Radicalization Detection Using Hybrid Deep Learning with Whale Optimization Technique

Abstract

Microblogging and Social media platforms like twitter, Facebook, etc. are very much popular among the youth. One can easily post any thought anytime using these platforms. Many times these posts belong to radical messages. These radical posts are one of the major social issues. This problem affects people everywhere in the world. This fosters a hostile, contentious, and discouraging atmosphere, which easily impacts the youth. Social media is the primary platform for radical people. They are using this as a weapon to spread their propaganda. It is important to quickly find and stop these radical messages on social media. In this article we proposed hybrid deep learning model DCLSNet using whale optimization technique for timely detection of radical message. We compared the performance of different baseline deep learning models with this model. This model outperforms than the baseline deep learning models. The F1-Score is 0.96 of DCLSNet. Further we used BERT, DistilBERT and RoBERTa transformer. BERT, RoBERTa and DistilBERT F1-Scre is 0.94, 0.95 and 0.92 respectively. These transformers have to be fine-tuned on the training data and then their performance is almost as good as DCLSNet in term of accuracy and F1-Score. But the complexities of these transformers are very higher than the proposed model. The proposed hybrid model is consuming low computing resources. It can be used by the administrators to detect the radicalization timely.