Um benchmark para comparação de métodos para análise de sentimentos

Abstract

In the last few years thousands of scientific papers have explored sentiment analysis, several startups that measures opinions on real data have emerged, and a number of innovative products related to this theme have been developed. There are multiple methods for measuring sentiments, including lexical-based approaches and supervisedmachine learning methods. Despite the vast interest on the theme and wide popularity of some methods, it is unclear which method is better for identifying the polarity (i.e., positive, negative or neutral) of a message. Such a comparison is key for understanding the potential limitations, advantages, and disadvantages of popular methods. This study aims at filling this gap by presenting a benchmark comparison of 21 widelyused sentiment analysis methods and tools to better understand their strengths and weaknesses. Our evaluation is based on a benchmark of 21 labeled datasets, covering messages posted on social networks, movie and product reviews, as well as opinions and comments in news articles. Our results highlight limitations, advantages, and disadvantagesof existing methods, showing that their performances varied widely across datasets. Finally, we propose initial efforts in combining these methods with the aim of maximize the results of sentiment classification. Despite of this introductory attempt, we show that this is a promising strategy that needs further investigation.