International Journal of Scientific Research and Engineering Development

Cyberbullying has emerged as one of the most alarming consequences of widespread social media adoption, affecting millions of adolescents and young adults globally. Conventional moderation methods relying on manual review and keyword filtering are fundamentally inadequate to handle the volume, velocity, and linguistic diversity of harmful content generated on modern platforms. This paper presents GuardNet, an AI-powered real-time cyberbullying detection framework that integrates Natural Language Processing (NLP), deep learning-based text classification, and a multi-dimensional toxicity profiling engine to identify and categorize harmful content across social media platforms. The proposed system employs a fine-tuned BERT-based transformer model augmented with a 5-axis Harm Profiling Module, achieving a detection accuracy of 96.8% with a response latency of under 1.5 seconds per post. Upon detection of high-severity content, GuardNet autonomously notifies platform administrators, generates victim-support resources, and logs incidents for longitudinal behavioral analysis. Experimental evaluation on benchmark datasets including the Hate Speech and Offensive Language Dataset and the Cyberbullying Detection Dataset demonstrates superior performance over existing baseline models. GuardNet offers a scalable,privacypreserving moderation layer applicable across educational platforms, social networks, and online gaming communities, with significant potential to reduce the psychological harm caused by digital harassment.

A.Agesta Jenifer, Nitin R.S,"An AI-Powered Real-Time Cyberbullying Detection System on Social Media Using Natural Language Processing and Deep Learning" International Journal of Scientific Research and Engineering Development, V9(3): Page(1771-1777) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.