Robertanet: Enhancedroberta Transformer Based Model for Cyber Bullying Detection with Glove Features

Abstract

Social media has become a vital platform for communication, but it also encourages harmful behaviours such as cyberbullying, trolling, and hate speech. Manual moderation is slow and expensive, making automatic detection essential. This work presents an enhanced RoBERTa transformer model combined with GloVe word embeddings to detect cyberbullying in tweets. The system is tested against various machine learning and deep learning models and achieves about 95% accuracy, along with high precision, recall, and F1 scores. Cross-validation further confirms its reliability. The results show that advanced transformer models supported by effective feature representation can provide a strong solution for detecting cyberbullying on social media. Beyond academic evaluation, such a system can be valuable in protecting vulnerable users, especially teenagers, from psychological harm. Moreover, the approach has potential to be adapted across multiple social platforms, making it a promising step towards safer digital communities.