Exploiting RLPI for Sentiment Analysis on Movie Reviews

Abstract

The rapid growth in internet usage has made people to share their opinions publicly. Public opinions generally influence the crowd to a great extent. It becomes important to analyze the sentiment expressed as opinion to derive useful conclusions. Sentiment Analysis (SA) on movie reviews deals with summarizing the overall sentiment of the reviews. In literature, many researchers worked on sentiment analysis on IMDb reviews by identifying relevant features and classifying the reviews. In this paper, we show that exploiting Regularized Locality Preserving Indexing (RLPI) as a feature selection method shows better results compared to other feature selection methods like Information Gain, Correlation and Chi Square when tested with classifiers like SVM, KNN and Naive Bayes. RLPI reduced the overall complexity by extracting discriminating features from the input data and improved classification accuracy.