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

2025-08-26

AI Summary

Existing toxic speech datasets largely lack demographic fine-grained annotations—particularly age—hindering intergenerational analysis of linguistic behavior. To address this, we introduce the first large-scale, age-annotated German toxic comment dataset, drawn from Instagram, TikTok, and YouTube. Our methodology innovatively integrates platform-provided age estimates with human–LLM collaborative annotation (3,000+ expert human labels and 30,000+ LLM-assisted labels), covering categories including insults, misinformation, and critiques related to public broadcasting fees. Analyses reveal significant generational patterns: younger users exhibit more emotionally charged language, whereas middle-aged and older users disproportionately propagate misinformation and derogatory content. With a toxicity annotation rate of 16.7%, this dataset establishes a critical benchmark for modeling linguistic variation, developing fairness-aware content moderation systems, and enabling age-sensitive platform governance.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Application Domains: Humanities & Computational Social Science
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

A lack of demographic context in existing toxic speech datasets limits our understanding of how different age groups communicate online. In collaboration with funk, a German public service content network, this research introduces the first large-scale German dataset annotated for toxicity and enriched with platform-provided age estimates. The dataset includes 3,024 human-annotated and 30,024 LLM-annotated anonymized comments from Instagram, TikTok, and YouTube. To ensure relevance, comments were consolidated using predefined toxic keywords, resulting in 16.7% labeled as problematic. The annotation pipeline combined human expertise with state-of-the-art language models, identifying key categories such as insults, disinformation, and criticism of broadcasting fees. The dataset reveals age-based differences in toxic speech patterns, with younger users favoring expressive language and older users more often engaging in disinformation and devaluation. This resource provides new opportunities for studying linguistic variation across demographics and supports the development of more equitable and age-aware content moderation systems.

Problem

• Lack demographic context toxic speech datasets
• Need study age-based differences online communication
• Develop equitable age-aware content moderation systems

Innovation

• First large-scale German dataset with toxicity and age annotations
• Combined human expertise with state-of-the-art language models
• Revealed age-based differences in toxic speech patterns