ScamRadar: Identifying Blockchain Scams When They are Promoting

The vigorous growth of cryptocurrencies has infiltrated into social media, manifested by extensive advertisement campaigns popping up on platforms such as Telegram and Twitter. This new way of promotion also introduces an additional attack surface on the blockchain community, with a number of recently-identified scam projects distributed via this channel. Despite prior efforts on detecting scams, no existing works systematically study such promotion channels for detecting on-going scam projects. In this paper, we take the first step to propose a novel approach that leverages promotion messages to detect on-going blockchain-themed scam projects. We extract features spanning four dimensions: word frequency, sentiment analysis, potential profit margin and URL information. For automated detection, we first construct two datasets, and then train a classifier on them that achieves a high accuracy (over 90%). Through our large-scale study, we have identified 69, 148 highly suspicious scam messages, relating to 8, 247 blockchain addresses. By tracing the confirmed scam addresses, we further identify a cumulative illicit income of around 900 ETH. Our findings should help the blockchain community understand and promptly detect scam projects when they are promoting.