Reconhecimento automático de temas abordados e desvios temáticos em comissões da Câmara dos Deputados.

Abstract

Access to information is indispensable for creating a politically participative society. In Brazil, many initiatives aim to ensure transparency in the actions of the Legislative Branch. However, Brazilian National Congress committees still receive only a small fraction of the media attention dedicated to the plenary sessions. This scenario is harmful to civil society since the committees are the real stage for the political clashes and debates between the Brazilian parliamentarians. Using Natural Language Processing techniques, especially the generative statistical model Latent Dirichlet Allocation (LDA), this work presents an approach for automatic recognition of addressed themes and thematic deviations in events of the Brazilian Chamber of Deputies's permanent committees. The obtained results prove the applicability of this statistical model in the monitoring of current political debates, defining latent topics aligned with the themes of the committees and allowing the detection of the events whose debates were affected by thematic deviations.