Student: Willian Pitter Cardoso Lima
Title: An Analysis of Political Parties Based on Speeches at the Brazilian National Congress
Advisors: Laura Silva de Assis (advisor) and Douglas de Oliveira Cardoso (co-advisor)
Committee: Laura Silva de Assis (Cefet/RJ), Douglas de Oliveira Cardoso (IPT), Eduardo Bezerra da silva (Cefet/RJ) e Rafael Lima de Carvalho (UFT)
Day/Time: December 05, 2023 / 10 a.m.
Abstract: Speeching is an intrinsic part of the work of congressmen, as they expose facts as well as their points of view and opinions on several subjects. This paper aims to analyze relations between congressmen according to the speeches given by members of the lower house of the national congress of Brazil. The period considered in this study comprises the mandate between 2011 and 2015. In order to accomplish this goal, the proposed methodology based on Natural Language Processing, Term Frequency–Inverse Document Frequency and Universal Sentence Encoder were used to assess pairwise relationships between congressmen which were then analyzed from the perspective of Complex Networks. In this work, to represent the problem under study, a complete graph was constructed in which each node represents a deputy, and the weights associated with the edges that connect these nodes represent the
similarities between their political positions. Node clustering was used to evaluate multiple speech-based measures of distance between each pair of congressmen, as well as the resulting cohesion of their political parties. Experimental results showed that one of the proposed measures, based on aggregating similarities between each pair of speeches, is superior to a previously established alternative of considering concatenations of these elements relative to each individual when targeting to group parliamentarians organically.