Research On Helpfulness Of Online User Review

In the era of Web2.0, every user is content consumer and content creator at the same time. User generated content, such as user reviews, has grown exponentially due to the enormous base of user number. How to ease this information overload becomes more and more important. Finding high quality subsets of user reviews, as a major research concern, is challenging and has attracted many excellent researchers. In this paper we try to tackle this problem from both classification approach and ranking approach, based on a dataset consisting of online book reviews.Book reviews are different in two ways from reviews on commercial site. At first, book reviews are usually longer than product reviews. A typical good review contains more than one thousand words. Review length reflects the amount of information contained in the review. Secondly, book belongs to experiencing good. Users usually expect to find similar experiences or new things they haven’t found by themselves. And as a kind of creative work, writing style also plays an important role in a review.The contribution of this paper is mainly two fold. Firstly, we constructed a dataset consists of book reviews and gave a comprehensive analysis of its statistical character and user voting behavior. Secondly, based on review text and reviewer’s information, we proposed a novel weighting scheme of words to relate the review with its reviewed book. The experiment results on classification and ranking were reported and showed that the features we used achieved a better performance.