A pre-evolutionary advisor list generation strategy for robust defensing reputation attacks

Shu-juan Ji, Haiyan Ma, Shu-lian Zhang, Ho-fung Leung, Dickson K. W. Chiu, Chun-jin Zhang, Xianwen Fang

Abstract: Highlights • Our approach better estimate trustworthiness of sellers and reducing risk. • Pre-evolution algorithm searching for optimal customized advisor lists. • Evaluating by buyer's own experience and seller's recent reviewers. • Superior in robustness, accuracy and stability under multifarious attacks. Abstract Trust and reputation systems are vital in large open distributed electronic commerce environments. Although existing various mechanisms have been adopted to guarantee trust between customers and sellers (or platforms), self-interested agents often impose various attacks to trust and reputation systems. As these attacks are usually deceptive, collusive, or strategic, it is difficult to keep trust and reputation systems robust to multifarious attacks. Many defense strategies employ a robust trust network (such as a trustable advisor list) for protecting buyers. However, in the evolution of a trust network, existing strategies consider only historical ratings of given buyers and advisors, while neglecting the timeliness of these ratings. Besides, only a single trust network is utilized to evaluate all sellers, leading to problems such as lack of pertinence and quite large deviation of evaluation. This paper proposes a novel pre-evolutionary advisor generation strategy, which first pre-evolves an optimal advisor list for each candidate seller before each trade and then evaluate each seller according to its corresponding list. After evaluating and selecting the seller, the buyer's own advisor list is evolved based on the pre-evolved optimal advisor list of chosen seller. Two sets of experiments have been designed to verify the general performance of this strategy, including accuracy, robustness, and stability. Results show that our strategy outperforms existing ones, especially when attackers use popular attack strategies such as Sybil, Sybil and Camouflage, and Sybil and Whitewashing. Besides, our strategy is more stable than compared ones, and its robustness will not change with the ratio of dishonest buyers.