From Reviews to Ratings: A model to predict ratings based on review text

User generated ratings have a profound effect on the success of a business on popular websites like Yelp. While a particular business may have an averaged rating associated to it, it doesn’t necessarily abide by the tastes of all users. Although some users may find the business particularly outstanding, it might not cater to another user’s tastes. Besides being related to user preferences, such varied inclinations might be related to a businesss ratings on hidden topics. In this project, we predict the rating for a given business customized to a given user. We present two methods for predicting ratings. The first one involves clustering the users probabilistically, based on hybrid features that seem the most important in determining a users interest. The second method makes use of LDA in extracting latent subtopics from review texts to determine the star rating. Finally, we present our findings, compare our model to the existing state of art techniques and explain the results thus obtained.