Application of Latent Factor Model on a Restaurant Menu Recommendation System

Abstract

A restaurant menu recommendation system is becoming a necessity following the trend of eating out at the restaurants. The task of such a system is to generate a top- N list of menus that may be of interest to a customer, in which the customer previous rating behavior is used as the learning model. In this paper, we apply the SVD (Singular Vector Decomposition) latent factor model as the learning technique of the recommendation system. Beforehand, we implement the mean imputation technique to fill in the missing rating entries so that SVD can also deal with the new customer that has no rating record in the system. Evaluation on a real-world restaurant menu recommendation dataset shows that our recommendation system is able to generate a top-10 list of menu recommendations to a target customer and that the results of the low-rank approximation using SVD are comparable with that of the full-rank.