Automatically predicting the helpfulness of online reviews

Abstract

Online shopping websites provide platforms for consumers to review products and share opinions. Online reviews provided by the previous consumers are major information source for both consumers and marketers. However, a large number of reviews for a product can make it impossible for readers to read through all the reviews in order to collect information. So it is important to classify and rank the reviews based on their helpfulness to make them easily accessible by readers. This will not only help consumers finish their information search and decision making more easily, but also be valuable for product manufacturers or retailers to get informative and meaningful consumer feedbacks. Due to the lack of editorial and quality control, the reviews of products vary dramatically in quality: from very helpful to useless and even spam-like. The helpfulness of reviews is currently assessed manually by the votings from readers. This paper describes a machine learning approach to predicting the helpfulness of online reviews. The experiments conducted in the study were based on data collected from Amazon. We also discuss the determinants of the helpfulness of online reviews.