Intelligent Cyberbullying Detection by CNN-BiGRU Using Word2Vec and GloVe Word Embeddings

Abstract

In this paper, we propose an intelligent cyberbullying detection model using deep learning and word embeddings. Leveraging the 2024 Ejaz-Choudhury-Razi dataset, our approach combines Convolutional Neural Networks (CNN) for local pattern extraction and Bidirectional Gated Recurrent Units (BiGRU) to catch temporal features. Word2Vec and GloVe embeddings catch word semantics. The two proposed hybrid models achieve over 84% accuracy, precision, recall, and F1-score, with an AUC exceeding 91% and specificity above 82%, demonstrating its effectiveness in detecting cyberbullying on social media.