Sentiment Analysis Applied to News from the Brazilian Stock Market

Investments in the stock market have grown in Brazil in recent years, especially considering the individual number of investors. According to data from April 2020, the Brazilian stock market reached the historic mark of 2.38 million active investors, and with this scenario, there is an increasing need to study the Brazilian financial market, seeking to better understand its fluctuations. Recent work in the literature indicates that a companys stock values can be influenced by published news. Therefore, this work contributes to the automatic sentiment analysis applied to news written in Portuguese and related to the Brazilian stock market. For this, we performed three sentiment analysis strategies: two based on machine learning, using the Naive Bayes classifier and a Multilayer Perceptron neural network; and the other based on the lexical approach. Also, we proposed two dictionaries, focused on the financial domain and adapted to Portuguese. Our results show that the Naive Bayes classifier and the Multilayer Perceptron overcomes the best lexical approach. It is worth mentioning that the accuracy achieved by the best lexical approach was with the adapted dictionary proposed here.