Multilingual Cyberbullying Classification for Social Platforms

The prevalence of cyberbullying is increasing on social media platforms, impacting individuals from many cultural backgrounds and linguistic communities. This surge is particularly evident in Asian nations. This paper explores the development of a sophisticated multilingual cyberbullying categorization system specifically designed for various social networks. The research uses modern natural language processing methods to address the worldwide issue of online harassment. It carefully examines linguistic trends in different languages. The novelty of this paper lies in its analysis of multiple languages which enables the method to be more accurate, versatile and culturally aware of different cyberbullying instances. Also, the utilization of diverse models enhances its efficacy. Specifically, the BERT method demonstrates a remarkable accuracy of 0.95 in the English dataset, whilst the RNN approach gets a noteworthy accuracy of 0.75 in the Bangla dataset. The RF algorithm demonstrates a strong accuracy of 0.94 in the Arabic dataset, whereas the LR method shows a more subtle accuracy of 0.60 in the Hinglish dataset. The language-specific indicators highlight the model's exceptional adaptability, underscoring its vital contribution to promoting a more secure digital environment. The need of using a multilingual strategy becomes apparent, as it effectively deals with the many language subtleties that underpin cyberbullying, therefore making substantial progress in promoting inclusion and flexibility in combating online harassment across various social platforms.