Multi-Class Label Classification of Extremist Tweets

Abstract

Extremists use tweets for group strengthening, propaganda, brainwashing and fundraising by reaching people’s mind. A tweet is a post of utmost 280 characters on Tweeter, a popular micro blogging service. Our objective is to identify extremist affiliation based on opinions expressed on tweets, in order to prevent brainwashing on public and trace terrorist activities. To overcome these issues, sentiment analysis with an automated process of understanding an opinion on given subject from written or spoken language is adapted. Traditional methods of filtering are not scalable for classifying extremist and non-extremist tweets. Overcoming conventional approach, machine learning based classification system applied to this problem, was still limited to unrigorous and vast categorization of tweets into positive and negative. Another restriction was the negligence of overall dependency related to sentences. This paper, therefore aims at distinguishing tweets as extremist and non-extremist and also investigating other types of extremism by employing Deep Learning and machine learning methods.