Graphical Structure Model for Abnormal Account Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies lack an effective method for preventing and controlling abnormal accounts, which can disrupt platform operations and pose risks, as they often rely on manual reporting and are inefficient in identifying and mitigating batch registration and authentication activities by machines.
Innovation Solution
A graphical structure model is used to integrate node and edge features from account relationship networks, calculating embedding vectors and prediction probabilities to identify and prevent abnormal accounts by building a network based on registration and authentication data, including device, network, and identity information, and training a model to determine abnormal nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual reporting and processing methods are used for abnormal accounts, then operational simplicity is maintained, but detection efficiency and response speed deteriorate
Solution Approach 1:
The patent replaces manual mechanical reporting and processing with an automated graphical structure model system that automatically detects abnormal accounts through embedding vector calculations and prediction probabilities, significantly improving detection efficiency while reducing reliance on human intervention
Solution Approach 2:
The system enables self-service detection by automatically monitoring account behaviors, calculating risk scores, and identifying abnormal accounts without requiring manual reporting, allowing the platform to autonomously detect and respond to abnormal activities
2Productivity
If batch registration and authentication by machines are allowed, then user registration speed is improved, but account security and platform reliability deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-calculating embedding vectors for accounts, devices, and networks during registration, and by training the graphical structure model in advance to predict abnormal behaviors before they occur, enabling proactive security measures that maintain both registration speed and account security
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring account behaviors, updating prediction probabilities based on observed patterns, and adjusting the graphical structure model to improve detection accuracy over time, thereby maintaining security while allowing efficient batch processing
3Measurement precision
If comprehensive account relationship network analysis is implemented, then detection accuracy is improved, but computational complexity and processing time worsen
Solution Approach 1:
The patent applies segmentation by dividing the complex account relationship network into separate graphical structure models for different entity types (accounts, devices, networks), allowing parallel processing and reducing computational complexity while maintaining comprehensive detection accuracy through integrated analysis
Solution Approach 2:
The system uses parameter changes by transforming raw account data into embedding vectors that capture essential features in a compressed form, and by adjusting prediction thresholds dynamically, thereby reducing processing time while maintaining high detection accuracy
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
Implementations of the present specification disclose a method, an apparatus, and a device for abnormal account prevention and control based on a graphical structure model. A solution includes: building in advance a suitable account relationship network based on data relating to account registration and/or authentication, performing feature integration and defining a graphical structure model, training the graphical structure model by using labeled samples, calculating embedding vectors and a prediction probability of nodes after multiple iterations in a hidden feature space by using the trained graphical structure model, and performing abnormal account prevention and control on the nodes based on the embedding vectors and prediction probability.