The field of Multicriteria Decision Aid (MCDA) aims to model in a formal way the preferences of a decision maker (DM) in order to bring informations that can help her in a decision problem. MCDA is interested in situations where the available options (called alternatives) are evaluated on multiple points of view. This work suggests elicitation methods: ways of questioning a DM or a group of DMs in order to obtain one or several preference models. These methods rely on socalled disaggregation techniques, which use exemplary decisions as a basis for building the preference model. In our context, the preference models are sorting models: they determine a way of assigning alternatives to preferenceordered categories. We are interested in a class of sorting models called MR Sort. We present a method that helps a group of DMs converge to a unique sorting model. It uses mathematical programs. We also analyze in detail the difficulties due to numerical imprecision when implementing these programs, and we propose an algorithm allowing to compare two MR Sort models. We introduce a novel way of interrogating the DM in order to take her hesitations into account, through the expression of degrees of credibility, when she gives assignment examples. Results of the method let the DM examine possible compromises between credibility and precision of the conclusions. We propose a method to choose portfolios. It encompasses two dimensions: absolute evaluation, in order to ensure that the selected alternatives are sufficiently good, and balance of the resulting portfolio. We also explain how this method compares to affirmative action. We describe the reusable software components that we have submitted to a web services platform, as well as functionalities developed in a library that implements the methods this work proposes. A data scheme exists that aims to standardize encoding of data related to MCDA methods, in order to ease communication between software components. We propose a new approach aiming to solve some drawbacks of the current approach. We develop as a perspective a proposal that aims to integrate preference modeling into the framework of realistic epistemology.
