Method and device for studying and judging illegally resold product oil
Through machine learning and graph neural network technology, funds and connected data in the crime of illegal reselling of refined oil are analyzed, abnormal transaction patterns and association relationships are identified, and the efficiency and accuracy of identifying and combating illegal reselling of refined oil are solved in the existing technology. , achieved precise strikes and prevention.
Patent Information
- Application Number
- CN202411419233.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology is difficult to quickly and accurately identify and crack down on the crime of illegal reselling of refined oil, the capital transaction data is complex, the connection data is messy, and the correlation between personnel and vehicle information is weak, resulting in low analysis and judgment efficiency and poor accuracy.
By obtaining fund transaction data, connection data and vehicle trajectory data, preprocessing and feature extraction, combining machine learning models and graph neural networks, identify abnormal transaction patterns and association relationships, build heterogeneous graphs and transaction maps, and analyze funds and connection behaviors , identify key personnel and gangs.
It has achieved accurate crackdowns and effective prevention and control of the crime of illegal reselling refined oil, improved the accuracy and efficiency of analysis and analysis, formed a comprehensive risk portrait, and guided the business departments to carry out targeted analysis.
Smart Images

Figure CN118941307A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of illegal business of refined oil, and in particular to a method and device for analyzing and judging illegal reselling of refined oil. Background Art
[0002] Illegal reselling of refined oil is a serious crime with wide-ranging implications and adverse consequences. In order to effectively combat the crime of illegal reselling of refined oil, scientific and effective research and judgment work is needed. Therefore, we propose a method for illegal reselling of refined oil based on data processing and machine learning algorithms, aiming to achieve rapid and accurate identification and combat of the crime of illegal reselling of refined oil.
[0003] In illegal reselling of refined oil and other illegal activities, criminals use bank cards, Alipay, WeChat, virtual currency and other channels to conduct fund transactions and collect sales proceeds. Telephone communication and social media are used to communicate upstream and downstream transactions. The illegal reselling of refined oil involves a large amount of fund flow, suspicious transaction accounts, communication phone numbers and social accounts. The fund transaction data is complex and the communication data is messy. After obtaining feedback on the transaction account data of key suspects through the fund control platform and the communication data feedback of key suspects through the operator, it is necessary to manually analyze and extract the fund characteristics and notes summaries of millions of transaction flow data in order to classify the accounts; the call time, duration and location between the communication objects must be analyzed and identified before the call numbers can be classified; the personnel and vehicle information is difficult to effectively integrate, the correlation between personnel and vehicle information is weak, and the correlation is difficult to be large, making it difficult to analyze the relationship between the gangs involved; the vehicle trajectory information data is huge, and the trajectory pattern is complex and diverse, making it difficult to identify with simple rules. At the same time, there may be errors in GPS positioning data, resulting in noise and missing trajectory information. The analysis and operation process of data such as funds, calls, personnel and vehicles, and vehicle trajectories consumes a lot of time and energy, seriously affecting the efficiency of the analysis and judgment on the illegal reselling of refined oil, as well as the accuracy and objectivity of the analysis and judgment on the illegal reselling of refined oil. Summary of the invention
[0004] The purpose of the present invention is to address the deficiencies of the prior art and propose a method and device for analyzing and judging illegal reselling of refined oil products, aiming to sort out the capital rules of illegal reselling of refined oil products and the operation mode of criminal gangs that illegally resell refined oil products. Through the analysis of capital behavior and transaction behavior habits, the core accounts, operating accounts and customer accounts are identified, and suspicious upstream accounts, intermediate accounts and lower-level accounts are mastered to determine the next direction of investigation and in-depth excavation of upstream and downstream crimes and criminal gangs that illegally resell refined oil products. Through communication behavior analysis, abnormal communication records are identified and communication gang relationships are expanded. Suspicious personnel and vehicles are associated through personnel and vehicle relationship information. At the same time, combined with vehicle trajectory data, suspicious vehicle activities are analyzed. Ultimately, precise strikes and effective prevention and control of illegal reselling of refined oil products are achieved.
[0005] The technical solution of the present invention is as follows: In the first aspect, a method for determining illegal reselling of refined oil products comprises the following steps: (1) Obtain the financial transaction data, communication data, location tracking data of refined oil vehicles, and identity data associated with refined oil vehicles of the persons involved in the case to construct a data set; (2) Preprocess the data set; extract the characteristics of fund data, analyze the stability of fund flow, identify abnormal transaction patterns and recipient groups; extract the characteristics of communication data, analyze abnormal call records, compare communication activity, identify potential connections based on the characteristics of fund data, and combine the identity data associated with refined oil vehicles to form a personnel-vehicle association data table, and integrate it with the geographic location tracking data of refined oil vehicles to form a refined oil vehicle driving trajectory data table; (3) Extract the personnel-vehicle usage relationship characteristics and vehicle trajectory characteristics from the personnel-vehicle association data table and the refined oil vehicle driving trajectory data table, and use the machine learning model to determine whether there is illegal reselling of refined oil by combining the capital data and communication data; (4) Based on the recognition results of step (3), a heterogeneous graph neural network is used to construct a heterogeneous graph. The nodes represent the persons involved in the case, and the edges represent the communication relationships and financial transaction relationships. A transaction graph is constructed based on the financial data, and the transaction behaviors and relationships are analyzed. The transaction center nodes and different roles in the transaction graph are identified, and key personnel are identified and divided into levels. Combined with the communication data and financial data, a graph clustering algorithm is used to divide the community, analyze the internal and external structures of the community, determine the core community where the key personnel are located, and further divide the key personnel in the core community into levels to determine the multi-level criminal network.
[0006] Furthermore, the communication data includes call time, call duration and call location data obtained from the communication operator platform, and the vehicle-related identity data includes the identity card, household registration or driver's license data of the person involved in the case obtained from the public security network data lake, which is used for identity verification, background investigation and personnel-vehicle association analysis.
[0007] Furthermore, the dataset preprocessing includes deduplication of the dataset, using an intelligent denoising algorithm to improve data accuracy, applying the XGBoost regression algorithm to accurately predict and fill in missing values, using time series analysis and cross-feature generation to extract deeper information, using the Yeo-Johnson power transformation algorithm to process non-normally distributed data to improve data consistency, and using the SMOTE technology to balance positive and negative samples.
[0008] Furthermore, key features such as transfer amount, transfer frequency, and recipient information are extracted from the fund data, and the change in the transfer amount is analyzed to identify anomalies. Key features such as call duration, call frequency, and number of callers and called parties are extracted from the communication data to discover abnormal call records. Time series analysis is performed on transfer and call data to identify potential criminal activities.
[0009] Furthermore, the identity association relationship features and usage time relationship features of personnel and vehicles are extracted from the personnel-vehicle association data table, and the vehicle trajectory features and vehicle usage frequency features are extracted from the vehicle driving trajectory data table, and the extracted features are screened, reduced in dimension, and normalized.
[0010] Furthermore, the machine learning model is specifically as follows: a classification model is constructed using a subspace random forest classification algorithm, and a sub-feature space containing different types of original data is constructed using the extracted funds, communications, vehicle associations and vehicle driving trajectory data. Multiple decision trees are constructed based on multiple sub-feature spaces, and each tree performs classification independently. The final classification result is obtained through a voting mechanism, and the importance of each data feature is evaluated. The feature selection is further optimized, and based on the optimized feature data, a deep neural network classification model is constructed using a multi-layer perceptron MLP and a K nearest neighbor algorithm KNN to determine and analyze the illegal reselling behavior patterns of refined oil.
[0011] Furthermore, the machine learning model uses adaptive enhancement and gradient boosting to perform ensemble learning. It iteratively trains multiple weak classifiers and adopts a dynamic weight adjustment mechanism to adjust the sample weights according to the error rate of each classifier. Finally, all weak classifiers are linearly combined into a strong classifier. A voting mechanism and a weighted fusion strategy based on model confidence are used to integrate the results of multiple weak classifiers to obtain the final prediction result.
[0012] Furthermore, based on the machine learning model, the fund transaction behavior and personnel communication behavior are analyzed. Through the scattered and collected fund characteristics, the abnormal account transaction data that meets the illegal reselling of refined oil is obtained; through the personnel communication relationship map, the key communication numbers and personnel subject information that meet the illegal reselling of refined oil are obtained, and the abnormal communication mode is discovered; by analyzing the personnel and vehicle association data, the personnel, vehicles and personnel-vehicle association abnormal data that meet the illegal reselling of refined oil are obtained, including personnel who frequently change vehicles and vehicles with abnormal frequency of use, multiple personnel using the same vehicle and one person using multiple vehicles, and the use time and location distribution of abnormal vehicles. By analyzing the vehicle trajectory characteristics, the DTW algorithm is used to calculate the similarity between the current trajectory of the vehicle and the historical trajectory to identify the abnormal trajectory behavior, and the abnormal vehicle and abnormal vehicle trajectory information are obtained, including vehicles that frequently enter and exit suspicious locations and stay for abnormally long periods of time, and the driving behavior of vehicles in different time periods; the fund transaction data, communication data and personnel-vehicle association data are combined to form a comprehensive portrait of personnel and vehicles, revealing potential criminal networks.
[0013] In a second aspect, the present invention also provides a device for determining illegal reselling of refined oil, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, the method for determining illegal reselling of refined oil is implemented.
[0014] In a third aspect, the present invention further provides a computer-readable storage medium having a program stored thereon, and when the program is executed by a processor, the method for determining illegal reselling of refined oil is implemented.
[0015] Beneficial effects of the present invention: The present invention establishes a decision tree that describes the relationship between each characteristic attribute value and its account type, and conducts more comprehensive and three-dimensional processing and analysis and research and application of account fund transactions, key personnel communication relationships, personnel vehicle association relationships, and vehicle driving trajectory information, deeply outlining the transaction behavior characteristics, communication relationship characteristics, personnel vehicle association relationship characteristics, and vehicle driving trajectory characteristics of the fund account, forming a comprehensive risk portrait of key personnel who illegally resell refined oil. Combined with the matching with the account portrait, through the communication relationship analysis, the role of the account in the transaction structure is determined; through the directed graph analysis, the upstream and downstream rules of the funds of the illegal resale of finished oil gangs are found, and the criminal facts of the members of the illegal resale of refined oil gangs are determined; through the clustering algorithm, the gang transaction behavior pattern and the abnormal behavior characteristics of fund transactions are summarized, the rationality of account identification account transaction behavior is determined, and the anomalies between account identity and transaction behavior are discovered in time; through the neural network, the abnormal association relationship pattern of personnel and vehicles is identified, and the illegal activity gang is determined; through the deep learning method, the abnormal change pattern of the vehicle driving trajectory is identified, and the potential illegal activity route is discovered. In summary, it effectively guides the business departments to further refine and carry out targeted analysis and judgment on key individuals and gangs who illegally resell refined oil products, and improves the effectiveness and pertinence of the analysis and judgment on capital transaction behavior, gang operation mode, communication behavior, personnel and vehicle relationship, and vehicle driving trajectory. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 This is the overall modeling idea diagram of the present invention.
[0017] Figure 2 The figure is a flow chart of the method of the present invention.
[0018] Figure 3 This is a diagram of the gang’s personnel hierarchy.
[0019] Figure 4 This is a diagram of the vehicle's driving trajectory.
[0020] Figure 5 This is the clustering diagram of base station location information.
[0021] Figure 6 This is a structural diagram of a device for analyzing and judging illegal reselling of refined oil provided by the present invention. DETAILED DESCRIPTION
[0022] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.
[0023] The present invention mines data features of accounts, fund data, and call data of various fund transaction activities in illegal reselling of refined oil products, analyzes abnormal fund behavior, transaction behavior inertia, and communication behavior characteristics, classifies the entire illegal reselling of refined oil account group, sorts out account fund behavior and communication behavior, and establishes analysis models to analyze and judge fund behavior and communication behavior, so as to serve the case crackdown.
[0024] Various types of financial data including bank cards, Alipay, WeChat, virtual currency, etc. are obtained from the fund control platform to identify fund flows and abnormal transactions; communication data including call time, call duration, call location, etc. are obtained from the communication operator platform for identification of the contact network and geographic location tracking of the persons involved in the case; public information of the persons involved in the case, such as identity card, household registration, driver's license, etc., is obtained from the public security network data lake for identity authentication, background investigation and personnel-vehicle association analysis; as well as data that may be related to the judgment of illegal resale of refined oil, such as vehicle GPS data, media reports, clue reports, etc., which may be related to the judgment of illegal resale of refined oil, to provide more comprehensive evidence support for the judgment of illegal resale of refined oil cases.
[0025] The model aims to sort out the financial patterns of illegal reselling of refined oil crimes and the operating modes of criminal gangs involved in illegal reselling of refined oil. It identifies core accounts, operating accounts and customer accounts through financial behavior analysis and transaction behavior habit analysis, and grasps suspicious upstream accounts, intermediate accounts and lower-level accounts, so as to determine the next direction of investigation and in-depth excavation of upstream and downstream crimes and criminal gangs involved in illegal reselling of refined oil.
[0026] The overall modeling idea of the illegal reselling of refined oil is as follows: Figure 1 As shown in the figure, the flow chart of the method for judging illegal reselling of refined oil is as follows Figure 2 shown.
[0027] The first step is to process the collected classified capital data, build a data set, integrate capital data, communication data, public information and other related data, and build a complete data set for subsequent analysis and modeling. The integrated data set is deduplicated, intelligent denoising is used to improve data accuracy, the XGBoost algorithm is used to accurately predict and fill missing values, advanced feature engineering (using time series analysis and cross-feature generation) is performed to extract deeper information, the Yeo-Johnson transformation is used to process non-normally distributed data to improve data consistency, and the SMOTE technology is used to balance positive and negative samples to ensure that the proportion of labeled data is about 50%, so as to facilitate subsequent machine learning algorithm modeling and improve the accuracy and robustness of algorithm prediction.
[0028] The specific process of cleaning and preprocessing fund data is as follows: 1) Traverse all the "XXX Bank Account Information Table" and "XXX WeChat Transaction Account Information Label" in the folders and compressed files at the same level as the data cleaning process of the fund control platform, process all "transaction card numbers" containing underscores, and delete abnormal suffixes.
[0029] 2) Traverse all the "Transaction Details" in the folders and compressed packages at the same level as the data cleaning model program of the fund control platform. If the "Transaction Account Name" and "Transaction Certificate Number" are empty, query the "Transaction Card Number" in the "Account Information Table" through the "Transaction Card Number" in the "Account Transaction Details Table", and fill in the "Account Opening Name" and "Account Opener Certificate Number" in the "Account Opening Basic Information Table" into the corresponding "Transaction Account Name" and "Transaction Certificate Number" in the "Account Transaction Details Table" to complete the information. If the transaction card number is empty, fill it in with the corresponding matching query of the transaction account number.
[0030] 3) If the "Opponent Account Number" is the same as the "Transaction Card Number" or the "Opponent Account Number is empty", replace the "Opponent Account Number" with the format of "Transaction Card Number" + "Opponent Account Name" + "Summary" + "Remarks" + "Cash Symbol".
[0031] 4) If the “Opponent Account Name” is empty, the “Opponent Account Name” will be the “Opponent Account Number”.
[0032] 5) After the above operations, remove the blank data of "Transaction Card Number", "Transaction Certificate Number", "Transaction Account Name", "Transaction Amount", "Transaction Date" and "Payment and Receipt Mark", and export the new classified transaction flow EXCEL table data.
[0033] The specific process of cleaning and preprocessing personnel-vehicle association data and vehicle trajectory data is as follows: 1) Traverse all the "XXX Personnel and Vehicle Information Tables" in the folders and compressed packages at the same level as the data cleaning process of the public security network data lake, and merge these tables into a unified personnel and vehicle association data set.
[0034] 2) Remove duplicate data with the same column names. For data missing from the "ID Number" column, if both the "Personnel Name" and "License Plate Number" of the data exist and the "Personnel Name" and "License Plate Number" have appeared in other valid data rows, they can be removed.
[0035] 3) Traverse all the "XXX Vehicle Trajectory Data Tables" in the folders and compressed packages at the same level as the data cleaning process of the public security network data lake, and merge these tables into a unified vehicle trajectory data set.
[0036] 4) De-duplicate the merged vehicle trajectory dataset to ensure that each trajectory record is unique.
[0037] 5) Format the time field uniformly as "YYYY-MM-DD HH:mm:ss".
[0038] 6) For data with missing "latitude" and "longitude", if the "checkpoint location" or "geographic location" of the data exists and the "checkpoint location" or "geographic location" has appeared in other valid data rows, the corresponding "latitude" and "longitude" will be added.
[0039] The second step is to extract and filter the features of the processed fund flow data and call record data. The specific process is as follows: 1) Traverse the cleaned and pre-processed fund transaction data and communication data, identify and extract the features of the data, and solidify the features. For vehicle association data and vehicle trajectory data, extract and output abnormal behavior patterns.
[0040] 2) Export the EXCEL data table of fund transaction feature flow, communication feature, personnel-vehicle association information feature, and vehicle trajectory information feature.
[0041] Extract key features such as transfer amount, transfer frequency, and payee information from fund data, analyze the change range of transfer amount, evaluate the stability of fund flow, and identify abnormal transaction patterns and common payee groups; extract key features such as call duration, call frequency, and number of callers from communication data, discover abnormal call records, compare communication activity, and identify potential connections; combine fund data with communication data to extract joint features such as "high-frequency transfers and high-frequency calls" to identify potential criminal behavior; perform time series analysis on transfer and call data to identify patterns related to specific events. Integrate the ID card, household registration, and driver's license data obtained from the public security network data lake to form a personnel-vehicle association data table, and integrate the driver's license data with the vehicle GPS data to form a vehicle driving trajectory data table. Extract key features such as personnel name, ID number, household registration location, age, gender, vehicle model, license plate number, the relationship between personnel and multiple vehicles, and the time relationship between personnel and vehicles from the personnel-vehicle association data table; extract key features such as latitude and longitude, driving path, checkpoint information, and vehicle usage frequency from the vehicle driving trajectory data table. Combined with the knowledge of illegal reselling of refined oil products, the features are screened and reduced in dimension to improve the accuracy and efficiency of subsequent analysis. The screened features are normalized to facilitate subsequent model training.
[0042] The third step is to use the feature data integrated by feature engineering to build machine algorithm model. According to the characteristics and requirements of the problem, select the appropriate machine learning algorithm, including logistic regression, support vector machine, decision tree, random forest, and neural network. Use the data set processed by feature engineering to train the selected machine learning algorithm to obtain a model for predicting the persons and events involved in the illegal reselling of refined oil. The performance and stability of the model are evaluated through cross-validation, precision, recall rate, F1 value and other evaluation indicators. The specific establishment of the machine learning model is as follows: Based on the extracted and screened feature data, a machine learning model for illegal reselling of refined oil products is established. The support vector machine (SVM) algorithm, random forest (RF) algorithm and deep neural network (DNN) algorithm are used to build classification models respectively, and the results of the three algorithms are integrated through a variety of ensemble learning methods, combined with multi-level integration strategies and dynamic weighting mechanisms to improve classification performance. Feature data for illegal reselling of refined oil products: Funds data: Extract features such as transfer amount, transfer frequency, and recipient information, and capture abnormal patterns in fund flows through kernel function mapping.
[0043] Communication data: Extract features such as call duration, call frequency, and number of callers and called parties, analyze communication frequency and patterns, and identify suspicious communication behaviors.
[0044] Personnel-vehicle association data: Utilizes the relationship between personnel and vehicles, usage time and other features, and identifies abnormal patterns of vehicle usage through kernel function mapping.
[0045] Vehicle driving trajectory data: extract features such as longitude and latitude, driving path, checkpoint information, and vehicle usage frequency, and analyze abnormal changes in vehicle trajectory.
[0046] By combining financial data and communication data, joint features such as "high frequency transfers and high frequency calls" are extracted to identify potential criminal behavior. At the same time, time series analysis is performed on transfer and call data to discover patterns related to specific events.
[0047] The support vector machine algorithm maps features to a high-dimensional space through a kernel function and searches for the optimal hyperplane for classification. By adjusting the kernel function type and regularization parameters of the SVM model, the generalization ability of the model can be improved and the risk of overfitting can be reduced.
[0048] Based on the processed feature data, the improved subspace random forest classification algorithm is used to build a classification model to improve the efficiency of model building. Using the extracted funds, communication modes, vehicle associations and vehicle driving trajectory data, a sub-feature space containing different types of original data is first constructed, and multiple decision trees are constructed based on multiple sub-feature spaces. Each tree is classified independently, and the final classification result is obtained through a voting mechanism. Using the characteristics of random forests, the importance of each feature is evaluated, and the feature selection is further optimized to improve the model performance. The random forest algorithm can process high-dimensional data and has a strong ability to resist overfitting. It improves classification accuracy by integrating the results of multiple decision trees. For the optimized feature data of capital flow, communication mode, vehicle association and vehicle driving trajectory data, the multi-layer perceptron MLP and K nearest neighbor algorithm KNN are used to build a classification model, design a suitable deep neural network structure, and optimize the parameters such as activation function, loss function and learning rate. The adaptive learning rate optimizer Adam is used to accelerate convergence and improve model performance. Through the deep learning model, multi-source feature data are integrated to capture complex illegal reselling behavior patterns of refined oil. Deep neural networks can handle nonlinear and complex feature relationships and improve the classification ability of the model by deepening the network layers.
[0049] The support vector machine, random forest, deep neural network and other algorithms are integrated, and the adaptive enhancement (AdaBoost) and gradient boosting (Gradient Boosting) are used for ensemble learning. Multiple weak classifiers are iteratively trained, and the dynamic weight adjustment mechanism is used to adjust the sample weight according to the error rate of each classifier. Finally, all weak classifiers are linearly combined into a strong classifier. Through the ensemble learning method, the advantages of different models are combined to reduce the variance and bias of a single model. The voting mechanism and the weighted fusion strategy based on model confidence are adopted to integrate the results of multiple base classifiers to obtain the final prediction result.
[0050] By evaluating the performance of the ensemble learning model and comparing it with a single model, we ensure its higher accuracy and stability in the analysis of illegal reselling of refined oil products.
[0051] The fourth step is to use the trained model to predict the integrated data set. The present invention first establishes the judgment rules for abnormal data (i.e., illegal reselling of refined oil related transaction data), and then finds the abnormal part of the transaction data in each account based on this series of judgment rules. According to the established machine learning algorithm, the fund transaction behavior and personnel communication behavior are analyzed, and the abnormal transaction data of the account that meets the judgment rules for illegal reselling of refined oil is obtained through the fund characteristics such as scattered in and out, and the personnel who may be involved in illegal reselling of refined oil are identified. Through the personnel communication relationship map, the key communication numbers and personnel subject information that meet the judgment rules for illegal reselling of refined oil are obtained, and the relationship network of the personnel involved is constructed to further explore the connection and communication mode between the personnel. By analyzing the association data of personnel and vehicles, the abnormal data of personnel, vehicles and personnel-vehicle association that meet the judgment rules for illegal reselling of refined oil are obtained, including personnel who frequently change vehicles and vehicles with abnormal frequency of use, multiple personnel using the same vehicle together and one person using multiple vehicles, and the use time and location distribution of abnormal vehicles. By analyzing the characteristics of vehicle trajectories, the DTW algorithm is used to calculate the similarity between the current trajectory of the vehicle and the historical trajectory to identify abnormal trajectory behavior, and obtain abnormal vehicle and abnormal vehicle trajectory information, including vehicles that frequently enter and exit suspicious locations and stay for abnormally long periods of time, and vehicle driving behaviors in different time periods. Combine financial transaction data, communication data, and personnel-vehicle association data to form a comprehensive portrait of personnel and vehicles, revealing potential criminal networks. Build a real-time monitoring system to dynamically analyze call and vehicle usage patterns, and combine machine learning algorithms to quickly identify abnormal behaviors. By graphically displaying the relationship between personnel and vehicles, it helps identify complex interaction patterns and intuitively display potential risks. According to the topological structure of the relationship network, the people involved in the case are divided into levels, such as leaders, middlemen, grassroots members, etc., to reveal the organizational structure and operation mode of illegal reselling of refined oil.
[0052] In the fifth step, the obtained capital data and communication data are subjected to correlation analysis, and a heterogeneous graph neural network is used to construct a heterogeneous graph, where nodes represent individuals and edges represent communication relationships and transaction relationships. A graph neural network (GNN) is used to propagate information and embed learning on the heterogeneous graph to extract node features. The learned features are used to classify the nodes, identify key personnel and their levels, and form user portraits. A transaction graph is constructed based on the capital data to analyze transaction behaviors and relationships. The graph analysis algorithm PageRank is used to identify the transaction center nodes, and the graph analysis algorithm HITS is used to identify different roles in the transaction graph, and the nodes in the transaction graph are sorted and scored. Based on the scoring results, key personnel are identified and hierarchical divisions are performed. Combining communication data and capital data, a graph clustering algorithm is used to divide the community. The internal and external structures of the community are analyzed to determine the core community where key personnel are located. The key personnel in the core community are further divided into hierarchies, such as Figure 3As shown. The obtained abnormal data of personnel, vehicles and personnel-vehicle association, abnormal vehicles and abnormal vehicle trajectory information are feature extracted, and the data of personnel using vehicles is converted into time series format, including usage time, usage frequency, vehicle change, etc. The processed data is input into RNN and TCN models, and the training model captures the time-dependent characteristics of personnel using vehicles, identifies the pattern of personnel frequently changing vehicles, especially frequent changes in a short period of time, which may indicate an attempt to cover up the whereabouts. At the same time, it is found that the use of vehicles by personnel in abnormal time periods (such as late at night or suddenly using them after a long period of non-use) may be related to illegal activities; the graph neural network (GNN) is used to identify the dynamic changes of the personnel-vehicle association network and discover abnormal relationship changes. The long time series deep learning method (LSTM) is used to extract time-dependent features, identify the behavior of vehicles frequently entering and exiting illegal transaction hotspots, especially the high-frequency entry and exit within a specific time period; the anomaly detection technology is used to discover the behavior of vehicles staying in abnormal places for a long time, identify the abnormal change pattern of vehicle driving trajectory, and discover potential illegal activity routes.
[0053] Based on the data processing method of deep learning, according to the correlation between capital data and communication data, personnel portraits are constructed, and the results of correlation analysis are used to identify suspected key personnel and gang members. According to the different dimensions of user portraits, key personnel are divided into levels. The internal organizational structure and business methods of the illegal reselling of refined oil gangs are obtained, so as to expand and dig out more clues to find illegal reselling of refined oil gangs. By capturing the time-dependent characteristics of personnel using vehicles, potential illegal reselling of refined oil behavior subjects are identified, abnormal personnel-vehicle relationship changes are discovered in time, illegal activities are warned in advance, and the effectiveness of monitoring and management is improved. By fusing and analyzing the personnel-vehicle relationship and vehicle trajectory data, the time series deep learning model is used to capture the multi-dimensional time-dependent characteristics. Combining the results of RNN and TCN models, the abnormal behaviors of personnel and vehicles are comprehensively evaluated to improve the recognition accuracy. At the same time, the abnormal behaviors are clustered to form a behavior pattern library, which provides a basis for subsequent risk assessment and decision-making. A real-time data update and monitoring mechanism is established, and the prediction ability of the time series model is used to dynamically monitor the behavior changes of personnel and vehicles and timely warn of potential risks.
[0054] The present invention integrates data collection, data processing, feature engineering, machine learning algorithm modeling, personnel identification and hierarchical division into a complete illegal oil reselling system. Through the analysis and prediction of real-time data, the system can monitor the occurrence of illegal oil reselling activities in real time, issue early warnings to relevant departments in a timely manner, and improve the effectiveness of crackdown and prevention. According to the actual operation situation and user needs, the system is continuously optimized and updated to improve the system performance and user experience.
[0055] The specific rules for judging illegal reselling of refined oil products proposed by the present invention are as follows: 1. Research and analysis of capital behavior Analysis of fund transaction behaviors includes the downstream fund aggregation behaviors in the form of "scattered in and gathered out", the concentrated fund outflow behaviors to the upstream, and the bank card deposit and withdrawal behaviors, and analysis of deposit and withdrawal and fund transaction characteristics; for key suspicious transaction accounts, analysis is made as to whether they are low-level sales personnel, middle-level distribution personnel, or high-level regional business owners, and through analysis of the number of transactions, transaction time, and transaction inflow and outflow amounts, the top-level boss accounts, middle-level distributor accounts, low-level sales personnel accounts, and customer accounts of illegal reselling of refined oil gangs are identified; use of personnel and account summary information to identify account transaction time, transaction amount, transaction frequency, etc., which reflect the illegal and criminal behavior of illegal reselling of refined oil gangs.
[0056] The criminal gangs that illegally resell refined oil products have the characteristics of scattered inflows and outflows in the process of fund transactions, and different gang members have different manifestations of scattered inflows and outflows according to their different levels. The main manifestation is that the account transactions of the bottom-level sales personnel in the gang are mostly WeChat transactions, and the WeChat transaction flow has the characteristics of "scattered inflows and outflows". There are multiple consecutive incoming items, but the amount of the incoming items is small. After multiple incoming items, there will be a large amount of outflows, and the amount of outflows is about the sum of the previous multiple incoming items. The scattered inflows and outflows of the fund transactions in the accounts of middle-level distributors are mostly in the amount of several thousand yuan, and the amount of outflows is mostly more than 10,000 yuan, and there will be multiple consecutive outflows. The scattered inflows and outflows of the funds in the local bosses' accounts are not representative, but it still shows that large amounts of incoming items come from the accounts of middle-level distributors.
[0057] 1. Analysis of illegal reselling of refined oil products Total transaction volume: more than 100,000 / month, more than 1 million / month; Transaction number determination: more than 300 transactions per month; Transaction difference determination: the difference between the total amount of inflow and outflow is within 10%; Determination of the number of counterparties: more than 30 people / month, more than 50 people / month; Determination of scattered in and out: determined by the number of transactions.
[0058] Third-party feature determination: Determine based on whether the third-party keywords are found in the opponent’s account name, summary description, and remarks.
[0059] Analyze the accounts of bottom-level sales personnel, middle-level distributors, and top-level bosses involved in the illegal reselling of refined oil.
[0060] Account judgment of the bottom-level sales staff: Most transactions are mainly WeChat transactions, and the WeChat transaction flow has the characteristics of "scattered in and concentrated out". The amount of incoming funds is mostly 100-300 yuan, and the amount of outgoing funds is mostly 2000-4000 yuan. And after multiple consecutive funds are received, there is a large outgoing amount, which is about the sum of the previous multiple incoming amounts.
[0061] Mid-level distributor accounts: The incoming transaction amounts are mostly between RMB 2,000 and RMB 4,000, and the outgoing transaction amounts are mostly between RMB 10,000 and RMB 50,000, with multiple consecutive outgoing transactions.
[0062] Local boss account: an account that meets the characteristics of scattered in and out transactions or has a transaction amount that is an integer multiple of 100,000 and has more than 5 transactions on the same day.
[0063] 2. Analysis of the operation mode of illegal reselling of refined oil products The illegal reselling of refined oil is usually carried out in a gang-like manner, forming a pyramid structure to collect oil purchase funds from each layer and earn the difference. The bosses who illegally resell refined oil often organize local oil dealers to carry out illegal reselling of refined oil. Local oil dealers are usually in the middle of the illegal reselling of refined oil. They sell private oil to lower-level retail oil dealers and collect oil purchase money, and buy private oil from upper-level oil dealers and pay for the oil. Retail oil dealers sell private oil to scattered terminal oil buyers. The illegal reselling of refined oil business gangs purchase oil from oil dealers and distribute it layer by layer, earning the difference layer by layer, forming a pyramid-shaped refined oil illegal business gang operation model with a fixed hierarchical structure and obvious transaction characteristics, forming a cross-regional refined oil illegal business crime form.
[0064] (II) Analysis of the communication behaviors of personnel illegally reselling refined oil products Analyze communication behaviors, including call duration, communication frequency, number of callers and call time periods, to identify abnormal communication patterns; for key suspicious numbers, analyze whether they are low-level sales personnel, middle-level distributors or high-level managers. By analyzing the number of calls, call time and contact objects, identify the communication network of top-level managers, middle-level distributors and low-level sales personnel in illegal reselling of refined oil gangs. Use personnel and communication record summary information to identify abnormal call duration, call frequency and call time periods, reflecting the illegal and criminal behavior of illegal reselling of refined oil gangs. As shown in Table 1, this is a statistical table of the proportion of words related to communication content.
[0065] Table 1
[0066] Determination of communication object: Analyze the communication relationship between the number holder and the suspicious person, including phone calls, text messages, WeChat, overseas communication software (Telegram), etc. There are more than 10 communication records with specific objects.
[0067] Determination of communication content: Analyze communication content, such as text messages, WeChat chat records, overseas communication software (Telegram) group chat records, etc., to identify whether it involves information such as refined oil transactions, prices, quantities, and transportation; identify whether the communication content contains keywords such as "oil", "gasoline", "diesel", "gas station", and other words related to the illegal resale of refined oil, and whether the corresponding communication content accounts for more than 50% of the total communication content.
[0068] Determination of communication time: Analyze the communication time to determine whether the communication behavior coincides with the time node of the fund transaction behavior, including the communication behavior before and after the inflow and outflow of funds. If the interval between the communication time and the suspicious fund transaction behavior is within 20 minutes, it is determined to be a suspicious communication record. If most of the communication time of an account is concentrated between midnight and 2 am, it can be preliminarily determined that its communication behavior has a special purpose.
[0069] Communication frequency determination: Analyze the communication frequency to determine whether the communication behavior is abnormally frequent, including frequent communication in a short period of time, which may have special purposes. The number of communications with a specific object exceeds 10 times / day, and the average communication time interval with a specific object is less than 5 minutes.
[0070] Determination of communication location: Analyze the communication location to determine whether the communication behavior is consistent with the location of the fund transaction behavior, such as communication near a gas station or other places related to refined oil. The number of communications with specific people at a specific location exceeds 50 times / week, and the average communication distance with the specific location is less than 5km.
[0071] Communication network analysis: Through comprehensive analysis of communication objects, communication content, communication time, communication frequency, communication location and other information, a communication network is constructed to identify core figures and key links, thereby revealing the structure and operation mode of criminal gangs that illegally resell refined oil.
[0072] (III) Analysis on the relationship between personnel and vehicles involved in illegal reselling of refined oil Analyze the personnel-vehicle association data, including personnel's vehicle use, vehicle ownership, etc., and analyze the role of personnel and vehicles in the reselling network. Analyze whether the personnel and vehicle information of key suspicious personnel and vehicles shows a gang pattern, provide a basis for the gang's expansion, and reflect the association network of illegal reselling of refined oil gangs.
[0073] Personnel information determination: Analyze the identity information of the person, such as name, ID number, address, etc., to see whether it is consistent with or related to the information of known persons involved in the case.
[0074] Vehicle information determination: Analyze the type of vehicle to see if it matches the type of vehicle used to illegally resell refined oil, such as tankers, vans, etc. Analyze the owner of the vehicle to see if he is a known person involved in the case or has a connection with a person involved in the case.
[0075] Determination of person-vehicle association information: Analyze the closeness of the social relationship between the vehicle owner and the person. For example, if the vehicle owner is a relative, friend, colleague, etc. of the person, the closer the relationship, the stronger the association.
[0076] Kinship relationships: parent-child relationship (100%), husband-wife relationship (90%), sibling relationship (80%), and other kinship relationships (70%).
[0077] Friendship: close relationship (90%), general friends (70%), ordinary friends (50%).
[0078] Relationships with colleagues: close relationship (80%), general colleague (60%), ordinary colleague (40%).
[0079] (IV) Analysis of the driving trajectory of vehicles illegally reselling refined oil like Figure 4 As shown, the vehicle trajectory data includes vehicle driving routes, checkpoint entry and exit times, vehicle driving time periods and other features, identifies suspicious vehicle driving trajectories, analyzes illegal reselling routes of refined oil transportation, compares multiple vehicle driving trajectories, and discovers potential related networks; evaluates vehicle usage during abnormal time periods (late at night or holidays), and identifies potential illegal trading activities.
[0080] Vehicle driving trajectory determination: Analyze whether the vehicle's driving route frequently changes driving direction and detours to specific areas, or makes unreasonable route choices within a specific time period. If a vehicle detours a specific area more than three times in a single trip, it will be determined to be a suspicious vehicle.
[0081] Vehicle checkpoint passing judgment: In order to avoid inspection, vehicles may choose to bypass checkpoints, resulting in multiple passes through adjacent checkpoints in a short period of time. If a vehicle passes through adjacent checkpoints more than 3 times per hour, it will be judged as a suspect vehicle. A vehicle may hide oil products in a remote location to avoid inspection, resulting in a long period of time without passing the checkpoint. The judgment basis is: Vehicle traffic frequency change rate: (current week traffic frequency - last week traffic frequency) / last week traffic frequency * 100% Normal driving: The change rate of traffic frequency is small, such as less than 10%. Abnormal driving: The change rate of traffic frequency is large, such as more than 50%, or it suddenly drops from 3-4 times per week to 0 times.
[0082] Judgment of vehicles traveling together: If a vehicle appears at the same checkpoint as a suspected vehicle more than 5 times a week and the coincidence degree of the driving routes with the suspected vehicle is more than 80% within a specific past time period, it is judged as a vehicle traveling together.
[0083] (5) Research and analysis on the operation mode of illegal refined oil reselling gangs 1. Analysis and research on gang parsing based on clustering of online chat information Carry out targeted analysis of online social groups for illegal refined oil reselling criminal gangs in their use of social networks for information dissemination and criminal business activities, and explore the characteristics of the online activities of illegal refined oil reselling gangs.
[0084] (1.1) Analysis of personnel groups Based on the names of social groups of the persons involved in illegal refined oil business activities, conduct preliminary data cleaning and grouping of group chats, and identify and obtain the corresponding main body information of suspicious WeChat IDs and WeChat IDs.
[0085] (1.2) Cluster analysis of group chats According to the key information presented by the online social groups of illegal refined oil reselling criminal activities, use the GloVe + k-means++ algorithm to clean and preprocess the text data in the WeChat group chat records. For example, stop words (such as "of", "is", "in", etc.) contribute little to text analysis and should be removed to reduce noise. Punctuation marks generally do not carry semantic information and can be removed to simplify the text, etc., and each text is converted into a word sequence (for Chinese texts, use the word segmentation tool jieba to segment the text into independent word sequences). At the same time, use the GloVe algorithm to construct a word vector model, summarize all the cleaned text data to form a training corpus, and use the GloVe algorithm to train the corpus to generate word vectors. To improve the clustering discrimination of the algorithm, select the improved version k-means++ algorithm of k-means. The relevant formula is as follows: Each individual within the group x i The probability P( x i ) of being inherited into the next generation group:
[0086] Among them, M is the number of individuals within the group, f ( x i ) is the fitness of each individual in the group.
[0087] The cumulative probability of each individual q i :
[0088] Among them, n is the total number of groups.
[0089] Use the k-means++ algorithm to perform cluster analysis on the obtained word vectors, and generate cluster analysis results by specifying the number of clusters k and initializing the cluster center points.
[0090] For each cluster, you can understand the theme and background of the cluster by looking at the words and chat records in it. For example, a cluster may contain words and chat records related to criminal activities such as illegal operation of refined oil products, indicating that the cluster may be a criminal gang engaged in illegal operation of refined oil products. Based on the clustering results, you can preliminarily judge the possible criminal gangs and conduct further investigation and verification, such as finding relevant evidence and questioning witnesses.
[0091] 2. Group analysis and judgment based on base station location information clustering like Figure 5 As shown, by screening and identifying the call record data of suspected illegal refined oil business operators, cluster analysis is performed on the suspected illegal refined oil business dens based on the call base station locations, and the call base station-related objects of tank truck drivers are identified. A cluster identification of personnel calls based mainly on the call base station locations is formed, and the locations of suspicious call base stations are output, and a list of key overlapping locations and map locations are locked.
[0092] Corresponding to the aforementioned embodiment of a method for determining illegal reselling of refined oil, the present invention also provides an embodiment of a device for determining illegal reselling of refined oil.
[0093] See also Figure 6 An embodiment of the present invention provides an illegal reselling of refined oil judging device, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement an illegal reselling of refined oil judging method in the above embodiment.
[0094] An embodiment of the device for analyzing and judging illegal reselling of refined oil provided by the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, the processor of any device with data processing capability reads the corresponding computer program instructions in the non-volatile memory into the internal memory and runs them. From the hardware level, if Figure 6 As shown, it is a hardware structure diagram of any device with data processing capability where the illegal reselling of refined oil research and judgment device provided by the present invention is located, except Figure 6In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.
[0095] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.
[0096] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.
[0097] An embodiment of the present invention further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, a method for determining illegal reselling of refined oil in the above embodiment is implemented.
[0098] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.
[0099] The present invention also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method for determining illegal reselling of refined oil is implemented.
[0100] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.
Claims
1. A method for determining illegal reselling of refined oil, characterized in that: The method comprises the following steps: (1) Obtain the financial transaction data, communication data, location tracking data of refined oil vehicles, and identity data associated with refined oil vehicles of the persons involved in the case to construct a data set; (2) Preprocess the data set; extract the characteristics of fund data, analyze the stability of fund flow, identify abnormal transaction patterns and recipient groups; extract the characteristics of communication data, analyze abnormal call records, compare communication activity, identify potential connections based on the characteristics of fund data, and combine the identity data associated with refined oil vehicles to form a personnel-vehicle association data table, and integrate it with the geographic location tracking data of refined oil vehicles to form a refined oil vehicle driving trajectory data table; (3) Extract the personnel-vehicle usage relationship characteristics and vehicle trajectory characteristics from the personnel-vehicle association data table and the refined oil vehicle driving trajectory data table, and use the machine learning model to determine whether there is illegal reselling of refined oil by combining the capital data and communication data; (4) Based on the recognition results of step (3), a heterogeneous graph neural network is used to construct a heterogeneous graph. The nodes represent the persons involved in the case, and the edges represent the communication relationships and financial transaction relationships. A transaction graph is constructed based on the financial data, and the transaction behaviors and relationships are analyzed. The transaction center nodes and different roles in the transaction graph are identified, and key personnel are identified and divided into levels. Combined with the communication data and financial data, a graph clustering algorithm is used to divide the community, analyze the internal and external structures of the community, determine the core community where the key personnel are located, and further divide the key personnel in the core community into levels to determine the multi-level criminal network.
2. The method for determining illegal reselling of refined oil according to claim 1, characterized in that: The communication data includes call time, call duration and call location data obtained from the communication operator platform, and the vehicle-related identity data includes the identity card, household registration or driver's license data of the persons involved in the case obtained from the public security network data lake, which is used for identity authentication, background investigation and personnel-vehicle association analysis.
3. The method for determining illegal reselling of refined oil according to claim 1, characterized in that: The data set preprocessing includes deduplication of the data set, using an intelligent denoising algorithm to improve data accuracy, applying the XGBoost regression algorithm to accurately predict and fill in missing values, using time series analysis and cross-feature generation to extract deeper information, using the Yeo-Johnson power transformation algorithm to process non-normally distributed data to improve data consistency, and using the SMOTE technology to balance positive and negative samples.
4. The method for determining illegal reselling of refined oil according to claim 1, characterized in that: Extract key features of transfer amount, transfer frequency, and recipient information from fund data, analyze the change in transfer amount to identify anomalies, extract key features of call duration, call frequency, and number of callers and recipients from communication data, discover abnormal call records, conduct time series analysis on transfer and call data, and identify potential criminal behavior.
5. The method for determining illegal reselling of refined oil according to claim 1 is characterized in that: The identity association relationship features and usage time relationship features of personnel and vehicles are extracted from the personnel-vehicle association data table, and the vehicle trajectory features and vehicle usage frequency features are extracted from the vehicle driving trajectory data table. The extracted features are screened, reduced in dimension, and normalized.
6. The method for determining illegal reselling of refined oil according to claim 1 is characterized in that: The machine learning model specifically includes: using a subspace random forest classification algorithm to build a classification model, using the extracted funds, communications, vehicle associations and vehicle driving trajectory data to build a sub-feature space containing different types of original data, building multiple decision trees based on multiple sub-feature spaces, each tree independently classifies, and obtaining the final classification result through a voting mechanism, evaluating the importance of each data feature, further optimizing feature selection, and based on the optimized feature data, using a multi-layer perceptron MLP and K nearest neighbor algorithm KNN to build a deep neural network classification model to judge and analyze the illegal reselling of refined oil behavior patterns.
7. The method for determining illegal reselling of refined oil according to claim 1 is characterized in that: The machine learning model uses adaptive enhancement and gradient boosting for ensemble learning. It iteratively trains multiple weak classifiers and adopts a dynamic weight adjustment mechanism to adjust the sample weights according to the error rate of each classifier. Finally, all weak classifiers are linearly combined into a strong classifier. A voting mechanism and a weighted fusion strategy based on model confidence are used to integrate the results of multiple weak classifiers to obtain the final prediction result.
8. The method for determining illegal reselling of refined oil according to claim 1 is characterized in that: Based on the machine learning model, the fund transaction behavior and personnel communication behavior are analyzed. Through the scattered and collected fund characteristics, the abnormal account transaction data that meets the illegal reselling of refined oil is obtained; through the personnel communication relationship map, the key communication numbers and personnel subject information that meet the illegal reselling of refined oil are obtained, and the abnormal communication mode is discovered; by analyzing the association data of personnel and vehicles, the abnormal data of personnel, vehicles and personnel-vehicle association that meet the illegal reselling of refined oil are obtained, including personnel who frequently change vehicles and vehicles with abnormal usage frequency, multiple personnel using the same vehicle and one person using multiple vehicles, and the use time and location distribution of abnormal vehicles; by analyzing the vehicle trajectory characteristics, the DTW algorithm is used to calculate the similarity between the current trajectory of the vehicle and the historical trajectory to identify the abnormal trajectory behavior, and the abnormal vehicle and abnormal vehicle trajectory information are obtained, including vehicles that frequently enter and exit suspicious locations and stay for abnormally long periods of time, and the driving behavior of vehicles in different time periods; Combine financial transaction data, communication data and personnel and vehicle association data to form a comprehensive portrait of people and vehicles, revealing potential criminal networks.
9. A device for judging illegal reselling of refined oil, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a method for determining illegal reselling of refined oil products as described in any one of claims 1-8 is implemented.
10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by the processor, a method for determining illegal reselling of refined oil products as described in any one of claims 1-8 is implemented.
Citation Information
Patent Citations
Case handling investigation and evidence collection data cloud storage and retrieval analysis method
CN111090779A
Anti-money laundering studying and judging method
CN112668614A
System, method and device for realizing similar target person identification processing based on fund transaction relation data, processor and medium thereof
CN113641906A
Criminal gang mining analysis method based on optimized knowledge graph
CN116822628A
Product oil suspicious transport vehicle GPS track studying and judging method
CN118330698A
Cited By
Indirect abnormal transaction identification method and device and indirect abnormal transaction model construction method and device
CN120806967A
