Efficient Pairwise Preference Elicitation Allowing for Indifference

Many methods in Multi-Criteria Decision Analysis for choice problems rely on eliciting pairwise preference information in their attempt to efficiently identify the most preferred solution out of a larger set of solutions. That is, they repeatedly ask the decision maker which of two solutions is preferred, and then use this information to reduce the number of possibly preferred solutions until only one remains. However, if the solutions have a very similar value to the decision maker, he/she may not be able to accurately decide which solution is preferred. This paper makes two main contributions. First, it extends Robust Ordinal Regression to allow a user to declare indifference in case the values of the two solutions do not differ by more than some personal threshold. Second, we propose and compare several heuristics to pick pairs of solutions to be shown to the decision maker in order to minimize the number of interactions necessary.