Experimental analysis of automated negotiation agents in modeling Gaussian bidders

Abstract

Automated negotiating agents are usually designed and implemented in a general way so that they can negotiate successfully in front of a vast variety of opponents. In the real world, most opponents are single-peaked. Gaussian agents that use such distribution function to rate the negotiation items are important sorts of such opponents. Modeling the opponents is of great importance since it enables us to adjust our next decisions accordingly. This can bring us short-time compromises, ideal eventual utility, more satisfaction, and so on. In negotiating with Gaussian opponents, the estimation of the opponent's peak point is the core. In this regard, we have paid particular attention to how accurate the existing automated agents attended in Automated Negotiating Agents Competition (ANAC) during 2010–2019 can model Gaussian bidders and showed the result of the experiments.