Abstract
Previous work on computational trust has delivered robust aggregation engines that proved competent in estimating the trustworthiness of agents. However, most of these approaches were evaluated through simulation using simple and static agents' populations that did not consider the social account of trust. In this paper, we show by experimental analysis that these approaches tend to perform poorly when the populations of agents form social structures that evolve with the situation and the relationship existing between partners. Moreover, we present an approach to social-aware computational trust based on Social Tuner, a new component that we developed that infers the benevolence of the trustee toward the truster from the available evidence. We show by experimental analysis that this approach outperforms the social-less aggregation engines in a relevant way. The experiments are run using a model of agents' behavior grounded on the literature of social trust and benevolence that we have developed and describe in detail in this paper.