Abstract:
Suspended accounts are high-risk accounts that violate the rules of a social network. These accounts contain spam, offensive and explicit language, among others, and are incredibly variable in terms of textual content. In this work, we perform a detailed linguistic and statistical analysis into the textual information of suspended accounts and show how insights from our study significantly improve a deep-learning-based detection framework. Moreover, we investigate the utility of advanced topic modeling for the automatic creation of word lists that can discriminate suspended from regular accounts. Since early detection of these high-risk accounts is crucial, we evaluate multiple state-of-the-art classification models along the temporal dimension by measuring the minimum amount of textual signal needed to perform reliable predictions. Further, we show that the best performing models are able to detect suspended accounts earlier than the social media platform.
