Résumé (EN)

In this work we introduce and study a general model of preference game. Our game belongs to the large class of games in which each agent aims at compromising between her innate preferences and the choices of the neighbours. Our game encompasses two models, i.e., the full information and the partial information settings. While in the former each agent is fully aware of the strategies of the neighbours, in the latter each agent has only access to a single representative strategy, which can be regarded as the output of an aggregation function. Both settings are quite general and allow for a systematic study of the fundamental questions related to the existence and efficiency of (approximate) equilibria in the strategic interaction of agents in social networks. Interestingly, our bounds on the price of anarchy are very general and are derived in terms of key features of the game (i.e., social impact, boundary and stretch). Moreover, our results show that the two settings behave very similarly, thus confirming our conjecture that in general the lack of information does not affect the existence and efficiency of the (approximate) equilibria.