Opponent modelling in automated multi-issue negotiation using Bayesian learning

In this paper, we show that it is nonetheless possible to construct an opponent model, i.e. a model of the opponent’s preferences that can be effectively used to improve negotiation outcomes. We provide a generic framework for learning both the preferences associated with issue values as well as the weights that rank the importance of issues to an agent. The main idea is to exploit certain structural features and rationality principles to guide the learning process and focuses the algorithm on the most likely preference profiles of an opponent. We present a learning algorithm based on Bayesian learning techniques that computes the probability that an opponent has a particular preference profile. Our approach can be integrated into various negotiating agents using different strategies.