Learning as a methodology for partner selection in agent-based negotiation

Agent-based negotiation protocols, like contract net, require that every agent capable of performing a given task has to be considered in the negotiation. This can lead to extra bids and rounds in the negotiation process, and it could happen that we end up with a poor solution if all agents were to act defensively and favor price decrease to quality improvement. In real world scenarios, however, it is not uncommon that applicants to a certain contract have to undergo a preliminary phase, in which they are evaluated for their overall characteristics. Hence, if a given company does not fulfil these beforehand requirements, it would probably be excluded from the negotiation before it actually occurs. Our proposition is to establish these requirements through a learning mechanism, in order to select the most appropriate agents to negotiate with in every possible situation. Therefore, this article discusses a methodology to reach such premises. The chosen scenario is a travel service multi-agent system in which airliners need to decide which hotels to admit to negotiation, based on a dataset of historical customer bookings. The goal is to establish the requirements that hotel agents should comply to, in order to define an integrated travel service product through the negotiation process, resulting in customer satisfaction.