A Scalable Opponent Model Using Bayesian Learning for Automated Bilateral Multi-Issue Negotiation

Learning an opponent's preference is critical to achieving a winwin situation in automated bilateral multi-issue negotiations. Most of the existing opponent preference-learning techniques are not scalable to many kinds of opponents with different strategies due to their strong assumptions on an opponent's concession pattern. This study enables a more general assumption into the Bayesianlearning-based opponent model to address the mentioned disadvantage. The proposed method is experimentally compared with state-of-the-art opponent models and found to have higher accuracy and greater scalability in most cases.