Study On Sentiment Polarity Analysis Of Subjective Text

With the massive amount of information and resources available on the Internet, classifying, mining and analyzing this data, especially the sentiment of the online review data posted by users, can give us a new window of better understanding of consuming habits and public opinions of various users, and play an important role in decision-making for the enterprises and the government.For the beginning, this paper introduces the motivation, research challenges and promising applications of sentiment analysis, and then describes the major area concern, sentiment classification, by reviewing and analyzing some recent works. As compared to topic-based classification, sentiment classification has some unique features, resulting in specific solutions to their own problems. After examining key concepts and different types of solutions involved in common classification problems in sentiment analysis, this paper proposes two approaches to determining the sentiment expressed by movie and product reviews. The sentiment polarity of a review can be positive, or negative. This paper discusses experiments in the sentiment polarity classification of a collection of product reviews, using well known classification techniques—Naive Bayes and Support Vector Machine—with unigram presence and frequency as the features. And then a comparison between different methods applied to sentiment polarity classification is presented. The result shows that the unigram presence feature has the best performance among these methods. In the meanwhile, another approach is presented for the similar problem as well, which counts positive and negative terms, but also takes into account contextual valence shifters, such as negation used to reverse the sentiment polarity of a particular term. To compute the corpus-based values of the sentiment polarity of individual terms this paper use bootstrapping and term association scores (PMI-IR) with a small group of positive and negative terms, showing that both PMI-IR and contextual valence shifters improve the accuracy of the classification.