Populismo em palavras: inteligência artificial e análise de discursos presidenciais no Brasil (2003–2022)

Abstract

This study focuses on the development of NLP tools for text analysis. Using populism as a case study, the research evaluates the effectiveness of classical models, deep learning, and generative AI (Gemini) in classifying Brazilian presidential speeches from 2003 to 2022. The results show that while traditional models have limited performance, Gemini proves to be a superior and effective screening tool. The analysis of its "errors" reveals the nuances of populist rhetoric, such as Lula's conciliatory populism and Bolsonaro's implicit performance, which are not captured by minimalist definitions. The study concludes that the best approach for political discourse analysis combines the scale of computational tools with the depth of human qualitative analysis, highlighting the value of hybrid methods in data science.