SmartWalk: Enhancing Social Network Security via Adaptive Random Walks

摘要: Random walks form a critical foundation in many social network based security systems and applications. Currently, the design of such mechanisms is limited to classical paradigm using fixed-length random for all nodes on graph. However, walk induces poor trade-off between other desirable properties. In this paper, we propose SmartWalk, enhancing system which incorporates adaptive We utilize set supervised machine learning techniques predict necessary length structural characteristics Using experiments multiple real world topologies, show that desired starting from specific node can be well predicted given local features node, knowledge small training nodes. describe node-adaptive path-adaptive usage models, where adaptively changes intermediate path, respectively. experimentally demonstrate applicability number privacy systems, including Sybil defenses, anonymous communication link preserving up two orders magnitude improvement performance.