DI-UMONS : Dépôt institutionnel de l’université de Mons

Recherche transversale
Rechercher
(titres de publication, de périodique et noms de colloque inclus)
2019-06-21 - Colloque/Présentation - communication orale - Anglais - page(s)

Belhacène Khaled, Mousseau V., Ouerdane Wassila, Pirlot Marc , Sobrie Olivier , "Ranking with Multiple reference Points: Efficient elicitation and learning procedures" in 25th International Conference on Multiple Criteria Decision Making, Istanbul, Turkey, 2019

  • Codes CREF : Intelligence artificielle (DI1180), Modèles mathématiques d'aide à la décision (DI1151), Théorie de la décision et des choix collectifs (DI4326), Recherche opérationnelle (DI1150)
  • Unités de recherche UMONS : Mathématique et Recherche opérationnelle (F151)
  • Instituts UMONS : Institut de Recherche en Technologies de l’Information et Sciences de l’Informatique (InforTech)
  • Centres UMONS : Centre de Recherche en Technologie de l’Information (CRTI)

Abstract(s) :

(Anglais) In the context of multicriteria ranking problems, we consider a ranking procedure based on reference points recently proposed in the literature. This method, known as Ranking with Multiple reference Points (RMP) makes use of the following preference parameters to specify the decision maker judgment: (i) a set of reference points and (ii) an importance relation on criteria coalitions. Using the RMP method, alternatives are compared to reference points considered successively in a certain order. An alternative is considered better than another if the coalition of criteria on which it is better than a reference point is more important than the coalition of criteria on which the other is better than the same reference point. Implementing the RMP method in a real world decision problem requires to elicit the model preference parameters. This can be performed indirectly by inferring the parameters from stated preferences, as done in previous research papers. Learning an RMP model from stated preferences proves however to be computationally extremely costly and can hardly be put in practice using state of the art algorithms. We propose a Boolean satisfiability (SAT) formulation of the inference of an RMP model from a set of pairwise comparisons, which is much faster than the existing algorithms based on mathematical programming. We present the SAT clauses that aim at inferring the model and explain them. The SAT formulation is tested on artificial data sets. We set up an experimental framework and assess the computing time and the performance of the formulation in generalization. In the experiments, we vary the size of the learning set and the size of the model that has to be learn. Finally, we present a MAX-SAT formulation which allows inferring a RMP model when the learning set is not fully compatible with such a model. The MAX-SAT formulation is tested on artificial datasets.