Investigating different methods for query selection in preference elicitation
Loading...
Files
Date
2014
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of New Brunswick
Abstract
The elicitation of preferences from individual users has been an increasing area of
research. The goal is to learn as much as possible about a user's values (utilities) for
different outcomes in order to perform tasks on behalf of the user. The aim of this report
is to build on existing work on assisting agents in choosing the most effective query to
pose to a user. The report will explore different ways of using the estimated utility values
of each outcome in a preference graph to calculate the expected value of each possible
query. Our simulation results will be evaluated using three evaluation measures: the
number of new preferences actually learned as a result of the query chosen, the number of
pairs of outcomes for which the true preference is correctly predicted by the new
estimated utilities, and the root mean squared error between the estimated utilities and the
true utilities. The performance of some of the methods implemented (Default and
Normal-0. 05) were quite competitive with the already existing Minimum method.
However, the other methods implemented performed poorly when compared with the
existing method.