Active Preference Elicitation by Bayesian Updating on Optimality Polyhedra - Archive ouverte HAL Access content directly
Conference Papers Year : 2019

Active Preference Elicitation by Bayesian Updating on Optimality Polyhedra

Nadjet Bourdache
Patrice Perny
Olivier Spanjaard

Abstract

We consider the problem of actively eliciting the preferences of a Decision Maker (DM) that may exhibit some versatility when answering preference queries. Given a set of multicriteria alternatives (choice set) and an aggregation function whose parameter values are unknown, we propose a new incremental elicitation method where the parameter space is partitioned into optimality polyhedra in the same way as in stochastic multicriteria acceptability analysis. Each polyhedron encompasses the subset of parameter values for which a given alternative is optimal (one optimality polyhedron, possibly empty, per alternative in the choice set). The uncertainty about the DM's judgment is modeled by a probability distribution over the polyhedra of the partition. At each step of the elicitation procedure, the distribution is revised in a Bayesian manner using preference queries whose choice is based on the current solution strategy, that we adapt to minimize the expected regret of the recommended alternative. We interleave the analysis of the set of alternatives with the elicitation of the parameters of the aggregation function (weighted sum or ordered weighted average). Numerical tests have been performed to evaluate the interest of the proposed approach.
Fichier principal
Vignette du fichier
sum19.pdf (476.56 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-02419458 , version 1 (07-01-2020)

Identifiers

Cite

Nadjet Bourdache, Patrice Perny, Olivier Spanjaard. Active Preference Elicitation by Bayesian Updating on Optimality Polyhedra. SUM 2019 - 13th international conference on Scalable Uncertainty Management, Dec 2019, Compiègne, France. pp.93-106, ⟨10.1007/978-3-030-35514-2_8⟩. ⟨hal-02419458⟩
134 View
220 Download

Altmetric

Share

Gmail Facebook X LinkedIn More