Leveraging deep learning to combat cyberbullying on social media.

The rise of social media, while revolutionizing the way one communicates, has also produced a rampant issue of cyberbullying with severe aftermaths, including emotional harm, social exclusion, and even suicide in some cases. Efforts have been ongoing to stop this, yet cyberbullying continues its growth through anonymity and ready accessibility provided by online resources. Deep learning techniques appeared to be the most promising, as they allow for recognizing cyberbullying through analyzing textual content and images and various user behaviors, yielding more accurate and comprehensive detection procedures than the traditional ones applied. Among the models using transformer-based and recurrent neural networks, there are perfect models for detecting harassment and threats, impersonation, and cyberbullying. Multimodal data, including emojis and sentiment analysis, have been added to improve detection accuracy as they capture subtle nuances in online communication. However, challenges persist in terms of difficulties in annotating data, slang, and context-dependent interpretations of cyberbullying. The anonymity of users further complicates the identification of perpetrators. This short communication discusses the potential of deep learning in tackling cyberbullying, underlining recent progress, challenges, and the need for further research and collaboration between academia, industry, and policymakers to produce more effective, ethical, and culturally sensitive solutions for cyberbullying detection worldwide.