The detection of hate speech online is a difficult topic that has drawn a lot of interest recently from the natural language processing field. Even with a significant improvement in performance, there are still several obstacles to overcome, one of which is how to include contextual data into automated hate speech detection systems. In this research, it is analyzed whether a hateful Facebook comment is intended toward migrants or someone else, i.e., identifying the target of hate speech in Dutch social media. The scholars examine the impact of various context elements on performance when we incorporate them into BERTje, a Dutch transformer-based pre-trained language model. The pertinent conversational context is manually annotated. In the end, it is demonstrated that adding pertinent contextual information can greatly enhance the model’s performance.
https://aclanthology.org/2022.trac-1.5