Sentix: An aspect and domain sensitive sentiment lexicon

Abstract

Sentiment lexicons have often been used to aid sentiment analysis. Most of these sentiment lexicons are general-purpose lexicons which assign a fixed polarity to every word. However, it has been noted that the polarity of words depends on both the aspect and domain, thus a general-purpose sentiment lexicon would not be able to accurately classify the sentiment of words. This paper proposes a method to automatically construct an aspect and domain sensitive sentiment lexicon which assigns polarity to a word depending on its aspect and domain, and make available Sentix which is an aspect and domain sensitive sentiment lexicon spanning over 200 product domains. Experimental results have shown that our lexicon produces significantly better results compared to other commonly used lexicons. We also observe the long tail distribution behavior of product aspects, and propose the possibility of aspect ranking by comparing the number of domains and number of sentiment words present for an aspect. © 2012 IEEE.