A new feature selection approach in sentiment classification of Internet product reviews

Abstract

Due to the characteristics of the Internet product reviews, features which can truly represent the Internet product reviews can't be extracted just using traditional feature selection methods in sentiment classification. To address this problem, we propose a feature selection approach, by identifying product aspects, aspect evaluation words and modifiers, to look for more representative features for Internet product reviews. Experimental results show that only using aspect evaluation words and modifiers as features can help SVM classifier work well. The experimental results demonstrate the effectiveness of our proposed approach.