Extremist text detection in social web

Abstract

Online or cyber extremism is one of the critical problem for the security of Russia and other countries as social web is widely used for radical activity and propaganda. This paper considers the problem of extremist text detection in Russian social media. We propose models and methods for identification of extremist text in Russian, which apply deep linguistic parsing and statistical processing of texts. We also present the dataset of terrorist, religious hate, racism and other radical texts in Russian and results of experiments on this dataset. It was shown, that low-dimensional psycholinguistic and semantic features of texts allow detecting extremist texts with quite good performance while lexical features allow recognizing topics of the detected extremist texts.