Arguing about social evaluations: From theory to experimentation

In open multiagent systems, agents depend on reputation and trust mechanisms to evaluate the behavior of potential partners. Often these evaluations are associated with a measure of reliability that the source agent computes. However, due to the subjectivity of reputation-related information, this can lead to serious problems when considering communicated social evaluations. In this paper, instead of considering only reliability measures computed from the sources, we provide a mechanism that allows the recipient decide whether the piece of information is reliable according to its own knowledge. We do it by allowing the agents engage in an argumentation-based dialog specifically designed for the exchange of social evaluations. We evaluate our framework through simulations. The results show that in most of the checked conditions, agents that use our dialog framework significantly improve (statistically) the accuracy of the evaluations, over the agents that do not use it. In particular, the simulations reveal that when there is a heterogeneity set of agents (not all the agents have the same goals) and agents base part of their inferences on third-party information, it is worth using our dialog protocol.