Detecting Adversarial BENDs in the Information Environment

Our objective is to assess the feasibility of detecting adversarial information campaigns by using a blend of stance assessment, and bot/troll/cyborg/meme detection techniques for automated classification of BEND actors and maneuvers. BEND includes 16 social media information maneuvers used in information campaigns for influencing topic groups through content or social network manipulation. We propose a three step early detection approach. Step-1: identify and classify the actors by type – bot, troll, cyborg, news agency, satire site, other. Step-2: classify the actions of those actors (who they follow, retweet, mention, quote, messages sent, sites linked to, memes sent, and content into information maneuver types = i.e., the BEND maneuvers. Step-3: identify signs of collaboration and orchestration among actors and sources on an information campaign. Unlike traditional approaches to information campaigns, we consider: 1) both content and social network manipulation, and 2) sentiment and stance. Unlike traditional approaches to adversarial behavior that consider only automated bots – we consider multiple types of actors i.e. bots, cyborgs, trolls, news, false news, and the relations of those actors to each other and sources. Baseline testing and demonstration will use data focused on Russian and Chinese information campaigns around NATO, Sweden and the Philippines.