Cyberbullying Detection using Deep Learning Techniques on Bangla Facebook Comments

Bullying someone online is known as Cy-berbullying. As social media and the internet have grown in popularity, it has taken a very negative turn. People use the Internet's services to aggressively attack others while hiding behind a screen. It appears in various ways and is typically presented as text on most social networks. Cyberbullying frequently causes severe mental and physical distress, especially for women and children, and it even sometimes causes victims to attempt suicide. Because of the serious social consequences, online harassment draws attention. Therefore, researchers have begun to focus on identifying cyberbullies on social networks, particularly in the Bangla language. However, for accuracy and implementation, there is always room for improvement. This paper conducts an analysis using two deep learning algorithms, namely, Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU), to identify cyberbullies in Facebook comments written in Bangla. A total of 8736 comments were collected, and 7072 of them were retained for the final analysis after being filtered out due to irrelevancy. In the study, we used data preprocessing steps such as text cleaning, the removal of punctuation and special characters, tok-enization, the removal of stop words, and later joining the stemmed words. Cleaned textual data was fed to the deep learning models for prediction. Our research showed that, with an accuracy of 83.55 %, the GRU over LSTM model produced the best results on the current dataset.