Mapping (A)Ideology: A Taxonomy of European Parties Using Generative LLMs as Zero-Shot Learners

• DOI: 10.1017/pan.2025.7
• Publication Date: 2025-04-14T07:58:31Z
• Authors: Riccardo Di Leo; Chen Zeng; Elias Dinas; Reda Tamtam
• Keywords: Zero (linguistics)

Abstract We perform the first mapping of ideological positions European parties using generative Artificial Intelligence (AI) as a “zero-shot” learner. ask OpenAI’s Generative Pre-trained Transformer 3.5 (GPT-3.5) to identify more “right-wing” option across all possible duplets at given point in time, solely based on their names and country origin, combine this information via Bradley–Terry model create an ranking. A cross-validation employing widely-used expert-, manifesto- poll-based estimates reveals that scores produced by Large Language Models (LLMs) closely map those obtained through expert-based evaluation, i.e. , CHES. Given high cost scaling trained coders, scarcity expert data before 1990s, our finding AI produces comparable quality CHES supports its usage political science grounds replicability, agility, affordability.