The Role of Context in Detecting the Target of Hate Speech (ACL Anthology)

The natural language processing community has recently focused a lot of emphasis on the intrinsically difficult job of detecting hate speech online. Even while performance has significantly improved, there are still many obstacles to overcome, such as incorporating contextual data into automated hate speech detection systems. The goal of this article is to identify the target of hate speech on Dutch social media, namely whether a hostile Facebook comment is intended at migrants or someone else. We manually annotate pertinent conversational context and examine how various context elements affect performance when included into BERTje, a Dutch transformer-based pre-trained language model. Incorporating pertinent contextual information can greatly enhance the model’s performance.

https://aclanthology.org/2022.trac-1.5