Despite increasing efforts to promote biodiversity protection, a persistent gap remains between policy measures and public expectations. Aligning biodiversity strategies with citizens' concerns is crucial to fostering support and ensuring effective implementation. Building on the environmental governance and ecosystem services literature, this study leverages Large Language Models for embedding-based topic modeling to analyze a large-scale public consultation involving 80,000 participants, 6000 proposals, and 1.8 million votes in France, identifying eight key biodiversity-related themes: Management of natural and urban spaces, Resource management and pollution control, Wildlife protection and species management, Environmental education and awareness, Agriculture and sustainable food practices, Environmental policy and regulatory measures, Sustainable mobility, energy, and marine conservation, and Emerging and niche concerns. Findings also reveal significant polarization in areas such as hunting regulations, taxation of polluting products, and energy transition, while topics like environmental education and urban greening enjoy strong consensus. These insights highlight the need for differentiated strategies: prioritizing immediate action in consensus areas while fostering dialogue and adaptive policymaking in highly polarized domains. Methodologically, this study demonstrates how Large Language Models can efficiently process and analyze large-scale citizen consultation datasets, offering a scalable approach to capturing public preferences and societal debates in a policy context. • Citizen consultation data reveal contrasting priorities on biodiversity governance. • Regulating services triggers more polarization than cultural or provisioning services. • The polarization index identifies deep divides on hunting, energy, and dietary choices. • LLM-based clustering uncovers public themes from large-scale participatory data. • The scalable method enables thematic and polarization analysis in public consultations.
Trelohan et al. (Sat,) studied this question.