Tracking Online Hate During the COVID-19 Pandemic in the Netherlands: A Hybrid Machine Learning and Keyword-Based Analysis of Dutch Social Media

Hate speech on social media platforms like X has emerged as a growing concern, particularly during times of social stress such as the COVID-19 pandemic. While the Netherlands has seen a rise in online hostility, especially on Dutch-language platforms, studies focusing on hate speech targets and dynamics remain limited. This paper presents a large-scale analysis of Dutch social media during 2020—the first pandemic year—using a hybrid approach that combines sentence-based heuristics, supervised machine learning, and keyword-driven target identification. Our best-performing classifier, an SVM trained on a Dutch hate speech corpus with Bag-of-Words features, achieved an F1-score of 72.6% and identified 158,677 hate speech posts from a corpus of nearly four million tweets. Of these, 52,959 were further classified into specific target groups. The results show that political beliefs were the dominant target (56.2%), followed by religion, nationality, and ethnicity, with sexual orientation and gender accounting for much smaller shares. Compared to prior works our study has a more granular target labeling while achieving a good classification performance. These findings highlight the prevalence and patterns of hate speech in the Dutch Twittersphere and offer a replicable method for low-resource language settings.