FCM-type co-clustering of categorical multivariate data with exclusive partition

An FCM-type co-clustering model was proposed for handling cooccurrence matrices, in which co-clusters of objects and items are extracted using two different types of fuzzy memberships. Objects are partitioned into clusters in a similar concept with the conventional FCM, which uses the exclusive condition forcing each object to be exclusively assigned. On the other hand, memberships of items represent only the relative typicality degree in each cluster, and cannot be used for determining the clusters, to which each item belongs. This paper proposes a new approach for deriving the exclusive partition not only of objects but also of items in the FCM-type co-clustering model. In order to avoid each item to belong to multiple clusters, an additional penalty term for evaluating the degree of sharing is introduced into the FCM-type objective function, in which the aggregation degree of each cluster is maximized by forcing all items to be exclusively assigned.