Multilingual Hate Speech Detection

— Due to recent reports of social media abuse, social media companies have been urged to address the issue of hate speech on social media. Advances in artificial intelligence and natural language processing have enabled the automation of hate speech detection. Despite that, challenges in the field of hate speech detection remain, as hate speech is highly context-dependent. This paper highlights the challenges of hate speech detection in multilingual communities and a solution for these challenges. This study adopts a hyperparameters fine-tuning approach on the pretrained BERT model for the development of hate speech detection models in both the monolingual and multilingual scenarios. The findings of the research have revealed that the multilingual hate speech detection approximates or exceeds the performance of baseline monolingual hate speech detection models, achieving excellent performance on the English test data (Accuracy = 0.931, Precision = 0.877, Recall = 0.921, F-1 = 0.899) and the Malay test data (Accuracy = 0.872, Precision = 0.874, Recall = 0.868, F-1 = 0.871). The multilingual hate speech detection models can be applied to multilingual communities where members of the community use different languages interchangeably