Yelp Recommendation System

We apply principles and techniques of recommendation systems to develop a predictive model of how customers would rate businesses they have not been to. Using Yelp’s dataset, we extract collaborative and content based features to identify customer and restaurant profiles. We use generalized regression models, ensemble models, collaborative filtering and factorization machines. We evaluate the performance of the models using the Root Mean Squared Error (RMSE) metric and compare the models’ performances. For dealing with cold start problems, we use segmentation ensembles and 3 different imputation methods (mean, random values and predicted values) for filling in missing information.