SANTOS, M.A.; http://lattes.cnpq.br/2022936009335392; SANTOS, Matheus Alves dos.
Abstract:
Strong and truly representative democracies can be built only when civil society can understand and monitor the political activities of its representatives. However, despite remarkable progress towards government transparency, people in Brazil tend not to be aware of parliamentary activities. This scenario emerges from multiple socio-cultural aspects, as well as from the structural complexity surrounding the Legislative Branch. Therefore, methods and tools that provide access to information for civil society and, more importantly, contribute to its understanding and usage of such information are essential. In this regard, Natural Language Processing techniques have been increasingly employed to analyze huge textual datasets surrounding political contexts, such as speeches and law proposals. In this study, we evaluated both the individual and the combined use of two state-of-the-art techniques for latent topic modeling and text-based ideal point estimation, applying them to characterize the speeches and political views of Brazilian parliamentarians. Specifically, we used the BERTopic and the Text-Based Ideal Point models to analyze the 55th and 56th Legislatures of the Brazilian Chamber of Deputies, spanning the period from 2015 to 2022. In this process, we also built and published an open database containing speech transcriptions from events held by this legislative house between 2003 and 2022. The evaluation of these techniques was quantitative and qualitative, considering metrics such as coherence and diversity of latent topics, but also the face validity and the comparison to Political Science experts’ opinions. The performance of these models and our analysis of their results suggest that these techniques are viable, promising, and suitable for new political studies in Brazil. Nevertheless, due to the inherent features of the Brazilian Legislative Branch, our estimations differ from the original interpretations regarding these ideal points by replacing the traditional left-right dichotomy and demonstrating how “ideological” or “pragmatic” the analyzed individuals are.