Abuse and spam in Twitter have long been a pressing issue, and in response, Twitter regularly purges (i.e., suspends in mass) accounts that violate Twitter Rules. However, there is no available information about the characteristics and activities of these regularly purged users. We have developed a novel and comprehensive measurement mechanism to identify millions of purged Twitter users and collect their tweets. We have identified 2.4M purged users and collected 1M tweets made by them over eight months. Using our dataset, we perform a retrospective analysis to characterize their account properties and behavioral activities. We analyze their tweet content to identify their role and abuse strategy over-time.
Our analysis shows that the abuse on Twitter is pervasive globally and not confined in mere spamming. Alarmingly, more than 60% of the purged users survived on Twitter for more than two years. We observe that politics is a major theme among the purged users irrespective of language and location, and these politically motivated users spread controversial content consistently over time. However, the spammers reorient their agenda across time to participate in multiple marketing campaigns. We also discover interaction and associated communities among purged users. Our analysis sheds new light on the evolving nature of abuse in Twitter that can help researchers understanding the characteristics and behavior of emerging malicious users to develop an effective defense system.