HICC : a dataset for German hate speech in conversational context

We present HICC: A dataset of German Hate speech comments In Conversational Context. The dataset comprises the target hate speech comments to be detected, the root post that starts the conversation, and all direct replies to the root that were published before the target. We especially focus on comments that are difficult to identify as hate speech in isolation, but need context for classification. We show that LLM-based classification can successfully incorporate conversational context to identify such comments that do not explicitly contain hateful language. Our dataset is gathered from the social media platform X (formerly Twitter), manually annotated, and made freely available for research.