AI-Based Cyberbullying Detection Among College Students on Social Media Using Natural Language Processing

Abstract

As digital connectivity deepens its foothold in academic life, social media has become a dominant channel through which college students communicate, collaborate, and consume information. Yet this ubiquity carries a darker dimension: the unchecked spread of cyberbullying. Manually reviewing the enormous torrent of daily online interactions for abusive content is neither scalable nor sustainable. This paper introduces an automated content moderation framework that draws on Natural Language Processing (NLP) and machine learning to flag cyberbullying incidents at scale.At its core, the framework applies TF-IDF vectorisation to convert raw messages into weighted feature vectors, then feeds these into a dual-component classifier pairing Random Forest with a domain-adapted BERT model. Alongside binary predictions, the platform generates enriched diagnostic outputs — covering sentiment polarity, message length distribution, and term frequency profiles — to deepen understanding of harassment dynamics.Evaluation results confirm that the hybrid architecture achieves strong detection performance. The proposed framework offers a scalable pathway toward proactively moderating harmful content on digital platforms and cultivating healthier online communities.