Interactive portfolio selection involving multicriteria sorting models

Given a set of items, the portfolio selection problem involves selecting a subset of the items subject to resource constraints. We propose in this paper a multiobjective interactive approach based on a constrained Non-Compensatory Sorting model which integrates preferences both on items and portfolios in the same device. More precisely, we combine two evaluations models. The first one assigns items into two categories ( Good / Bad ) and model resource limitation using weighted cardinality constraints in such a way that the portfolio is composed of the items assigned to the good category. The second evaluation level compares portfolios on a set of portfolio-related criteria. We learn the constrained sorting model, based on a learning set, with SAT/MaxSAT language, which proves to be efficient for the preference learning task.