The characteristics of rumor spreaders on Twitter: A quantitative analysis on real data

In this paper we study a dozen of rumors on Twitter to find new insights in user characteristics and macro patterns in the process of rumor spreading. The collection and curation of data has left us with 12 rumor datasets out of 56,852 tweets from 43,919 users. The analysis over data shows users with lower ratio of following-to-follower are more probable to spark the rumor diffusion while users with the higher ratio are those who keep the flame alive. Furthermore, most users participate in the process of rumor spreading only once which implies the nature of rumor spreading is not a recurrent activity. However, among those users who engage with multi posts, the extreme change of state from rumor spreader to anti-rumor spreader happens to users with higher ratio of following-to-follower. We discuss these findings by employing the theory of planned behavior. Finally, analyzing the process of rumor spreading at the macro level revealed the existence of two distinctive patterns. Further investigations showed the extent of time gap between the beginning of rumor and anti-rumor diffusion plays the major role in emerging of these patterns. This phenomenon is explained by the shift in subjective norm toward rumors on social media.