An Aggregation-Disaggregation Approach for Automated Negotiation in Multi-Agent Systems
Amal El Fallah Seghrouchni, Pavlos Moraı̈tis, Alexis Tsoukiàs · 2000
This paper describes an automated negotiation procedure based on an aggregationdisaggregation approach. Negotiation is based on an oriented graph, namely a negotiation graph. In this graph, nodes represent a state of the world while arcs represent multiple actions of the involved agents. Agents are equipped with a multi-criteria decision making model. Based on the negotiation graph, an agent can make an offer (a proposal of a path) using his strictly individual preference model (multiple criteria aggregation based decision step), while the same agent can receive the counter-offer of a counterpart. In this paper, such a model takes the form of a multiple-attribute additive value function. On these grounds and using a multiple linear regression model (disaggregation step), agents are able to make a first estimation for the parameters of the preference model of their counterpart (that is the trade-offs and the shape of the value functions). Using such an estimation, agents create an enhanced preference model including the estimated preference model of their counterpart in their own model. Then, they compute a new offer on the basis of the enhanced model. The procedure loops until a consensus is reached, that is all the negotiating agents make the same offer.