Uma ferramenta para análise de sentimento para o Twitter

Abstract

The growth of social networks has provided some advantages among which we can highlight the sharing of information among users in a short space of time. The posting of opinions on different events that are currently on the rise and common among users of social networks results in a growing volume of data on a daily basis.However, this information generated through these posts can be very useful if they are treated correctly, thus generating knowledge that can be used in different areas. This article presents a study on sentiment analysis, more specifically based on the social network Twitter because it contains resources that facilitate data collection. For the composition of the work, a dataset was used with data collected from Twitter by people who work for the government of Minas Gerais, that is, the users' posts, known as tweets, that data went through a natural language processing in order to remove stopwords and unwanted characters. The Naive Bayes machine learning algorithm was used to classify sentiment in posts. The tool used presented, despite being used a simple test base, a satisfactory accuracy of 86.85% in the test phase. Subsequently, the tool was used in retrieved tweets and also proved to be quite usual, managing to predict all retrieved tweets.