The opinions of a few: A cross-platform study quantifying usefulness of reviews

Review platforms employ a voting mechanism, in which the crowd is invited to upvote reviews that they find useful. In this research, we investigate the attributes of over 1.2M reviews written by more than 327K users over three different review platforms. We find that useful reviews, on average, are authored by a small group of authors and that the best predictor for the success of a review is the author's history. The emotional effect of a review, measured by the amount of expressed emotions according to Plutchik wheel of discrete emotions, is of a lesser explaining value, but still performs better than textual features, and more so for services (as opposed to products). Regression analysis for the exact score reveals that apart from the cross-platform predictive power of the author's impact and review length, search goods and experience goods differ in what makes them useful. Structural review features are significant for experience goods but insignificant for search products. Discrete negative emotions correlate positively with the usefulness of reviews, more so for experience goods.