ESTIMATION OF REVIEW HELPFULNESS BY CONTENT COVERAGE AND WRITING STYLE

Customer reviews are helpful information to decide what to buy in EC sites. However, some customer reviews are not helpful. Therefore, users must choose helpful ones. This is difficult especially for users who do not have enough knowledge about the products. In this paper, we propose methods to classify customer reviews into helpful and unhelpful. To classify them, we focus on content coverage and writing style of reviews. Coverage expresses how many important words are contained by the reviews. Writing style is expressed as “formal” or “informal”, and it is classified by a machine learning technique. We made test data from user votes in Amazon.co.jp, and evaluated our methods. The accuracy of our rule-based method was 0.69. However, the method using coverage does not work well when many reviews have not been written yet. We analyzed the effects of the number of reviews, and confirmed that our rule-based method is less affected and achieves better accuracy than the method using only coverage.