Comparação de tópicos nos discursos dos deputados federais nos anos de 2014 e 2022 com métodos de aprendizado de máquinas

Abstract

The Chamber of Deputies of Brazil is a Brazilian political and institutional body at the federal level, and is one of the components, along with the Federal Senate, of the Legislative Branch. Brazil's Chamber of Deputies is where the laws that govern Brazil are drafted. To this end, speeches are made in this space, where the deputies discuss the bills they and their various peers have created. The speeches therefore have an enormous wealth of data to be used, data that can be widely used by the population and scientists from various fields to understand the representativeness and issues discussed within this space. This data is then made available to the public free of charge and openly, but in a raw form. There is then room for scientific and technological exploration of methods, tools, analyses and methodologies that bring relevant information from such data. This is very important for the general population to better understand what is happening within the Brazilian democratic process under the rule of law. Immersed in this context is the emergence of Machine Learning methods, which help in the modeling and analysis of such problems. In this scenario, this study implements a \textit{pipeline} for the analysis of topics, or terms that appear together, in speeches by Federal Deputies and the attempt to find similar topics from a socio-political point of view between different periods of Brazilian politics using Machine Learning methods. This makes it easier to analyze changes in the Brazilian political scene, especially by observing the different Brazilian political actors over time. The proposed \textit{pipeline} is a tool for social scientists to easily access and analyze such data. The \textit{pipeline} proposed here can then be divided into 3 parts: the collection of the previously raw speeches and their organization in a structured and easily accessible format, the extraction of topics from the speeches of the Federal Deputies using the \textit{Propagation on Bipartite Graphs} algorithm and finally a reduced manual analysis of the topics collected in the previous stage in order to find topics that are similar in comparison with those found in another time period. This study proposes three parties as a case study: the Workers' Party, the Brazilian Democratic Movement and the Liberal Party. For the former, it is possible to see a return to the need to defend rights that were once considered established and fixed, such as the demarcation of indigenous lands. The Liberal Party, for its part, has undergone a major change in its discourse, moving the speeches to the right of the brazilian political scenery.