Twitter, MySpace, Digg

Sentiment analysis is a growing area of research with significant applications in both industry and academia. Most of the proposed solutions are centered around supervised, machine learning approaches and review-oriented datasets. In this article, we focus on the more common informal textual communication on the Web, such as online discussions, tweets and social network comments and propose an intuitive, less domain-specific, unsupervised, lexicon-based approach that estimates the level of emotional intensity contained in text in order to make a prediction. Our approach can be applied to, and is tested in, two different but complementary contexts: subjectivity detection and polarity classification. Extensive experiments were carried on three real-world datasets, extracted from online social Web sites and annotated by human evaluators, against state-of-the-art supervised approaches. The results demonstrate that the proposed algorithm, even though unsupervised, outperforms machine learning solutions in the majority of cases, overall presenting a very robust and reliable solution for sentiment analysis of informal communication on the Web.