AI seminar: A personalized approach to address unfair ratings in multiagent reputation systems

Speaker: Jie Zhang

In multiagent systems populated by self-interested agents, consumer agents would benefit by modeling the reputation of provider agents, in order to make effective decisions about which agents to trust. One method for representing reputation is to ask other agents in the system (called advisor agents) to provide ratings of the provider agents. The problem of unfair ratings exists in almost every reputation system, including both unfairly high and unfairly low ratings. We begin by surveying some existing approaches to this problem, characterizing their capabilities and categorizing them in terms of two main dimensions: public-private and global-local. The impact of reputation system architectures on approach selection is also discussed. Based on the study, we propose a novel personalized approach for effectively handling unfair ratings in an enhanced centralized reputation system. Experimental results demonstrate that the approach effectively adjusts the trustworthiness of advisor agents according to the percentages of unfair ratings provided by them. We then argue for the merits of our model as the basis for designing social networks to share reputation ratings of providers in multiagent systems.