Mining Musical Content from Large-Scale Peer-to-Peer Networks

Peer-to-Peer (p2p) networks are an invaluable resource for various multimedia information retrieval (MIR) tasks, such as user and song similarity, recommendation and trend prediction, mainly due to their size and ability to capture user preferences. This paper presents a study performed on musical content collected from the a large scale p2p network. Using the song files shared by users in the Gnutella network, we present a method for constructing a song similarity graph. With this graph, several MIR applications are presented, targeting estimation of song and peer similarity, approximation of missing metadata for improved content retrieval and music recommendation. We show that despite the high “noise” that inherently exists in the user generated content, it is possible to leverage this dataset and create scalable and efficient applications.