A Phishing Account Detection Model via Network Embedding for Ethereum
Jintao Luo, Jiwei Qin, Ruijin Wang, Li Lü
Abstract
As the first blockchain platform to support smart contracts, Ethereum has gained popularity and breeds various cybercrimes. Many phishing accounts on Ethereum take advantage of the blockchain’s anonymity to participate in illegal acts. To this end, to solve the transaction security problem caused by phishing accounts on Ethereum, this brief proposes a network embedding-based phishing account detection model. Firstly, we crawl the history of transactions from both labeled phishing and non-phishing accounts and use a new method to build these transactions as a transaction network, where each transaction edge is accompanied by additional transaction information for all source nodes of the target node. Then we propose a new random walk-based network embedding algorithm named bias2vec to obtain the embeddings of nodes as the feature inputs. Finally, we classify the accounts into phishing and non-phishing ones by different classifiers such as lightGBM and XGBoost. Experiments show that our proposed phishing account detection model achieves effective detection performance under different classifiers.