Optimal Preference Clustering Technique for Scalable Multiagent Negotiation(Short Paper)

Abstract

We propose protocol for automated negotiations between multiple agents over multiple and interdependent issues. We consider the situation in which the agents have to agree upon one option (contract) among many possible ones (contract space). Interdependency between issues prevents us from applying negotiation protocols that have linear time complexity cost like Hill Climbing implementing mediated text negotiation protocol(HC). As a result most previous works propose methods in which the agents use non linear optimizers like simulated annealing to generate proposals. Then a central mediator can be used to match the proposals in order to find an intersection. But this matching process usually has exponential time cost complexity. We propose multi round HC(MR-HC) for negotiations with multiple and interdependent issues. In each round the normal HC is used to determine a negotiation deal region to be used by the next round. We propose that the agents should cluster their constraints by the cardinality of the constraints in order to get socially optimal contracts before applying MR-HC. To showcase that our proposed clustering technique is an essential one, we evaluate the optimality of our proposed protocol by running simulations at different cluster sizes.