There are words that change its polarity from domain to domain. For example, the word deadly is of positive polarity in the cricket domain as in "Shane Warne is a 'deadly' leg spinner". However, 'I witnessed a deadly accident' carries negative polarity and going by the sentiment in cricket domain will be misleading. In addition to this, there exist domain-specific words, which have the same polarity across domains, but are used very frequently in a particular domain. For example, blockbuster, is specific to the movie domain. We combine such words as Domain Dedicated Polar Words (DDPW). A concise feature set made up of principal polarity clues makes the classifier less expensive in terms of time complexity and enhances the accuracy of classification. In this paper, we show that DDPW make such a concise feature set for sentiment analysis in a domain. Use of domain-dedicated polar words as features beats the state of art accuracies achieved independently with unigrams, adjectives or Universal Sentiment Lexicon (USL).
展开▼