Hook, Line and Rug Pull: Untangling the Web of Crypto Scam Sites

Abstract

The rapid growth of cryptocurrencies has been accompanied by a rise in illicit activities within the digital asset ecosystem. Among these, rug pulls have emerged as a prevalent form of fraud in which project creators promote and sell tokens under false pretenses, attract funds, and then abruptly withdraw liquidity, leaving investors with worthless assets. This study introduces a novel detection approach based on linguistic and structural cues from these websites. Our findings reveal that legitimate cryptocurrency websites differ significantly from rug pulls in both language use and structure. Tree-based classifiers demonstrate strong potential for identifying rug pull schemes, achieving classification accuracies of over 82%. These results not only enhance fraud detection capabilities but also contribute to a deeper understanding of the characteristics and tactics behind rug pull schemes.