ABSTRACT

Sentiment analysis determines the polarity of text, whether it belongs to a positive or negative polarity. One motivation for sentiment analysis research is the need for users and e-commerce companies to know the public opinion from blogs, online forums, reviews about certain products, services, topics, and so forth. Polar words like good, bad, excellent, boring, and so forth are key indicators for recognizing the overall polarity of the document as positive or negative orientation. Phrases can convey sentiment information more efficiently than individual words. For example, the word unpredictable may have a negative polarity in an automobile review, with the phrase “unpredictable steering,” but it could have positive polarity for a movie review with the phrase “unpredictable story.” Phrases are very important for sentiment analysis as individual words are incapable of incorporating contextual and syntactic information, which is very important for sentiment analysis. In this chapter, various semantic orientation–based approaches are discussed for sentiment analysis, which is previously reported in the literature. Next, challenges in the sentiment analysis problem are discussed. Further, a new efficient semantic orientation–based approach is proposed for sentiment analysis. The proposed approach works as follows. Initially, various features like unigrams, part-of-speech (POS) pattern–based features, dependency features, and modified dependency features are extracted. Next, supervised and semisupervised methods for the computation of semantic orientation of these features based on mutual information are investigated. Finally, the overall sentiment orientation of the document is determined by aggregating the polarity values of all the phrases in the document. Experimental results show the effectiveness of the proposed methods.