Software robots, or simply bots, have often been regarded as harmless programs confined within the cyberspace. However, recent events in our society proved that they can have important effects on real life as well. Bots have in fact become one of the key tools for disseminating information through online social networks (OSNs), influencing their members and eventually changing their opinions. With a focus on classification, social bot detection has lately emerged as a major topic in OSN analysis; nevertheless more research is needed to enhance our understanding of such automated behaviors, particularly to unveil the characteristics that better differentiate legitimate accounts from bots. We argue that this demands for learning behavioral models that should be trained using a large and heterogeneous set of behavioral features, so to detect and characterize OSN accounts according to their status as bots. Within this view, in this work we push forward research on bot analysis by proposing a machine-learning framework for identifying and ranking OSN accounts based on their degree of bot relevance. Our framework exploits the most known existing methods on bot detection for enhanced feature extraction, and state-of-the-art learning-torank methods, using different optimization and evaluation criteria. Results obtained on Twitter data show the significance and effectiveness of our approach in detecting and ranking bot accounts.
Learning to rank social bots
Perna, Diego;Tagarelli, Andrea
2018-01-01
Abstract
Software robots, or simply bots, have often been regarded as harmless programs confined within the cyberspace. However, recent events in our society proved that they can have important effects on real life as well. Bots have in fact become one of the key tools for disseminating information through online social networks (OSNs), influencing their members and eventually changing their opinions. With a focus on classification, social bot detection has lately emerged as a major topic in OSN analysis; nevertheless more research is needed to enhance our understanding of such automated behaviors, particularly to unveil the characteristics that better differentiate legitimate accounts from bots. We argue that this demands for learning behavioral models that should be trained using a large and heterogeneous set of behavioral features, so to detect and characterize OSN accounts according to their status as bots. Within this view, in this work we push forward research on bot analysis by proposing a machine-learning framework for identifying and ranking OSN accounts based on their degree of bot relevance. Our framework exploits the most known existing methods on bot detection for enhanced feature extraction, and state-of-the-art learning-torank methods, using different optimization and evaluation criteria. Results obtained on Twitter data show the significance and effectiveness of our approach in detecting and ranking bot accounts.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.