Pheromone-based graph embedding algorithm for Ethereum phishing detection

Abstract

Phishing scams pose significant risks to Ethereum, the second-largest blockchain-based cryptocurrency platform. Traditional methods for identifying phishing activities, such as machine learning and network representation learning, struggle to capture the temporal and repetitive transaction patterns inherent in Ethereum’s transaction network. To address these limitations, we propose a Pheromone-based Graph Embedding Algorithm (PGEA), which leverages pheromone mechanisms and a taboo list inspired by ant colony behavior to enhance subgraph sampling. This approach improves the identification of phishing activities by ensuring subgraph homogeneity and isomorphism during the sampling process. In our methodology, Ethereum transaction data is collected from known phishing addresses to construct a transaction network graph. The PGEA guides subgraph sampling, producing sequences that are transformed into node embeddings using word2vec. These embeddings are then classified using a Support Vector Machine (SVM) to distinguish between legitimate and malicious nodes. Experimental results demonstrate the superiority of our model over existing methods. PGEA achieves an accuracy of 87.18%, precision of 91.01%, recall of 84.82%, and F1 score of 86.91%, outperforming baseline approaches such as Deepwalk, Node2vec, and Graph2vec. These results highlight the efficacy of PGEA in detecting phishing addresses, contributing to a more secure Ethereum ecosystem.