score for ID card similarity between the new vehicle owner and the original vehicle owner; performing a character similarity algorithm on the mobile phone number field and the standardized residential address field in the standardized shared attribute dataset to obtain a partial matching score for communication information between the new vehicle owner and the original vehicle owner; performing a common contact ratio calculation on the social relationship network field in the standardized shared attribute dataset to obtain a social relationship network similarity score between the new vehicle owner and the original vehicle owner; inputting the basic ID card similarity score, the partial matching score for communication information, and the social relationship network similarity score into a weighted allocation model to generate an identity overlap index between the new vehicle owner and the original vehicle owner.

[0033] Specifically, the method proposed in this application describes the calculation process of the identity overlap index. First, data preparation is performed: the ID card number, mobile phone number, residential address text, and social network of two user nodes are extracted from the knowledge graph as raw shared attribute data. Standardization processing includes: unifying the ID card number into an 18-character string, removing spaces and area codes from the mobile phone number, using an address parsing library to convert the residential address into a standardized structure of province, city, district, street, and house number, and normalizing the contact names in the social network. Then, sub-item calculations are performed: the basic score for ID card similarity uses a Boolean matching algorithm; if two ID card numbers are completely identical, the score is 100; if the first 6 digits are identical, the score is 60; otherwise, it is 0. The matching score for the communication information portion is obtained by calculating the matching of the last few digits of the mobile phone number, comparing the standardized residential address level by level, and calculating the similarity using a character similarity algorithm, then weighting the results. The social network similarity score is obtained by calculating the ratio of the number of contacts in the intersection and union of the two users' social networks. Finally, the weighted allocation model receives these three scores, performs a weighted sum based on preset or learned weights, and outputs the final identity overlap index.

[0034] Furthermore, an address normalization algorithm is performed on the residential address field to remove special characters and convert it into a standard format address string; the Levenstein distance is applied to calculate the similarity of the mobile phone number field, and the cosine similarity is applied to calculate the similarity of the standardized residential address field; the weights of the ID card similarity base score, the communication information partial matching score, and the social relationship network similarity score are assigned based on preset weight coefficients.

[0035] Specifically, this application provides an embodiment of some computational details for the proposed method. The address normalization algorithm specifically involves: calling a geocoding service or a built-in address dictionary to convert non-standard addresses into standard-level strings and removing irrelevant special characters. When calculating the matching score for the communication information portion, for the mobile phone number field, the Levinstein distance is used to calculate the difference between two number strings; the smaller the distance, the higher the similarity, which is then converted into a corresponding score. For residential addresses that have been standardized into word vectors or string sequences, cosine similarity is used to calculate the cosine value of the angle between two address vectors in the semantic space, which is then used as the address similarity. The preset weight coefficients in the weight allocation model can be set, for example, as follows: a weight of 0.5 for the basic ID card similarity score, a weight of 0.3 for the communication information portion matching score, and a weight of 0.2 for the social relationship network similarity score. These coefficients are configured during system initialization and can be optimized based on the analysis results of historical fraud cases.

[0036] Further, step S23 includes: extracting original ETC behavior record data from the ETC user knowledge graph database; performing missing value imputation and outlier removal processing on the original ETC behavior record data to generate a clean ETC behavior dataset; outputting a traffic trend similarity score based on time series correlation analysis performed on the passage timestamp sequence in the clean ETC behavior dataset; outputting a route preference overlap score based on high-frequency path clustering analysis performed on the passage path coordinates in the clean ETC behavior dataset; outputting an arrears behavior deviation score based on statistical comparative analysis of the arrears status field in the clean ETC behavior dataset; outputting a discount pattern consistency score based on pattern consistency analysis of the discount usage tag field in the clean ETC behavior dataset; and inputting the traffic trend similarity score, route preference overlap score, arrears behavior deviation score, discount pattern consistency score, and economic relationship overlap score into a multi-dimensional evaluation model to generate the historical behavior similarity index.

[0037] Specifically, the method proposed in this application details the calculation method for historical behavior similarity indicators. First, data preprocessing is performed: all recent ETC passage records of the two users are extracted from the knowledge graph as raw data. Each record includes fields such as timestamp, path coordinates, cost, arrears status, and whether a discount was used. A data cleaning submodule is used to handle missing and outlier values, for example, filling missing timestamps with the average of previous and subsequent records, and removing abnormal path coordinates that are impossible to achieve at high speeds, forming a clean dataset. Then, multi-dimensional similarity calculations are performed: the passage trend similarity score is obtained by aggregating the passage timestamp sequences of the two users by week or month, forming two time series, calculating their Pearson correlation coefficient or dynamic time-normalized distance, and mapping it to a score of 0-100. The route preference overlap score is obtained by clustering the passage path coordinates of the two users to find their respective high-frequency passage paths, and then calculating the intersection ratio of these high-frequency path sets. The arrears behavior deviation score is obtained by statistically analyzing the number of arrears and the average arrears amount of both parties within the same time period, calculating the difference between these statistical values; the smaller the difference, the higher the score. The consistency score for discount patterns is calculated by determining the proportion of times each party triggered discounts to the total number of trips on the same road segments traversed by both parties. Finally, the multi-dimensional evaluation model uses these five scores as features to output a comprehensive historical behavior similarity index.

[0038] Furthermore, the missing travel timestamps are filled using the nearest neighbor interpolation method; the standard deviation threshold algorithm is applied to identify and remove the coordinates of abnormal travel paths; and the historical behavior similarity index is calculated based on the linear weighted summation formula, wherein the travel trend similarity score, route preference overlap score, overdue payment behavior deviation score, and discount mode consistency score are weighted according to preset rules.

[0039] Specifically, the method proposed in this application provides optional implementations for data cleaning and index synthesis. For missing travel timestamps, the nearest neighbor imputation method is used: in the clean dataset, find the K records most similar to the missing record in terms of path, cost, and other features, and fill the missing value with the mean or mode of the timestamps of these K records. For abnormal travel path coordinates, a standard deviation threshold algorithm is applied: calculate the standard deviation of travel time between the same entrance / exit pairs in all travel records, and consider records whose travel time exceeds the mean ± 3 times the standard deviation as abnormal, and mark their path coordinates as abnormal and remove them. When synthesizing the final historical behavior similarity index, a simple linear weighted summation formula is used as a specific implementation of the multi-dimensional evaluation model: the historical behavior similarity index is equal to the sum of the travel trend similarity score, route preference overlap score, overdue payment behavior deviation score, and discount mode consistency score multiplied by preset weight coefficients. These weights can be set according to business experience.

[0040] Furthermore, this application also proposes an ETC vehicle license plate occupancy rapid processing system for executing the aforementioned ETC card unbinding anti-fraud method based on relationship graphs and multi-dimensional risk assessment. The system includes: a request receiving module for responding to ETC device unbinding requests submitted by new vehicle owners; a query module for querying the original vehicle owner's ETC device account toll arrears status data in the ETC settlement system based on the vehicle license plate index; a relationship analysis engine for parsing vehicle transfer relationship data between the new and original vehicle owners; a fraud risk assessment model for generating fraud risk level assessment results based on the vehicle transfer relationship data; and a processing module for executing a three-level response based on the fraud risk level assessment results when the original vehicle owner's ETC device account has outstanding toll arrears records: High risk: reject the unbinding request and trigger a fraud alarm; Medium risk: suspend processing and prompt for supplementary ownership proof; Low risk: execute the unbinding review process.

[0041] Specifically, this application describes the system architecture for implementing the aforementioned method. The ETC vehicle license plate occupancy rapid processing system is a microservice architecture software system deployed on a cloud server. The request receiving module is an API service that provides a user interface and receives requests. The query module interacts with the database of the ETC settlement system through a database connection pool. The relationship analysis engine is an independent microservice that internally encapsulates a graph database client and indicator calculation logic. The fraud risk assessment model is based on the aforementioned gradient boosting decision tree model and is provided to the engine for invocation through a model inference service. The handling module makes decisions based on business rules and sends prompt messages to the notification service by calling the message queue, or directly updates the application status through database transactions. These modules work together through a service registration and discovery center to form a highly available, scalable, automated data processing and decision-making pipeline, realizing efficient and accurate ETC unbinding and fraud prevention functions.

[0042] The data collection, tag management, rule setting, and recommendation decision-making processes proposed in this application are all based on publicly available and legal ETC business data and user authorization information. Data collection originates from user-submitted ETC business applications, legal vehicle transfer records, and toll collection and payment records generated during ETC usage. All data collection is conducted within the scope authorized by the user agreement. Tag management refers to modeling and classifying entities and relationships such as user nodes and transaction relationships in a knowledge graph. This process is based on objective business data and does not involve subjective evaluations or discriminatory classifications of users. Rule setting refers to preset risk thresholds, transaction validity judgment conditions, and weight mapping tables. These rules aim to identify abnormal transaction patterns to prevent fraud. Their setting is based on historical data analysis and business risk control experience, with the purpose of maintaining the legitimate rights of ETC debt recovery and transaction order, and does not contain content that violates laws, social morality, or harms public interests. Recommendation decision-making refers to a tiered handling process automatically executed by the system based on risk assessment results. This decision-making logic is transparent, aiming to block fraud chains and protect the legitimate rights and interests of normal users. The decision-making process does not involve illegal or unethical purposes. Therefore, the technical solution provided in this application complies with the provisions of Article 5, Paragraph 1 of the Patent Law.

[0043] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment, characterized in that, include: S1. Responding to the ETC device unbinding request submitted by the new vehicle owner user through the ETC vehicle license plate occupancy quick processing system, wherein the ETC device unbinding request is for the original vehicle owner's ETC device corresponding to the vehicle license plate transferred from the original vehicle owner user. S2. Based on the vehicle license plate index and the personal information of the new vehicle owner, query the toll arrears status data of the original vehicle owner's ETC device account in the ETC settlement system, and analyze the vehicle transfer relationship data between the new vehicle owner and the original vehicle owner through the relationship analysis engine. S3. Input the vehicle transfer relationship data into a preset fraud risk assessment model, and the fraud risk assessment model generates a fraud risk level assessment result based on the vehicle transfer relationship data; S4. When the original vehicle owner's ETC device account has outstanding fees: If the fraud risk assessment result exceeds the high-risk threshold, the unbinding request will be rejected and a fraud alarm will be triggered. If the fraud risk assessment result is in the medium risk range, the process will be temporarily suspended and the new car owner will be prompted to supplement the ownership certificate. If the fraud risk assessment result is lower than the low risk threshold, the unbinding review process will be executed.

2. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment according to claim 1, characterized in that, The vehicle transfer relationship data specifically includes: direct and indirect transaction records between the new car owner and the original car owner, identity overlap index, historical behavior similarity index, and economic relationship overlap score.

3. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment according to claim 2, characterized in that, Step S2, which involves parsing the vehicle transfer relationship data between the new and original vehicle owners, specifically includes: S21. Based on the vehicle license plate index and the personal information of the new vehicle owner, locate the original vehicle owner node and the new vehicle owner node in the ETC user knowledge graph database, and extract the direct and indirect transaction records between the new vehicle owner and the original vehicle owner from the ETC user knowledge graph database based on the original vehicle owner node and the new vehicle owner node. S22. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the shared attribute fields between the new vehicle owner user node and the original vehicle owner user node, and the interpersonal relationships between the original vehicle owner user node and the new vehicle owner user from the ETC user knowledge graph database, and generate the identity overlap index between the new vehicle owner user and the original vehicle owner user based on the shared attribute fields and interpersonal relationships. S23. Based on the original vehicle owner user node and the new vehicle owner user node, obtain the historical ETC passage behavior datasets of the original vehicle owner user and the new vehicle owner user from the ETC user knowledge graph database, and compare and analyze the historical ETC passage behavior datasets of the original vehicle owner user and the historical ETC passage behavior datasets of the new vehicle owner user to obtain the historical behavior similarity index and economic relationship overlap score between the original vehicle owner user and the new vehicle owner user.

4. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment according to claim 3, characterized in that, Step S21 specifically includes: based on the original vehicle owner user node and the new vehicle owner user node located from the ETC user knowledge graph database, traversing the direct and indirect relationship paths between the original vehicle owner user node and the new vehicle owner user node, and identifying valid transaction records containing direct and indirect transaction records between the new vehicle owner user and the original vehicle owner user through a graph database path query algorithm. The transaction records specifically include transaction frequency statistics, total transaction amount, transaction time distribution characteristics, and relationship type weight coefficient. The valid transaction records must meet the following requirements: within a preset time window and the total transaction amount exceeds a preset amount threshold.

5. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment as described in claim 4, characterized in that, The direct relationship paths are traversed using breadth-first search and shortest path algorithms; the indirect relationship paths are traversed using depth-first search. Transaction records that are within a preset time window and whose total transaction amount exceeds a preset amount threshold are considered valid transaction records; the relationship type weight coefficient is dynamically generated based on a preset weight mapping table.

6. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment according to claim 3, characterized in that, Step S22 includes: extracting original shared attribute data containing the original vehicle owner's ID card number, mobile phone number, residential address, and social relationship network of the original vehicle owner and the new vehicle owner from the ETC user knowledge graph database; performing standardization processing on the original shared attribute data to obtain a standardized shared attribute dataset; performing a Boolean matching algorithm on the ID card number field in the standardized shared attribute dataset to obtain a basic score for ID card similarity between the new vehicle owner and the original vehicle owner; performing a character similarity algorithm on the mobile phone number field and the standardized residential address field in the standardized shared attribute dataset to obtain a partial matching score for communication information between the new vehicle owner and the original vehicle owner; calculating the proportion of common contacts on the social relationship network field in the standardized shared attribute dataset to obtain a social relationship network similarity score between the new vehicle owner and the original vehicle owner; inputting the basic score for ID card similarity, the partial matching score for communication information, and the social relationship network similarity score into a weighted allocation model to generate an identity overlap index between the new vehicle owner and the original vehicle owner.

7. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment as described in claim 3, characterized in that, Step S22 includes: performing an address normalization algorithm on the residential address field to remove special characters and convert it into a standard format address string; applying Levenstein distance to calculate the similarity of the mobile phone number field, and applying cosine similarity to calculate the similarity of the standardized residential address field; and assigning weights to the basic score of ID card similarity, the partial matching score of communication information, and the similarity score of social relationship network based on preset weight coefficients.

8. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment according to claim 3, characterized in that, Step S23 includes: extracting original ETC behavior record data from the ETC user knowledge graph database; performing missing value imputation and outlier removal on the original ETC behavior record data to generate a clean ETC behavior dataset; outputting a traffic trend similarity score based on time series correlation analysis performed on the traffic timestamp sequence in the clean ETC behavior dataset; outputting a route preference overlap score based on high-frequency path clustering analysis performed on the traffic path coordinates in the clean ETC behavior dataset; outputting arrears behavior deviation score based on the difference between the frequency and amount of arrears counted from the arrears event identifiers in the clean ETC behavior dataset; outputting a discount mode consistency score based on the discount usage rate calculation of the same road segment performed on the discount strategy trigger records in the clean ETC behavior dataset; and inputting the traffic trend similarity score, route preference overlap score, arrears behavior deviation score, discount mode consistency score, and economic relationship overlap score into a multi-dimensional evaluation model to generate a historical behavior similarity index between the original vehicle owner and the new vehicle owner.

9. The method for preventing fraud by unbinding ETC cards based on relationship graphs and multi-dimensional risk assessment as described in claim 3, characterized in that, Step S23 includes: filling missing travel timestamps using nearest neighbor interpolation; identifying and removing abnormal travel path coordinates using a standard deviation threshold algorithm; and calculating historical behavior similarity index based on a linear weighted summation formula, wherein travel trend similarity score, route preference overlap score, overdue payment behavior deviation score, and discount mode consistency score are weighted according to preset rules.

10. An ETC vehicle license plate occupancy rapid processing system, used to execute the ETC card unbinding and anti-fraud method based on relationship graphs and multi-dimensional risk assessment as described in any one of claims 1-9, characterized in that, include: The request receiving module is used to respond to the ETC device unbinding request submitted by the new vehicle owner user; The query module is used to query the toll arrears status data of the original owner's ETC device account in the ETC settlement system based on the vehicle license plate index; The relationship analysis engine is used to analyze the vehicle transfer relationship data between new car owners and original car owners. A fraud risk assessment model is used to generate fraud risk level assessment results based on vehicle transfer relationship data. The processing module is used to execute a Level 3 response based on the fraud risk assessment result when the original vehicle owner's ETC device account has outstanding toll payment records. High risk: The unbinding request will be rejected and a fraud alert will be triggered; Medium risk: Processing will be temporarily suspended and supplementary proof of ownership will be requested; Low risk: Perform the unbinding review process.

Images