The warm-start bias of Yelp ratings

Abstract: Yelp ratings are often viewed as a reputation metric for local businesses. In this paper we study how Yelp ratings evolve over time. Our main finding is that on average the first ratings that businesses receive overestimate their eventual reputation. In particular, the first review that a business receives in our dataset averages 4.1 stars, while the 20th review averages just 3.69 stars. This significant warm-start bias which may be attributed to the limited exposure of a business in its first steps may mask analysis performed on ratings and reputational ramifications. Therefore, we study techniques to identify and correct for this bias. Further, we perform a case study to explore the effect of a Groupon deal on the merchant's subsequent ratings and show both that previous research has overestimated Groupon's effect to merchants' reputation and that average ratings anticorrelate with the number of reviews received. Our analysis points to the importance of identifying and removing biases from Yelp reviews.