Rating-Based Cyberbullying Detection with Text, Emojis on Social Media

In the dynamic landscape of online interactions, cyberbullying has become pervasive, profoundly impacting user's digital well-being. Public figures, especially celebrities and influencers, face heightened vulnerability to online harassment, exacerbated by the post-pandemic surge in social media usage. To address this challenge, our research adopts a holistic approach to detect cyberbullying in text, considering both textual content and the nuanced expressions conveyed through emojis on social media platforms. We employed a diverse set of machine learning and deep learning models, including Support Vector Classifier, Logistic Regression, Random Forest, XGBoost, LSTM, Bi-LSTM, GRU, and Bi-GRU, to accurately classify non cyberbullying or cyberbullying text. Beyond classification, our study introduces an offensive rating system, assigning severity ratings on a 1-5 scale to identify cyberbullying instances. A critical aspect is the establishment of a threshold value which depends on user security and safety ethics of different social media platforms; texts surpassing this trigger an automatic recommendation to block the user, ensuring a proactive response to minimize harm. This recent contribution not only comprehensively addresses cyberbullying but also empowers society.