Effect of Similarity Distribution on the Convergence of Decentralized Similarity Overlays

Bukhari, I.F., Harwood, A. and Karunasekera, S.

Decentralized systems are used by a significant number of Internet users due to their specific features such as autonomy and content privacy. With the evolution of big data, the problem of filtering information based on user interests has become prevalent. Decentralized strategies for information filtering, such as similarity clustering, seem appealing due to its simplicity over model-based approaches (Matrix factorization). However, decentralized similarity clustering is not as well studied as centralized ones. In this work, we studied gossip-based similarity clustering with different real world similarity distributions. We analyzed convergence time and found the trade-off between convergence time and bandwidth utilization based on optimal protocol parameters (message size, neighbor-list size). The optimal settings of the protocol parameters not only give the minimum convergence time (approximately), from a worst case random structure, but also avoid wastage in bandwidth.

Cite as: Bukhari, I.F., Harwood, A. and Karunasekera, S. (2015). Effect of Similarity Distribution on the Convergence of Decentralized Similarity Overlays. In Proc. 13th Australasian Symposium on Parallel and Distributed Computing (AusPDC 2015) Sydney, Australia. CRPIT, 163. Javadi, B. and Garg, S.K. Eds., ACS. 13-22

(from crpit.com) (local if available)