Adaptive Attacks on Discrete Rating Reputation Systems Considering the Herd Effect

Reputation systems are critical as a cornerstone of digital interaction, serving to assess entities and exerting a corrective influence that encourages users to engage in proper conduct. It plays a pivotal role in various domains, especially in those employing discrete five-star rating systems such as numerous E-commerce platforms. However, reputation systems also encounter significant challenges, including susceptibility to malicious attacks and vulnerability to group effects like herd behavior, which offer attackers opportunities to realize their extreme objectives. Numerous researchers have designed and analyzed effective attack methods, with the ultimate aim of enhancing the defense mechanisms of reputation systems. Therefore, considering group effects in the rating process, in this paper, we model the herding of user ratings and introduce the corresponding discrete rating reputation system. Based on the proposed model, we give simple attack methods and detailed theoretical analysis to precisely predict the attack effect. In addition, we propose adaptive attack methods, which are fundamentally based on adapting to the randomized decisions of the affected users, with the goal of influencing group decisions as much as possible and thus obtaining a better attack. We also analyze these adaptive scenarios and give some qualitative results. Simulations and experiments validate the effectiveness of the attack method, while real rating data reflects the accuracy of the model. By examining the attack dynamics in more depth, this work seeks to inspire the further development of system defenses that can contribute to strengthening social trust.