Piegas: A Systems for Sentiment Analysis of Tweets in Portuguese

Abstract

Mining posts in social networks has great potential for new applications. However, the huge amount of data produced every day in these networks makes it impractical for people to manually undertake the task. The Piegas system determines automatically the sentiments, or polarity, of tweets written in Portuguese about a topic of interest. Two main requirements directed the design and development of the system: (i) good usability and (ii) good precision in determining the sentiments expressed about the topic of interest. With the first requirement in mind, the system, developed using Ruby on Rails and JavaScript, has a clean interface to enter the keywords that describe the topic of interest and to present the results in several levels of detail. To meet the second requirement, the system uses a Naive Bayes classifier to identify the sentiments of tweets, as the literature shows that this algorithm combines a good classificatory performance and low response time. The usability of the system was assessed using well-known ergonomic criteria, with satisfactory results. The systems also presented good results in experiments conducted to measure its precision in identifying the predominant polarity of tweets related to five different topics of interest.