Recommendation Quality Evolution Based on Neighbors Discrimination

An "automated recommender system" plays an essential role in e-commerce applications. Such systems try to recommend items (movies, music, books, news, etc.) which the user should be interested in. The spectrum of proposed recommendation algorithms are based on information including content of the items, ratings of the users, and demographic information about the users. These systems hold the promise of delivering high quality recommendations. However, the incredible growth of users and applications bring some key challenges for recommender systems. One of the concerns in current recommenders is that the quality of recommendations is strongly dependant on the neighborhood size and quality. In this paper, we propose a new peer-to-peer architecture based on prior selection of the neighbors. We investigate the evolution of different recommendation techniques performance, coverage and quality of prediction. Also, we identify which recommendation method would be the most efficient with this new peer-to-peer architecture.