Efficiently Eliciting Preferences from a Group of Users
Learning about users’ preferences allows agents to make intelligent decisions on behalf of users. When we are eliciting preferences from a group of users, we can use the preferences of the users we have already processed to increase the efficiency of the
- PDF / 248,575 Bytes
- 12 Pages / 429.442 x 659.895 pts Page_size
- 79 Downloads / 190 Views
Abstract. Learning about users’ preferences allows agents to make intelligent decisions on behalf of users. When we are eliciting preferences from a group of users, we can use the preferences of the users we have already processed to increase the efficiency of the elicitation process for the remaining users. However, current methods either require strong prior knowledge about the users’ preferences or can be overly cautious and inefficient. Our method, based on standard techniques from non-parametric statistics, allows the controller to choose a balance between prior knowledge and efficiency. This balance is investigated through experimental results. Keywords: Preference elicitation.
1 Introduction There are many real world problems which can benefit from a combination of research in both decision theory and game theory. For example, we can use game theory in studying the large scale behaviour of the Smart Grid [6]. At the same time, software such as Google’s powermeter can interact with Smart Grid users on an individual basis to help them create optimal energy use policies. Powermeter currently only provides people with information about their energy use. Future versions of powermeter (and similar software) could make choices on behalf of a user, such as how much electricity to buy. This would be especially useful when people face difficult choices involving risk; for example, is it worth waiting until tomorrow night to run my washing machine if there is a 10% chance that the electricity cost will drop by 5%? To make intelligent choices, we need to elicit preferences from each household by asking them a series of questions. The fewer questions we need to ask, the less often we need to interrupt a household’s busy schedule. In preference elicitation, we decide whether or not to ask additional questions based on a measure of confidence in the currently selected decision. For example, we could be 95% confident that waiting until tomorrow night to run the washing machine is the optimal decision. If our confidence is too low, then we need to ask additional questions to confirm that we are making the right decision.. Therefore, to maximize efficiency, we need an accurate measurement of confidence. R.I. Brafman, F. Roberts, and A. Tsouki`as (Eds.): ADT 2011, LNAI 6992, pp. 96–107, 2011. c Springer-Verlag Berlin Heidelberg 2011
Efficiently Eliciting Preferences from a Group of Users
97
Confidence in a decision is often measured in terms of regret, or the loss in utility the user would experience if the decision in question was taken instead of some (possibly unknown) optimal decision. Since the user’s preferences are private, we cannot calculate the actual regret. Instead, we must estimate the regret based on our limited knowledge. Regret estimates, or measures, typically belong to one of two models. The first measure, expected regret, estimates the regret by assuming that the user’s utility values are drawn from a known prior distribution [2]. However, there are many settings where it is challenging or impossible t
Data Loading...
