How does irony affect sentiment analysis tools?

Abstract

Sentiment analysis applications have spread to many domains: from consumer products, healthcare and financial services to political elections and social events. A common task in opinion mining is to classify an opinionated document into a positive or negative opinion. In this paper, a study of different methodologies is conducted to rank polarity as to better know how the ironic messages affect sentiment analysis tools. The study provides an initial under- standing of how irony affects the polarity detection. From the statistic point of view, we realize that there are no significant differences between methodolo- gies. To better understand the phenomenon, it is essential to apply different methods, such as SentiWordNet, based on Lexicon. In this sense, as future work, we aim to explore the use of Lexicon based tools, thus measuring and comparing the attained results.