A binary text classifier that labels Dutch parliamentary motion text as climate-relevant or not. The positive class aggregates the climate, energy, and agriculture policy domains. The model is fine-tuned from RobBERT v2 (Delobelle et al., 2020), a Dutch RoBERTa-based language model, on a labelled sample of Dutch parliamentary text drawn from the ParlaMint corpus.

It was developed to isolate climate-relevant parliamentary motions from a larger collection of Dutch parliamentary documents, as the first stage of a longitudinal study of climate framing in Dutch politics.

Evaluated on a held-out test set (n = 2,761):

The model is designed for filtering Dutch parliamentary motions and debate text for climate relevance. It is trained on the formal register of parliamentary documents (motions, debates), and performance on other domains such as news articles or social media is untested. The climate-relevant class deliberately aggregates climate, energy, and agriculture, so texts dealing only with energy or agriculture without an explicit climate dimension are also labelled relevant.

A labelled sample of Dutch parliamentary text from the ParlaMint corpus (Erjavec et al., 2025), with policy-topic labels following the Comparative Agendas Project taxonomy aggregated into a single climate-relevant category. Labels were iteratively refined through error analysis, excluding topically adjacent but non-climate categories such as fisheries, animal welfare, and earthquake damage.

Developed as part of the MSc thesis "From Rhetoric to Action: Climate Framing in Dutch Politics (2008-2025)", University of Groningen, 2026.