Characterizing and Detecting State-Sponsored Troll Activity on Social Media

The detection of state-sponsored trolls operating in influence campaigns is a critical and unsolved challenge for the research community, which has significant implications beyond the online realm. To address this challenge, we propose a new AI-based solution that identifies state-sponsored troll accounts by analyzing their sharing activity sequences, or trajectories, through a two-step process. First, we classify accounts' trajectories using an LSTM-based classifier as belonging to either a state-sponsored troll or an organic, legitimate user. Second, we utilize the classified trajectories to compute a metric, named ``Troll Score'', to quantify the extent to which an account behaves like a state-sponsored troll. To evaluate our approach, we examine the Russian interference campaign during the 2016 U.S. Presidential election. The results of our experiments show that our method can identify account trajectories with an AUC close to 99% and accurately classify Russian trolls and organic users with an AUC of 91%. Additionally, we assessed the generalizability of our solution to different influence campaigns originating from various countries and found promising results that will guide future research.