JXP: Global Authority Scores in a P2P Network

This document presents the \textit{JXP} algorithm for dynamically and collaboratively computing PageRank-style authority scores of Web pages distributed in a P2P network. In the architecture that we pursue, every peer crawls and indexes Web fragments at its discretion, driven by the thematic profile or overlay neighborhood of the peer. The JXP algorithm runs at every peer, and is initialized by a local authority computation on the basis of the locally available Web fragment. Peers collaborate by periodically ``meeting'' with other peers in the network. Whenever two peers meet they exchange their local information and use this new information to improve their local authority scores. Even though only local computations are performed, the JXP scores approximate the global importance of pages in the entire network. The storage demand of each peer is linear in the number of Web pages and the locally stored Web fragment. Experiments show the quality and practical viability of the JXP algorithm