Learning decision models with multinomial logit model through pair-wise preferences

Abstract

Our goal is to study the behavioral process of a decision maker (DM) that leads to his choice. To this end, we combine the established models of discrete choice with the recent algorithmic advances in the emerging field of preference learning. Our proposed model takes the learning information in form of the exemplary preference information, as revealed by a DM, and returns the DM’s choice probability. To accomplish our learning objective, we resort to the probabilistic models of discrete choice and make use of the maximum likelihood inference. First experimental results on suitable preference data suggest that our approach is not only intuitively appealing and interesting from an interpretation point of view but also competitive to state-of-the-art preference learning methods in terms of the prediction accuracy.