Predicting user rating for Yelp businesses leveraging user similarity

Users visit a Yelp business, such as a restaurant, based on its overall rating and often based on other factors like location, hours of location, type/cuisine or other attributes such as free Wifi. In addition to this, users gain useful insight for a Yelp business based on its top reviews and highlights. However, the average rating that a business has, or the top reviews/feedback as per certain users may not necessarily provide a user the right perspective that he/she seeks while selecting the most suitable Yelp business to visit. For example, in the case of restaurants, different people have different tastes and a high rating by person A might be due to a feature that person B does not appreciate (such as the spice content of food). Thus, even though a Chinese restaurant may have high ratings primarily because of several American raters, users with a different preference, say from the middle east may not find it palatable. A rating which takes into account the feedback of similar users is expected to help people in making the right selection and enhance their experience. With this as motivation, we have utilized an additive Collaborative Filtering model to predict the rating by a user for a business, and have combined that with estimated similarity between users. We show that as compared to only using average ratings by a user, or the average ratings for a business, our method provides more accurate predictions as evidenced by low Mean Squared Errors in predicted ratings on test data.