Topic Modeling of Committee Discussions in the Brazilian Chamber of Deputies

Ensuring that civil society can monitor and supervise the actions of its representatives is essential to build strong democracies. Despite significant advances in transparency, Brazilian National Congress committees are presently complex to follow and monitor due to the lack of open structured data about their discussions and the sheer volume of activity in these committees. This work presents two contributions to this context. First, we create and present an open dataset including structured speeches of the 25 Chamber of Deputies' standing committees over the last two decades. Second, we use Natural Language Processing techniques - especially Latent Dirichlet Allocation (LDA) - to identify themes addressed on these committees. Based on these latent topics, we explore similarities and differences between the standing committees, their relationships, and how their debates change over time. Our results show that committees accommodate conversations - including their main topic and opposing agendas - and describe how the topics discussed in the committees reverberate external events.