A longitudinal dataset and analysis of Twitter ISIS users and propaganda

Although ISIS was driven out of its last territory in Syria and its physical caliphate collapsed in 2019, recent terrorist attacks and incidents in various areas, such as Central Africa, and recently designated terrorists by the Department of State signify the presence and growth of ISIS affiliates. Yet, once under scrutiny, their whereabouts have been less known and studied since then. In this paper, we aim to reopen this line of investigation to further our understanding of the recent ISIS online presence. We collect and analyze a longitudinal dataset of tweets from potential ISIS affiliates and an existing dataset of ISIS Twitter activities around 2015—when Twitter cracked down on many ISIS accounts. We build a user classifier to identify ISIS supporters on Twitter. We further investigate different types of content and topics used by known ISIS accounts and compare these attributes among ISIS and its potential affiliates. We also identify messages that gained abnormal attention and support and presumably were used to spread propaganda and influence and manipulate the online community. Finally, we build and evaluate an image classifier using the residual network pretrained image classification models to categorize the photos attached to the candidate propaganda tweets of ISIS and identify the major themes in those media files.