Cerberus: a cross-site social bot detection system based on deep learning

Abstract

Online social networks attract billions of active users and deeply influence people's lifestyles. However, as public social networks with low requirements for registration and joining, it is inevitable that social bots are able to easily register and do harmful things such as controlling public opinion and spreading inaccurate information for profit. Nevertheless, single-site social bot detection systems often rely on historical behavioral data to identify bots, and the detection occurred after the social bots have implemented their attacks. To identify social bots as early as possible, this paper proposes Cerberus, a cross-site system for detecting social bots in social networks, which solves the cold-start problem of user identification caused by insufficient user data on a single platform at an early stage and thus identifies social bots as early as possible. In this paper, the system is designed to identify whether a user on Twitter is a bot or not by leveraging the user's profile and text contents on his or her Medium account linked to Twitter. The results from our experiments show that the AUC score of the system can reach 0.7522, which outperforms others.

Keywords

• Online Social Networks;Social Bot Detection;Cross-Site Linking;deep learning;Cold-Start Users