Personalizing EigenTrust in the Face of Communities and Centrality Attack

Abstract

EigenTrust (ET) is a renowned algorithm for reputation management in adversarial P2P systems. It incorporates the opinions of all peers in the network to compute a global trust score for each peer based on its past behavior, and relies on a set of pre-trusted nodes to guarantee that malicious nodes cannot subvert the system. In this paper, we show that ET is vulnerable to community structure and a novel targeted attack based on eigenvector centrality, since ET ranks nodes close to the pre-trusted ones higher than those further away. To address these shortcomings, we propose Personalized EigenTrust (PET) which (i) enables each user to choose her trusted peers from the social network of peers, thereby eliminating the need of pre-trusted nodes and making the system autonomous, (ii) is effective in networks operating under various transaction models based on distributions such as random, community-like and power-law, and (iii) is robust to many types of attacks including the targeted one based on eigenvector centrality. Our simulation results reveal that PET outperforms ET under diverse transaction models and attack strategies.