Analyzing Social Bots and Their Coordination During Natural Disasters

Abstract

Social bots help automate many sociotechnical behaviors such as tweeting/retweeting a message, ‘liking’ a tweet, following users, and coordinate or even compete with other bots. Social bots exist as advertising bots, entertainment bots, spam bots, and influence bots. In this research, we focus on influence bots, i.e., automated Twitter accounts that attempt to affect or influence behaviors of others. Some of these bots operate independently and autonomously for years without getting noticed or suspended. Furthermore, some of the more advanced influence social bots exhibit highly sophisticated coordination and communication patterns with complex organizational structures. This study aims to explore the role of Twitter social bots during the 2017 natural disasters and evaluate their coordination strategies for disseminating information. We collected data from Twitter during Hurricane Harvey, Hurricane Irma, Hurricane Maria, and Mexico Earthquake that occurred in 2017. This resulted in a total of over 1.2 million tweets generated by nearly 800,000 Twitter accounts. Social bots were detected in the data. Social networks of top bot and top non-bot accounts were compared to examine characteristic differences in their networks. Bot networks were further examined to identify coordination patterns. Hashtag analysis of the tweets shared by bots further helped in identifying hoaxes (such as, ‘shark swimming on freeway’) and non-relevant narratives (black lives matter, DACA, anti-Semitic narratives, Kim Jong-Un, nuclear test, etc.) that were disseminated by bots in several languages, such as French, Spanish, Arabic, Japanese, Korean, etc., besides English.