Us against the World: Detection of Radical Language in Online Platforms

Abstract

In this paper, we have investigated if we can detect radical comments in an online social network. We used comments from 6 subreddits, 3 of which are considered radical and 3 non-radical. Using various structural features of the texts in the comments, we were able to obtain an F1-score of 91% when using SVM with a linear kernel and a precision of almost 98% when using Random Forest.