Relacionamento dos políticos brasileiros: um estudo envolvendo Ciência de Redes e Aprendizado de Máquina

Abstract

Brazilian politics is characterized by high party fragmentation and often complex relationships among legislators. With more than twenty parties represented in the Chamber of Deputies across different legislatures, understanding how political alignments and disputes evolve over time presents a significant challenge for researchers and analysts. In this context, network science techniques offer effective tools to represent and analyze these relationships in a structured and systematic manner. This study aims to analyze the Brazilian political scenario by examining the relationships among federal deputies through complex network analysis. Voting data from 2005 to 2023 were analyzed, considering both party affiliation and ideological alignment of the legislators. The first objective was to propose a complex network structure capable of representing political interactions in Congress. Based on this network, classical network metrics (degree, betweenness centrality, and clustering coefficient) were applied, alongside visual analyses over time. We aim to compare a central objective was to compare different grouping methodologies to identify affinity groups: the Louvain community detection algorithm, widely used in network science, and the hierarchical Ward clustering method, based on machine learning techniques. The comparison between these approaches allowed for the evaluation of their ability to capture relevant political groupings. The results show that, regardless of the ruling government, there is a tendency for the formation of majority government support blocks, especially at the beginning of presidential terms, with a gradual reduction in cohesion over time. A stronger ideological consistency was observed on the left, while other groups displayed greater flexibility in shifting between government and opposition blocks. Despite the high number of political parties represented in Congress, the actual number of groupings identified in the networks is much smaller, suggesting that formal party fragmentation does not reflect the true structure of political alignment. The integrated use of different methodologies provided a more comprehensive and detailed understanding of Brazil’s political dynamics and offers a foundation for future studies.