Partitioning three-way dissimilarity data

This paper presents a methodology for partitioning two modes (objects and occasions) of three-way dissimilarity data based on the statistical modeling approach of fitting an expected clustering model, expressed in terms of dissimilarities and specified by a classification matrix, to the observed three-way two-mode data. Specifically, occasions are partitioned into homogeneous classes of dissimilarity matrices, and, within each class, a classification matrix, specifying a consensus partition of the objects, is identified. The parameters of the model are estimated in a least-squares fitting context and an efficient coordinate descent algorithm is given.