A network science perspective of a distributed reputation mechanism
Rahim Delaviz, Niels Zeilemaker, Johan A. Pouwelse, Dick H. J. Epema · 2013
Reputation mechanisms are widely used in online networks to rank users or products, but despite their importance, very few studies have been done or published on their real behavior. In this paper, we study an Internet-deployed distributed reputation mechanism called BarterCast that is specifically designed for peer-to-peer file-sharing systems. The BarterCast mechanism is based on building a weighted directed graph from the data transfers that have occurred among the peers, and on employing the Maxflow algorithm in this graph to evaluate reputations. In this paper, we study this mechanism from the network perspective and we provide a detailed analysis, which includes such network topology measures as the degree distribution, node interconnectivity, the clustering coefficient, community structure, and distance measures. Besides, we study the geographical spread and content sharing behavior of the system participants and correlate the results with their connectivity in the network. We interpret each evaluated measure in the scope of reputation and file-sharing mechanisms and propose relevant implications and prospective applications for future designs. All the measurements are based on data that we have collected during two years of crawling the Tribler file-sharing network, which employs BarterCast as its reputation mechanism.