Predicting the semantic orientation of movie reviews

Online reviews are one of the important information resources for people. This paper focuses on a specific domain-movie review and presents a new model for predicting semantic orientation of reviews, i.e., classifying positive reviews from those negative. Different from traditional algorithms for sentiment classifications, this model integrates grammatical knowledge and takes topic correlations into account. Features are extracted, and the similarity between these features and the topic are computed also. The similarity is taken as a factor when evaluating the polarity of opinions. It´s shown by the experiment results that the proposed model is effective.