A Real-Time Cyberbullying Monitoring and Intervention Framework for Social Media Platforms

Abstract

: Cyberbullying has emerged as a significant problem on social media platforms, affecting millions of users through toxic comments, hate speech, and online harassment. To address this challenge, we developed an intelligent system capable of automatically classifying toxic comments using Artificial Intelligence and Natural Language Processing (NLP). In this project, we implemented three models Logistic Regression, LSTM, and BERT to perform multi-label classification of comments into six categories: toxic, severe toxic, obscene, threat, insult, and identity hate. We utilized the Jigsaw Toxic Comment Classification dataset from Kaggle for model training and evaluation. Our results showed that Logistic Regression provided a strong baseline with an accuracy of 89%, LSTM improved performance with higher recall and better sequence modeling, and BERT achieved the highest accuracy of 94% and overall best performance, although it required longer training time. The study highlights the effectiveness of conventional and deep learning techniques for toxic comment identification and provides comprehensive analysis of accuracy, recall, and computational cost. The outcome of this project is a practical multi-label classification framework that can serve as a foundation for future real-time cyberbullying detection systems.