BraveCat: Iterative Deepening Distance-Based Opponent Modeling and Hybrid Bidding in Nonlinear Ultra Large Bilateral Multi Issue Negotiation Domains

摘要

In this study, we propose BraveCat agent, one of the ANAC 2014 finalists. The main challenge of ANAC 2014 was dealing with nonlinear utility scenarios and ultra large-size domains. Since the conventional frequency and Bayesian opponent models cannot be used to model the unknown complex nonlinear utility space or preference profile of the opponent in ultra large domains, we design a new distance based opponent model to estimate the utility of a candidate bid to be sent to the opponent in each round of the negotiation. Moreover, by using iterative deepening search, BraveCat overcomes the limitations imposed by the huge amount of memory needed in the ultra large domains. It also uses a hybrid bidding strategy that combines behaviors of time dependent, random, and imitative strategies.