Social Security Reform in Brazil: A Twitter Sentiment Analysis

Brazil's social security reform, approved in 2019, moved several social networks, revealing the high polarity of this theme. In this paper, we perform an automatic investigation of this polarity using sentiment analysis techniques. We collected 980,577 tweets published between January and November of that year and compared four supervised machine learning algorithms: Multinomial Naive Bayes, Logistic Regression, Support Vector Machines, and Random Forest. Due to the large volume of data collected, we employed a transfer learning approach to train these algorithms. As a result, the four algorithms predominantly classified the tweets as "neutral", that is, these posts do not explicitly expose their users' opinion about the topic. Of the tweets that showed polarity, there was a dominance of posts classified as "positive", i.e., their authors have positive feelings (support, affinity) concerning the Social Security Reform proposal.