Enhanced Cyberbullying Detection System In Social Media Forensic Using Neural Network
Abstract
Cyberbullying has become a growing concern due to the increasing use of social media, leading to psychological distress and social consequences for victims. Traditional cyberbullying detection methods often fail to accurately classify offensive content due to the informal, ambiguous, and context-dependent nature of online communication. This research presents an enhanced cyberbullying detection system utilizing DistilBERT, a lightweight transformer-based model, to improve classification accuracy while maintaining computational efficiency. The system preprocesses and analyzes text data, leveraging deep learning techniques to detect various forms of cyberbullying, including hate speech and harassment. A Flask-based web application is integrated to enable real-time detection, providing an interactive platform for users to analyze and report cyberbullying instances. The model is trained on a large dataset of tweets and achieves superior accuracy compared to existing approaches, including BiLSTM and traditional BERT models. The experimental results demonstrate that our system effectively reduces misclassification rates and enhances the reliability of cyberbullying detection, ensuring a more robust and scalable solution for online safety. Furthermore, the system incorporates uncertainty estimation techniques to improve the interpretability and confidence of predictions, making it more reliable for large-scale deployment in social media forensics and moderation systems.
Keywords
Cyberbullying Detection; DistilBERT; Deep Learning; Social Media Forensics; Machine Learning.