Behavior Change as a Signal for Identifying Social Media Manipulation

A research team from Indiana University, led by Gangani Ariyarathne and including Filippo Menczer, has published a novel approach to detecting social media manipulation by analyzing behavioral change patterns. Their paper, accepted to the 18th ACM Web Science Conference 2026, introduces a method that tracks how accounts evolve their posting behavior over time, addressing a key limitation of existing detection systems that rely on static features. The researchers demonstrate that authentic accounts maintain consistent behavioral patterns, while manipulative accounts exhibit either extreme stability or volatility in their actions, providing a new signal for identification.

The technical approach uses Behavioral Languages for Online Characterization (BLOC) to represent account activities as sequences of symbols, then segments these sequences to measure changes between consecutive time periods. By converting accounts into feature vectors capturing the distribution of behavioral change values, the team trained supervised classifiers that achieved good accuracy in two detection tasks: identifying social bots and uncovering coordinated inauthentic behavior campaigns. The research reveals that social bots show either very low or very high behavioral change, while coordinated accounts within the same campaign exhibit highly similar change patterns. This method represents a significant advancement in detecting evolving manipulation tactics that evade traditional detection systems.