Multiple criteria decision aiding aims to recommend decisions that are consistent with the decision-maker's value system by making trade-offs among multiple criteria to measure the performance of alternatives. A challenge is that decision-makers may be unable to provide certain information about their preferences. This study focuses on the preference disaggregation analysis of uncertain pairwise comparisons between reference alternatives familiar to the decision-maker. Probabilistic linguistic preference relations are introduced to portray uncertain preference information, in which linguistic terms and subjective probabilities are used to express the intensities of preferences and corresponding belief degrees, respectively. The preference information is converted into constraints on value functions by triangular fuzzy numbers, and the value functions compatible with the preference information are estimated by linear programming to model the decision-maker's value system. We conduct a consistency analysis of probabilistic linguistic preference information as well as a robustness analysis of compatible value functions based on Monte Carlo simulations. Based on the data collected from Amazon.com, the proposed model is applied to laptop recommendations where the decision-maker is a single person and treadmill recommendations where the decision-maker is a group. The comparative analysis verifies the effectiveness of the proposed model in dealing with the uncertainties and intensities of preferences.
Read full abstract