Análise de discursos parlamentares na Câmara dos Deputados

Abstract

This work aims to analyze speeches delivered by Brazilian federal deputies using Natural Language Processing techniques and topic modeling. More than 900,000 speeches from the Chamber of Deputies were collected, covering the period from 1950 to 2024, and an exploratory analysis of this data was conducted. However, the topic modeling analysis focuses exclusively on the year 2001, in order to validate the proposed technique. The analysis employs models widely used in the literature, such as Latent Dirichlet Allocation (LDA), as well as more recent models like BERTopic, to identify thematic patterns and the evolution of topics debated over time. To achieve this, data was automatically collected via web scraping, and a series of textual preprocessing steps were applied, such as lemmatization and stopword removal. The modeling enabled not only the identification of recurring topics but also the correlation between historical events and the content of the speeches. This approach highlights the potential of computational methods for analyzing large volumes of textual data in a political context.