Querying User Preferences in Automated Negotiation

An automated negotiating agent must take into account the preferences of its user to negotiate effectively. In practice, these preferences are not always fully known; therefore the agent needs to support user preference uncertainty. We present a general framework to tackle the problem of user preference uncertainty in automated negotiation. We model the user's preferences as a utility function that is unknown to the representative agent. The utility is parametrizable by a finite dimensional real vector. The agent possesses a prior belief on this parameter and can query the user for information. We are interested in determining which queries will most reduce uncertainty of the belief through what we call their information potential. We propose an optimization problem with the goal of finding a sequence of queries maximizing the information potential. We present an application of this framework to a special type of linear additive utilities defined on a multi issue negotiation domain. We establish optimal querying algorithms for this application, and experimentally assess the quality of their robust guarantees.