A novel hybrid recommendation system integrating content-based and rating information

Abstract

Collaborative filtering (CF), the most efficient technique in recommendation systems, can be classified into two types: neighborhood-based model and latent factor model. Both are only based on the user-item interaction, or rating information, and do not take into account the item’s content-based information which may contain valuable knowledge. In this work, we propose a hybrid content-based and neighborhood-based recommendation system which utilizes the genome tag associated with each movie in the MovieLens 20M dataset. Experiment results show that our proposed system not only achieves a comparable accuracy but also performs at least 2 times faster than the “pure” CF methods.