Mapping Toxic Comments Across Demographics: A Dataset from German Public Broadcasting

Work by: Jan Fillies, Michael Peter Hoffmann, Rebecca Reichel, Roman Salzwedel, Sven Bodemer, Adrian Paschke Accepted at EMNLP 2025 This dataset introduces the first large-scale German collection of social media comments annotated for toxicity and enriched with platform-provided age estimates. Developed in collaboration with funk (ARD/ZDF), it contains 3,024 human-annotated and 30,024 LLM-annotated anonymized comments from Instagram, TikTok, and YouTube. Annotations cover categories such as insults, disinformation, and criticism of broadcasting fees, enabling fine-grained demographic analysis of toxic speech patterns. The resource highlights age- and platform-specific differences in online toxicity and supports the development of equitable, age-aware moderation systems. Access: The dataset is restricted to verified researchers with ongoing, validated projects in the domain. Access requires entering into a separate legal agreement with funk and is granted solely for research with demonstrated scientific value; student work and non-validated projects are not eligible.