This study examines EU politicization in Slovakia through large-scale computational analysis of parliamentary
discourse spanning 1994-2023. We argue for privileging stance detection over sentiment analysis when studying
EU contestation, as these dimensions capture theoretically distinct aspects of political positioning. Employing
large language models to classify EU stance and sentiment in approximately 30,000 speeches from the National
Council of the Slovak Republic, we validate this approach through systematic comparison against an expert-
coded dataset of over 600 parliamentary speeches annotated using custom guidelines specifically developed
for EU stance analysis. Our validation exercise compares multiple LLM architectures, including proprietary and
open-source models, to assess whether computationally efficient alternatives can reliably estimate EU stance at
scale. These findings advance methodological approaches to automated political text analysis while
demonstrating the feasibility of leveraging cost-effective LLMs for large-scale parliamentary research, with
implications for understanding EU politicization dynamics in Central and Eastern Europe.
