A Greedy Algorithm for k-Member Co-clustering and its Applicability to Collaborative Filtering

Privacy preserving data mining is an important issue in network societies and co-clustering is a basic technique for analyzing intrinsic data structures in cooccurrence information among objects and items. In this paper, a greedy algorithm for k-member clustering, which achieves k-anonymity by coding at least k records into a solo observation, is enhanced to a co-clustering model. In the greedy algorithm, k-member clusters are sequentially extracted one-by-one, where each cluster is composed of homogeneous objects. In numerical experiments, the applicability of the proposed algorithm to collaborative filtering tasks is discussed.