Deep Learning Approach for Multi-Label Detection of Abusive Bangla Social Media Comments

This paper presents a multilabel classifi-cation model to detect and classify various abuse or bullying content from Bangla social media comments. We utilized an LSTM-based deep learning model to train a classification system using a dataset of 12,557 multilabel comments. The comments were categorized into five categories: Bully, Sexual, Religious, Threat, and Spam. The model achieved an overall accuracy of 93.64% and an AUe of 0.9802. These metrics highlight the model's effectiveness in identifying and classifying abusive content. The results show that the model is robust and dependable, making it a valuable tool for addressing and mitigating various forms of Bangla text-based abuse.