In group decision making problems, reaching a common ordinal classification satisfying fairly each of the group members is a highly challenging objective. A key problem it to limit the exchange of private preference information among the process stakeholders. In this paper we propose a multiagent system made up of a set of autonomous, behaviour-based agents that act on behalf of their beliefs, preferences and objectives. All agents aim to reach a common classification but have private preferences. The group classification is derived through a negotiation process in a cooperative context. A mediator agent is charged with the task of conducting the process. For each item, classifications are proposed to the all the involved user agents in order to collect their feedback. Our system has been developed with two objectives : the first is limiting the exchange of private information, while the second is to provide solutions that are fair to agents. Current ongoing works consist in evaluating our method, with respect to these two objectives.