Recommendation Quality Evolution Based on Neighborhood Size

Abstract

Automated recommender systems play an important role in e-commerce applications. Such systems recommend items (movies, music, books, news, web pages, etc.) that the user should be interested in. These systems hold the promise of delivering high quality recommendations. However, the incredible growth of users and applications poses some challenges for recommender systems. One of the concerns for current recommenders is that the quality of recommendations is strongly dependant on the size of the user's population. In this paper we investigate, with the scaling of neighborhood size, the evolution of different recommendation techniques performance, the increase of the coverage, and the quality of prediction. We also identify which recommendation method is the most efficient given reasonably small training datasets.