Detection of Hate Speech in Turkish Social Media Posts with BERT-Base Model

With the development of the Internet, the amount of data expressed by users worldwide has increased. Detecting hate speech in social media, preventing harmful language and promoting safe online communities is an important task. Bidirectional Encoder Representations from Transformers (BERT) is a popular language model that performs well in natural language processing tasks. In this study, the BERT-Base model is used to detect hate speech in the comments of Twitter users in Turkey. It was trained with a dataset of examples labeled as harmful and harmless and achieved 92.53% test accuracy. The developed model was presented to users in a live environment.