A feature terms extraction method based on polarity analysis of customer reviews for content-based recommendation

Abstract

Our paper proposes a method for extracting feature terms expressing feelings regarding the use of a product from customer reviews on e-commerce sites, based on content-based recommendation. Considering previous research indicating that negative events and impressions have a greater impact than positive ones, we define terms relating to factors over which customers argue the pros and cons in reviews as features related to feelings regarding the use of a product. Our approach involves extracting sentences expressing opinions from customer reviews, and recognizing each evaluated term as a candidate for product features. Using the positive opinion ratio of each candidate to measure the extent of how divided the opinions of reviewers are, we extract feature terms for the selected product by considering a feature score based on the positive opinion ratio. We present an experiment to evaluate the utility of the feature terms extracted using our proposed method.