Exploiting interest-based proximity for content recommendation in peer-to-peer networks

Feasibility of content recommendation over interest-aware unstructured peer-to-peer (P2P) systems where peers sharing similar contents are connected. The authors present a novel and simple general metrics, by extending the Sorgenfrei coefficient to measure content similarities among peers. The authors provide two simple approximations of the proposed measure, that can be calculated by aggregating only the pair wise Sorgenfrei similarities, relaying on certain assumptions of statistical independence in the input data. The authors conduct experiments using a massive set of P2P file-sharing data to show that our new similarity measure could be a good predictor of the recommendation quality in unstructured distributed systems. The feasibility of finding similar peers in a simple unstructured system is also examined by simulation. The authors conclude that in unstructured P2P networks, an efficient recommendation system can be built without relying on any centralised or structured architectural extensions.