Finding Users we Trust: Scaling up Verified Twitter Users Using their Communication Patterns

Abstract

We present a technique to identify Twitter users we trust to be regular Twitter users and not spam or fake accounts. The technique starts with an initial set of trusted users based on Twitter's verified users and recursively includes other users the trusted users communicate with. Conversations must be initiated by trusted users. We show that this technique produces a set of users that is over 200 times larger than Twitter's verified users. Our evaluation shows that the share of non-spam users within the resulting set of trusted users is more than 92% while the share of non-spam users within all tweeting users is 74%.