Enhancing Online Safety: Natural Language Processing Based Multi-Label Cyberbullying Classification in Bangla

Cyberbullying can have devastating consequences on its victims, leading to emotional distress, psychological harm, and even suicidal thoughts. Numerous research studies have been conducted on the detection and classification of cyberbullying in high-resource languages like English, but there is a significant scarcity of research in low-resource languages such as Bangla. The lack of research may leave vulnerable communities without effective tools to combat this growing online threat, potentially increasing the harm caused by cyberbullying in these regions. Therefore, it is imperative to conduct additional research in low-resource languages to establish a secure online environment for these regions. This paper provides a comprehensive study on Bangla text classification, focusing primarily on cyberbullying detection, where we explored various classifiers, with fastText Pretrained Models emerging as top performers, achieving an accuracy of 84% and 0.84 as weighted F1-score. Our approach excels in advancing cyberbullying detection in the Bangla language and surpasses previous accuracy benchmarks, offering valuable insights and tools for online safety in this context.