Evaluation of Multi-Agent Systems: The case of Interaction

This paper deals with the evaluation of Multi-Agent Systems (MAS) at the level of their interaction. Two problems that may be a bias in the evaluation and measurement of interaction are discussed. The first one is the difference between the quantities of information carried by a unit of interaction in two systems having different architectures. The second one concerns the interaction units that are received and cannot be exploited by the agent. In this work, an evaluation based on the weight of the information brought by an interaction is suggested. In order to achieve this, a MAS model, on which the evaluation is based, is defined. Then, the different problems and solutions which will help to evaluate the interaction are studied. Finally, the approach is applied on two different implementations that solve the same problem.