With the rise of new social media platforms designed for teenagers and adolescents, the importance of content moderation supported by algorithms is more necessary than ever. State-of-the-art hate speech detection algorithms are increasingly challenged by the rapid and creative evolution of modern language. To better understand the online discourse and phenomena of German far-right extremism on contemporary platforms, this research presents the first German TikTok dataset, consisting of 10,586 comments collected from comment sections and annotated for far-right extremism and hate speech. An extensive and novel annotation scheme comprising of 32 labels was developed in collaboration with domain experts in online extremism, specifically tailored to the TikTok platform. Three trained annotators meticulously annotated the dataset, with 13.76% of the collected data annotated to be hateful. A quantitative analysis was conducted, examining the primary keywords emerging within hate speech classes, identifying label combinations and distributions, and a sentiment analysis was performed. The dataset reveals extensive hate directed toward German politicians, particularly members of the Green Party, as well as women and immigrants. This research contributes to the field by introducing a new annotation schema, providing a fully annotated dataset, and analyzing the annotations and language used.
