Résumé en anglais

In this paper, we propose an interaction model which allows more realistic interactions for simulations of human societies and limits the communication cost. Interaction is a central concept when designing a multiagent system. Classically, interaction between two agents contains two elements: communication and the action which is the result of this information exchange. This view of interaction has underpinned most research in the multi-agent domain. However, if we draw an analogy with human behavior, the notion of interaction is more complex and we show that the classical ways of interaction managing are not adapted. To reach this objective, the environment could be used to mediate interaction between agents. A toy problem shows that our proposition helps the agents to adapt their perception of communications in the light of their interest and limits the communication cost in case of complex interactions between agents. A real application stemming from the transportation domain illustrates the use of our proposition to help agents to adapt their behavior to the context.