Statistical Correction of Average Customer Ratings for Product Ranking

Byron J. Gao, Frank Medjo

Abstract: Many e-commerce websites allow customers to contribute product ratings and reviews. Such customer feedback can be used to rank products and make recommendations. As a standard approach, products are typically ranked by their average customer ratings. A problem of this approach is that average ratings based on small samples exhibit very little statistical confidence. They can differ significantly from true average ratings resulting in misleading rankings of products. In this paper, we investigate a systematic approach that applies statistical correction to average customer ratings leading to more robust rankings of products. We also implement the approach with the Yelp API to demonstrate its utility.