A Blockchain Phishing Scam Detection Method Based on Ethereum Transaction Subgraph

Abstract

In recent years, the field of blockchain technology has witnessed significant growth, particularly in its extensive application within the financial sector. However, this progress has also brought a concerning issue whereby an increasing number of users are becoming vulnerable to phishing scams. Many extant studies are focused on this issue; nonetheless, a paucity of emphasis has been placed on the comprehensive characterization of Ethernet transaction data from multiple perspectives which enhances the precision and generalization capabilities of the model. Consequently, within the context of this paper, we introduce an innovative detection model. In an effort to mitigate overhead while ensuring the acquisition of substantial information, we initially preprocess Ethernet transaction data and establish transaction subgraphs. Then, we leverage a graph autoencoder to amalgamate local structural information, global structural information, and node features within the graph, culminating in the detection of phishing scams. Our model significantly enhances detection accuracy by exploiting multi-view features. We implemented this method on the dataset from TSGN, and the results show that the accuracy of our method is 95.39%, which is at least 4.07% higher than other related methods.