Consistent Music Recommendation in Heterogeneous Pervasive Environment

Abstract

Seamlessly integrating services in a heterogeneous environment is a hot topic in pervasive computing. Given information explosion, it is wise to provide users services of recommending personalized information, although recommendation quality in a P2P network usually can not be compared with that in a centralized environment. In this paper, we introduce a music collaborative filtering system combining centralized and P2P recommendation algorithms together, which aims to provide consistent music recommendation services in a heterogeneous pervasive environment. Instead of bothering users for explicit ratings, we first track their listening behaviors and then extract implicit ratings using a new extraction mechanism. Meanwhile, we adopt a double-criteria strategy for the centralized algorithm, which integrates song recommendation and artist recommendation together. Moreover, we design a novel scalable gossip-based P2P recommendation algorithm that takes advantage of centralized services as much as possible with contexts switching. In addition, we shed some lights on the serendipity problem that is common in most recommendation systems.