Reproduce “Misleading Repurposing on Twitter” on OpenResearch

About this paper

Tuğrulcan Elmas, Rebekah Overdorf, Karl Aberer

This research provides a formal definition and the first large-scale empirical analysis of "misleading repurposing" on Twitter, where accounts drastically change their identity to exploit existing follower bases. The study develops a machine learning classifier that achieved an F1-score of 94.0% and an AUC of 87.6% in detecting this behavior, estimating over 180,000 such repurposings on popular accounts and revealing distinct behavioral patterns like mass tweet deletions and prior follower accumulation.

Key insights

• Misleadingly repurposed accounts often possess significantly higher follower counts, suggesting malicious actors acquire or cultivate popular accounts.
• Such accounts frequently engage in follower growth schemes and extensively delete prior tweets to obscure their original identity and purpose.
• The phenomenon exploits legitimate platform features, allowing malicious actors to bypass credibility filters and making retrospective data analysis challenging for researchers.

Summary by alphaXiv.