Metodologia de identificação de polaridade em textos com base em projetos de lei brasileiros

Abstract

With the increase of discussion and interest in politics, both by society and by specialists, come up the need to analyze, study end follow the legislative’s and executive’s actions. But the large number of documents , draft laws, as well as their layout and writing pattern, makes this a difficult process. The necessity of identifying and clustering each draft law subject and separate these documents by their polarities in relation to the identified theme. The goal of this dissertation is to build an efficient methodology for document polarity and similarity identification, over the same subject. Brazilian draft laws on abortion liberalization were analyzed as case study. The first step is to identify and compare different machine learning techniques the can sort law projects on both against and in favor bias. The methodology analyses each part of the document, utilizing K-means grouping, to later apply a graph based process on every parameter combination certify the drafts have more connection among each other. To perform tests, previously classified draft laws manually sorted by a specialist were used and all results obtained with this method demonstrated very interesting an promising connections and peculiarities, thus helping to identify similarities and patterns in documents.