Revealing Online Threats Through Leveraging RNN Models for Advanced Cyberbullying Detection

Abstract

Researchers are increasingly focused on the challenge of cyberbullying detection, driven by its growing presence on social media platforms—a realm where individuals share thoughts, opinions, and aspects of their lives. Unfortunately, these platforms often become arenas for toxicity, with cyberbullying manifesting through posts that are offensive, violent, or intimidating. This phenomenon, transcending national boundaries, inflicts harm on victims, affecting all facets of their lives. In our study, we introduce a method for identifying hate speech utilizing Recurrent Neural Network (RNN) models, specifically applied to a dataset in the Darija dialect. We assessed the model's effectiveness using metrics such as accuracy, precision, recall, and F1-score, and employed confusion matrices and ROC-AUC curves for visual evaluation.