Detecting Phishing Scams on Ethereum Using Graph Convolutional Networks with Conditional Random Field

Abstract

As an innovative technology of great significance, blockchain has been widely used in various walks of life. Meanwhile, scams have experienced rapid growth on the blockchain, the phishing scam is a classic fraud among them. Ethereum is the largest blockchain platform that supports smart contracts, and its ecosystem has been under serious threat due to phishing scams. Consequently, phishing scam detection is particularly critical for Ethereum to avoid economic loss. However, accounts on Ethereum are represented as strings without features, which brings great difficulties for detection. To combat this challenge, we propose an effective method based on Graph Convolutional Network (GCN) with Conditional Random Field (CRF) to detect phishing scams. Firstly, we process accounts and their neighbors with transaction records to build transaction graphs. Due to lack of portrait information, we adopt DeepWalk to provide initial features for each node. Then graph representations are learned through GCN with CRF. The extensive experiments show that the proposed model performs better than comparison methods on both precision and recall, which indicates that our method can effectively identify phishing scams on Ethereum.