Beyond Words: Advancing AI-Based Cyberbullying Detection - A Comprehensive and Practical Approach
Sakirulai Olufemi Isiaq, Ademayowa Adeogun, Peter Phiri · 2025
The pervasive growth of social media has facilitated unprecedented connectivity but also intensified the prevalence of cyberbullying, a digital form of harassment with severe psychological consequences. Unlike traditional bullying, cyberbullying transcends spatial and temporal boundaries, often eluding detection by conventional text-based models. This study proposes an enhanced detection system that fuses textual and emoji-based sentiment analysis, leveraging the expressive function of emojis to capture nuanced emotional and contextual cues. Employing natural language processing (NLP), supervised machine learning, and deep learning algorithms, the model processes multimodal inputs to improve classification performance. Addressing gaps in benchmark studies, the research introduces a diverse, annotated dataset, supports multi-class classification, and applies a comprehensive evaluation framework, including precision, recall, F1-score, and ROC AUC. The final model is deployed via an intuitive interface, ensuring practical applicability. Results demonstrate improved detection accuracy, reinforcing the potential of AI in moderating online interactions. While promising, challenges remain, including cross-platform generalisation, multilingual support, and bias mitigation. This work contributes to the development of intelligent, scalable interventions aimed at fostering safer digital environments and advancing human environment interactions through ethical AI.