그래프 뉴럴 네트워크 기반 사이버 범죄 계좌 탐지를 적용한 감사 가능한 블록체인 시스템

Abstract

Cybercrimes that exploit the anonymity of blockchain are increasing in the blockchain system. They steal users’ assets and destabilize the blockchain network. Cybercriminal account detection methods have been studied to prevent cybercrimes in the blockchain system. However, the existing methods are only effective for cybercriminal account detection but do not take into account the blockchain system itself. In real-world scenarios, it is necessary to regulate cybercrimes in the blockchain system, such as blacklisting or blocking accounts, to protect users’ assets. In this paper, we propose an auditable blockchain system integrating an improved cybercrime detection model by supervised learning for graph classification. The proposed system provides a blacklist of cybercriminal accounts to prevent users from transferring their assets to cybercriminal accounts. It can also determine whether the unblacklisted account is a cybercriminal account or not based on a graph neural network. We conducted simulations to evaluate the performance of the proposed system with the cybercriminal account detection model. Simulation results show that the proposed detection model outperforms the existing methods by a 0.18 F1-score. In addition, the proposed blockchain system can transfer users’ assets in an average of ten milliseconds.