Trapping Malicious Crawlers in Social Networks

Shiju Li, Chul-Ho Lee, Do Young Eun

Abstract: In this paper, we study a problem of trapping malicious web crawlers in social networks to minimize the attacks from crawlers with malicious intents to steal personal/private information. The problem is to find where to place a given set of traps over a graph so as to minimize the expected number of users who possibly fall prey to a (possibly random) set of malicious crawlers, each of which traverses the graph in a random-walk fashion for a random finite time. We first show that this problem is NP-hard and also a monotone submodular maximization problem. We then present a greedy algorithm that achieves a ($1-1/e$)-approximation. We also develop an $(ε,δ)$-approximation Monte Carlo estimator to ease the computation of the greedy algorithm and thus make the algorithm scalable for large graphs. We finally present extensive simulation results to show that our algorithm significantly outperforms other baseline algorithms based on various centrality measures.