Possibilistic co-clustering based on extension of noise rejection scheme in FCCMM
Seiki Ubukata, Katsuya Koike, Akira Notsu, Katsuhiro Honda · 2017
When we have cooccurrence information among objects and items, fuzzy co-clustering performs fuzzy c-means-type clustering of objects and items simultaneously, where two types of fuzzy memberships for objects and items are estimated by maximizing the aggregation criteria of clusters such that mutually familiar object-item pairs have large memberships in a same cluster. This paper tries to improve a multinomial mixture models-induced fuzzy co-clustering (FCCMM) to a robust co-clustering model by introducing the possibilistic clustering concept. Because possibilistic c-means can be regarded as a cluster-wise independent noise clustering model, multiple noise FCCMM models with single cluster situations are independently performed for rejecting illegal influences of noise objects. The characteristic features of the proposed model are demonstrated through several numerical experiments.