3 présentations

• 
  10h30 - 10h55

  Constrained TCP-nets

  • Malek Mouhoub, prés., University of Regina
  • Samira Sadaoui, University of Regina
  • Shu Zhang, University of Regina

  A Conditional Preference Network (CP-net) is a widely used graphical model for expressing qualitative and conditional preferences over attributes values. CP-nets have been recently extended to Tradeoffs-enhanced Conditional Preference Networks (TCP-nets) in order to capture the relative importance among attributes. In this paper, we extend the TCP-net to hard constraints and call the new proposed model constrained TCP-net. More precisely, the constrained TCP-net has the ability to represent and manage a given application under constraints as well as qualitative and conditional preferences over the attributes and their values. Solving the constrained TCP-net consists of finding the set of Pareto optimal solutions satisfying all the constraints and maximizing all the preferences. This task is addressed in this paper using a variant of the branch and bound algorithm enhanced with constraint propagation and variable ordering heuristics.

• 
  10h55 - 11h20

  Multi-objective optimization in multi-attribute combinatorial reverse auctions

  • Shubhashis Kumar Shil, prés., University of Regina
  • Samira Sadaoui, University of Regina

  Combinatorial Reverse Auctions (CRAs) often involve multiple conflicting objectives. Winner determination in CRAs is a NP-complete problem. Considering multi-unit and multi-attribute along with buyer's and sellers' constraints adds more complexity. Our ambition is to develop a multi-objective genetic algorithm to find the Pareto-optimal solutions for our new winner determination problem.

• 
  11h20 - 11h45

  Allocation optimale des budgets de campagne en marketing interactif

  • Koukla Azeuli, prés., Polytechnique Montréal
  • Michel Gamache, Polytechnique Montréal
  • Alain Hertz, GERAD, Polytechnique Montréal

  Nous proposons un modèle non-linéaire d'optimisation de campagnes publicitaires en ligne à partir de données de navigation partielles. Ce modèle permet de déterminer les valeurs d'enchères à miser sur les mots-clés dans les moteurs de recherche permettant d'améliorer la rentabilité de la campagne. Nous avons également développé un algorithme de résolution basé sur la recherche taboue.