A Bi-LSTM Based Model for Social Media Extremism Detection in Context of Bangladesh
Sajib Saha, Zahidul Islam, Tanvirul Islam, Md. Monzur Morshed, Tithee Chakma Mama, Subhenur Latif, Md Walid Bin Jashim · 2025
In today’s modern world, digital technology has advanced social networking with revolutionary changes. People use some most popular platforms like Facebook, YouTube and some other social media platforms to engage with each other for business, information sharing and so on. Attacking someone on social media platforms through comments, in response to their speech, has become a common occurrence in Bangladesh as well as worldwide. Very little research has been done in this related field for Bengali NLP. In this study we primarily focus on classifying these comments into two categories: extremist and non-extremist. Comments were collected from two of the most popular platforms, YouTube and Facebook, used in Bangladesh to build our "BEC (Bangla Extremist Comments)" dataset. To classify the comments, a CNN + Bi-LSTM hybrid model was employed considering extremist comments detection systems using deep learning-based sentiment analysis techniques. The proposed model achieved an accuracy rate of 85% in identifying the extremist comments. This research opens opportunities for future researchers to take advantage of contributing and collaboration in this field.