Abstract:
Preference learning modernizes the statistical analysis of rank data. It aims to predict and model complex user choices and item orderings from high-dimensional behavioral data. In this context, one might want to cluster individuals or items to simplify the preference space and make the data easier to interpret. In this talk, we first present non-metric multidimensional unfolding as a particular instance of multidimensional scaling for preference data. Then, we introduce an alternating algorithm that produces non-metric unfolding and clustering of individuals while keeping the items fixed. The graphical output is an unfolding plot showing the cluster centers of the individuals overlaid with the items.