Online service provision is becoming increasingly decentralized as system designers pursue the benefits gained from utilizing nodes at the periphery of the network. However, distributing control means relying on the cooperation of participating agents, and it is a significant challenge to design mechanisms that incentivise optimal global behavior in a population of selfish, rational agents. This is particularly evident in peer-to-peer file-sharing, where a high incidence of selfish behavior in the form of downloading without uploading, leads to the network losing the benefits of a decentralized network. In this paper a notion of reputation based on simple social network analysis is used to significantly improve cooperation rates in the one-shot game of prisoner's dilemma, where without such a technique the dominant strategy would be for all agents to defect. © 2007 IEEE.
|Title of host publication
|2007 IEEE Congress on Evolutionary Computation, CEC 2007
|Number of pages
|Published - Sept 2007