Cyberbullying Detection From Romanized Bangla Text Using Deep Learning Techniques

Abstract

People can freely express their ideas and opinions on a wide range of topics directly through a variety of social media outlets thanks to advancements in the internet and social media. In Bangladesh, people use both English and Romanized Bangla to write comments on social media sites like Facebook and Instagram. These remarks can be both positive and negative. Writing in our mother tongue simply stands out more than writing in other languages. Researchers have put a lot of effort into detecting cyberbullying in English and Bangla among other languages. In this thesis, I present my work on identifying offensive language from comments taken from celebrities on various social sites. What usually happens on social sites is that users post comments, some of which are good and some of which are offensive. The social site user follows all comments except those filtered by the poster's profile. I created a data set of about 1200 comments, including good comments and offensive comments. Where I have leveled the good and offensive comments. I have leveled good comments as 0 and offensive comments as 1. When celebrities post on social media my algorithm will check with the data set and users will try to share their comments and show only positive comments. In real-time, all abusive comments will come to spam and social media harassment will decrease.