Evaluating the Quality of Opponent Models in Automated Bilateral Negotiations