On the Local Nature of Sybil Defense in Online Social Networks

This paper explores the fundamental limits of using only the structure of social networks to defend against sybil attacks. We derive the number of attack edges needed to foil sybil defenses based on each of the known statistical properties of social graphs: our results suggest that it may be impossible to use this properties to identify with high probability both sybil and honest nodes. We then explore a more modest goal for social defenses: to use local graph properties, such as community structure, to provide an honest node with a way to white-list a set of trustworthy nodes. While recent literature claims that any community detection algorithm may prove equally useful, we find that the choice of community detection protocol can dramatically affect whether sybil nodes are successfully kept from being white-listed. We demonstrate both analytically and experimentally that Local Page Rank can very effectively exploit a community’s local properties to keep sybil nodes at bay.