Análise de sentimentos: Comparando o uso de ferramentas e a análise humana

Abstract

Sentiment Analysis emerged from the need to treat and evaluate texts, opinions and comments made by users on the Internet, in order to understand how they relate to a given entity. Several analytical methods have been developed in an attempt to better translate the uncertain and the subjectivity of human feelings. This research used data from interactions in a social network to identify the sentiments involved in each student comment. From the collected text different methods of sentiment analysis were used in order to identify which method had better results compared to the real ones. The comparison showed differences between these results and the real ones: While the used tools classified more than 40% of the comments as neutral, the analysis of the messages' author showed that 71% of the comments were positive. In the classification, the occurrence of outliers was identified as well as differences on the intensity of the sentiments acquired with each method. No approach, including the one performed by one of the researchers, was considered efficient enough as the highest level of accuracy obtained was less than 70%.