Deep Learning for Cyberbullying Detection: A GloVe-based Comparative Analysis of CNN and LSTM Models

ICLR 2026 Conference Submission21190 Authors

Keywords: GLOVE-CNN-Cyberbulling detection -LSTM

Abstract: Social networks were created to fulfill human needs, driven by people’s eagerness to learn new things and stay informed about global events. To detect cyberbullying on social media, this research compares two deep learning architectures: GloVe+CNN and GloVe+LSTM. Textual data were represented using pre-trained GloVe embeddings, and CNN and LSTM were employed as classification layers to identify sequential and local patterns, respectively. Using a multiclass cyberbullying dataset, the models were trained and evaluated. The results show that although both architectures perform well, GloVe+LSTM outperforms CNN in terms of F1-score and recall, indicating better contextual understanding. The experimental results demonstrate the superiority of LSTM, in terms of accuracy.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 21190