An Impression-Based Strategy for Defending Reputation Attacks in Multi-agent Reputation System

Abstract

As self-interested agents and malicious agents often launch various attacks to reputation systems and these attacks are usually deceptive, collusive, or strategic, it is difficult to keep reputation systems robust against multifarious attacks. Many filtering strategies have been designed for providing robust reputation evaluation and minimizing honest buyers' purchasing risks. This paper presents a novel impression-based strategy, which first gives an impression-based algorithm for selecting a group of lenient buyers and strict buyers respectively. Secondly, taking these groups as classification seeds, all sellers are pre-classified into honest and dishonest ones, and then all buyers are classified into honest, dishonest, and uncertain ones. Thirdly, sellers' reputation is evaluated based on the discounted ratings of honest, dishonest, and uncertain buyers. Several sets of experiments are designed to verify the effectiveness, accuracy, and robustness of our strategy. Results show that our strategy is accurate and robust in defending various common attacks.