Predicting the Performance of Opponent Models in Automated Negotiation

Abstract

Abstract-When two agents settle a mutual concern by negotiating with each other, they usually do not share their preferences so as to avoid exploitation. In such a setting, the agents may need to analyze each other's behavior to make an estimation of the opponent's preferences. This process of opponent modeling makes it possible to find a satisfying negotiation outcome for both parties. A large number of such opponent modeling techniques have already been introduced, together with different measures to assess their quality. The quality of an opponent model can be measured in two different ways: one is to use the agent's performance as a benchmark for the model's quality, the other is to directly evaluate its accuracy by using similarity measures. Both methods have been used extensively, and both have their distinct advantages and drawbacks. In this work we investigate the exact relation between the two, and we pinpoint the measures for accuracy that best predict performance gain. This leads us to new insights in how to construct an opponent model, and what we need to measure when optimizing performance.

Decoupling Negotiating Agents to Explore the Space of Negotiation Strategies

“…To our knowledge, there is limited work in which the performance of different opponent models is compared. Two examples are the work by Papaioannou et al [34], who evaluate a set of techniques that predict the opponent's strategy in terms of resulting performance gain, as well as computational complexity; and Baarslag et al [4,5], who compare the performance and accuracy of preference modeling techniques. The BOA architecture focuses on opponent models which estimate the (partial) preference profile, because most existing available implementations fit in this category; however, in principle, our architecture can accommodate for the other types of opponent models as well.…”

“…For example, by re-implementing the ANAC agents in the BOA architecture, it becomes possible to compare the accuracy of all ANAC opponent models, and to pinpoint the best opponent model among them. Following this approach, we are able to identify a categories of opponent models that outperform others [4,5]; naturally, this helps to build better agents in the future.…”

“…The answer to the second question depends on the component under consideration: for an opponent model, it is straightforward to measure its effectiveness [4,5,25]: the closer the opponent model is to the actual profile of the opponent, the better it is. The performance of the other two components of the BOA architecture is better measured in terms of utility obtained in negotiation (as has been done for acceptance strategies in [6]), as there seems no clear alternative method to define the effectiveness of the acceptance condition or bidding strategy in isolation.…”

“…The acceptance condition combines simple, utility-based criteria (called AC const and AC prev in [6]), and can be easily decoupled in our architecture. The opponent model is a variant of the Bayesian opponent model [4,5,24], which is used to optimize the opponent utility of a bid. Since this usage is consistent with our architecture (i.e., the opponent model provides opponent utility information), the model can be replaced by a call to the generic opponent model interface.…”

“…The BOA framework also allows us to search the large space of negotiation strategies [4,5]. Section 5.1 describes techniques integrated in the BOA framework that aid in this search by scaling down the negotiation strategy space.…”