Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a 20-minute conference talk that introduces ScamSweeper, a novel framework for detecting illicit accounts in Web3 scams on the Ethereum blockchain. Learn how Web3 applications have become targets for scammers who mimic legitimate platforms and regular transaction activity to deceive users. Discover the limitations of current phishing account detection tools that rely on graph learning or sampling algorithms, particularly when dealing with large-scale transaction networks with temporal attributes that follow power-law distributions. Understand how ScamSweeper addresses these challenges by focusing on the dynamic evolution of transaction graphs, utilizing structure-temporal random walks to sample transaction networks and capture both temporal and structural features. Examine the framework's use of variational transformers to analyze dynamic evolution patterns over time. Review experimental results from a large-scale dataset consisting of Web3 scams, phishing accounts, and normal accounts from the first 18 million block heights on Ethereum, demonstrating ScamSweeper's superior performance with a 17.29% advantage in weighted F1-score for Web3 scam detection and 17.5% advantage in F1-score for phishing node detection. Gain insights into large-scale dataset collection methods, dataset attribute distinctions, computational cost reduction techniques, and practical applications for real-world Ethereum transaction analysis while learning to incorporate dynamic evolution analysis into malicious behavior research.
