Features combination for the detection of malicious Twitter accounts

Abstract

Microblogging social networks are easily subverted by automated fake identities that amass disproportionately large influence. In this paper we present an effort to profile and screen such kind of accounts from existing and original ground truth obtained from the Twitter platform. Seventy-one explanatory properties solely extracted from profile and timeline information are evaluated and used to compare the efficacy of common supervised machine learning methods at this classification task. Results confirm that feasible and largely effective detection devices can be constructed for the problem at hand.