Abstract
In this paper we propose an argumentative multiagent model based on a mediator agent in order to automate the resolution of conflicts between decision makers for identifying knowledge that need to be capitalized and that we call “crucial knowledge”. We follow both an argumentative approach and a multi-agent system based on a mediator agent. This new approach allows the mediator agent to elicit preference of decision makers which can be different or even contradictory while exploiting and managing their multiple points of view to identify crucial knowledge. Concrete experiments have been conducted on real data from an automotive company and on randomly generated data. We have observed that a non-argumentative approach is more sensitive to the variation of the number of knowledge than an argumentative one. Indeed, the classification results using the multiagent system are consistent with classifications of human decision makers in nearly 80% of studied cases.