Abstract

Scenarios like recommendation systems are characterized by the presence of agents that play the role of experts by observing how much end-users would be interested in an item. However, their observations are subject to noise. To improve the service offered to end-users, a central agent can collect all the observations and deal with the noise by merging them, before deciding which items to propose to an end-user. This merging gives rise to interdependencies among the experts (who have a value/cost for the items when it is chosen by the end-user), that, in turn, create externalities effects. In this talk, this kind of scenario is analyzed, different types of externalities due to interdependency are described, and possibility and impossibility results in designing efficient, incentive compatible, individually rational, and weakly budget balanced mechanisms are presented. In the case study considered in this presentation, a federated search engine collects ads from a set of advertising providers, and decides which ads to display to a specific end-user.