Learning opponent's indifference curve in multi-issue negotiation combining rule based reasoning with Bayesian classifier

Abstract

Learning an opponent's preferences in multi-issue negotiation with incomplete information is a challenge recently. A variety of means have been proposed to learn the preferences of issues. Most of these assume linear utility functions and learn the fixed weight of issues. However, according to the principle of microeconomics, which is decreasing marginal utility, the preferences of agents should follow indifference curve. Therefore, learning opponent's indifference curve is a crucial problem. To this end, we propose a method combining rule based reasoning with Bayesian learning. we first use rule based reasoning to learn the fundamental of agent types. Subsequently we use Bayesian classifier to get the probability of the opponent. The experiments demonstrate the selection of classifier is right and efficient.