Detecting Cyberbullying through Sentiment Ratings of Text and Emojis on Social Media Platforms

With the growing use of digital platforms, online bullying has become a serious and widespread issue that significantly affects users’ mental and emotional well-being. Public figures such as influencers and celebrities face even greater vulnerability due to their high visibility and constant exposure on online networks, especially after the sharp rise in social media usage following the pandemic. To tackle this challenge, our work adopts a well-rounded strategy that examines both the text and the expressive cues conveyed by emojis to detect cyberbullying. We utilize an array of machine learning and deep learning models—namely, Support Vector Classifier, Logistic Regression, Random Forest, XGBoost, LSTM, Bi- LSTM, GRU, and Bi-GRU to classify comments as bullying or non-bullying. Furthermore, we introduce a severity-based scoring system that rates offensive text on a scale of 1 to 5. When a message crosses a predefined severity threshold— determined by the safety standards of each platform—an automated recommendation to block the user is triggered. This approach not only enables precise identification of harmful content but also provides a proactive mechanism to promote safer online interactions.