A Collaborative Filtering Recommendation Mechanism for Peer-to-Peer Video Sharing

Abstract

Almost all collaborative filtering recommendation systems based on C/S mode have to face the problems of one-point-failure and unscalable. This study proposes a scalable collaborative filtering recommendation mechanism for video sharing in unstructured Peer-to-Peer (P2P) networks. The mechanism is named as CFRPV, which can recommend videos in distributed way. The CFRPV mechanism includes four parts: peer model definition, neighbor peer set construction, CF-based recommendation for videos and neighbor peer set update. In CFRPV, peer users rank all the videos that they had watched. Then a video can be represented as a point in video vector space and its rank is the value of this point. One peer's preference also can be represented by a vector of the ranked videos in the video vector space. All the peers construct and dynamic reconstruct neighbor peer set in real time through calculating preference similarity between each other. From their neighbor peer sets, peers receive video recommendations that had been filtered. Finally, simulation results are discussed.