Recommender Systems Designed for Yelp.com
Naomi Carrillo, Idan Elmaleh, Rheanna Gallego, Zack Kloock, Irene Ng, Jocelyne Perez, Michael Schwinger, Ryan Shiroma, Alexander Ihler, Sholeh Forouzan · 2013
Although people’s preferences are dicult to predict perfectly, providing reasonable predictions is highly useful, both economically and socially, in a broad spectrum of elds from marketing to demographics. Given such demand, developing and improving automated computational recommendation systems is a constant priority. The RecSys2013: Yelp Business Rating Prediction contest exemplies one idea of such a recommender system { it requires an algorithm that will make high-quality predictions about a collection of specic user and business pairs. We use common existing recommender systems, such as nearest neighbor predictions, weighted averages, matrix factorization, and clustering, further customizing these traditional methodologies to include the distinct features unique to the Yelp data set, such as number of check-ins, user gender, and review counts. Our methods achieved an accuracy of 1.24039 RMSE, placing us at 51st of 401 at the time of submission on the Kaggle leaderboard.