Framework for Evaluating Feature Selection in Opinion Mining

Abstract

Opinion mining is important in text mining applications in brand and product positioning, consumer attitude detection, customer relationship management and market research. Applications result in new generation companies, products for reputation management, online market perception and online content monitoring. Web expansion encourages users to contribute or express opinions through blogs, videos and social networking sites which provide information for sentiment analysis regarding a product or service. This study investigates various feature extraction methods performance and opinion mining classification algorithm. Evaluation is through the use of opinions from amazon.com with product reviews. Features extraction is from opinions using Term Document Frequency and Inverse Document Frequency (TDF×IDF). Feature transformation is through Principal Component Analysis (PCA) and kernel PCA. Feature selection techniques like Information Gain (IG), Mutual Information (MI) and Fisher Score select features. Extracted features are classified by Naïve Bayes, k Nearest Neighbour and Classification and Regression Trees (CART) classification algorithms.