Full/Regular Research Paper submission to (CSCI-RTCW): Multi Class Classification of Online Radicalization Using Transformer Models

Online Radicalization is a major security threat to a nation and has the power to influence young minds through online blogs and articles present on social media. It can occur in various forms such as political, social, criminal radicalization etc depending upon the intention and target of the propagator. Each type of radicalization aims to harm a different section of society and thus, requires a different type of treatment and mitigation plan. In our paper we have proposed the use of transformer based models such as BERT etc. to identify the type of radicalization in online text. We have also presented a comparative analysis of several transformers based classifiers for multi class classification of radicalization on social media. The results show that DistilBERT outperforms the other transformer models and has achieved an accuracy of 96.3 percent in this text classification task. As per our knowledge, this is a first of its kind study where the type of radical behaviour in text is being detected.