The proliferation of social media platforms has intensified the prevalence of cyberbullying, posing significant threats to users' mental health and digital well-being. Conventional detection approaches relying on keyword matching and rule-based filters fail to capture the contextual and semantic nuances of harmful language. This paper proposes an integrated framework that combines sentiment analysis with deep learning architectures to automate the identification of cyberbullying content in text. The methodology begins with multi-platform data collection, followed by preprocessing steps including tokenization, stop-word removal, and normalization. Sentiment orientation is subsequently determined through polarity classification. Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNN) are then employed to learn latent linguistic patterns. Experimental evaluation on labeled datasets demonstrates that the hybrid approach achieves an accuracy of 85–90%, outperforming traditional machine learning baselines across precision, recall, and F1-score metrics. The scalable, modular architecture supports real-time monitoring and holds strong potential for integration into social media moderation pipelines. This work contributes a transparent and ethically grounded solution for early detection and mitigation of online harassment.
Deepthi et al. (Mon,) studied this question.