MMMs-induced k-member co-clustering for k-anonymization of cooccurrence information
Katsuhiro Honda, Hikaru Sakamoto, Seiki Ubukata, Akira Notsu · 2016
k-anonymization is a basic technique for utilizing sensitive information in data mining without violating personal privacy, and can be efficiently achieved by a greedy k-member clustering, where each data record is coded reflecting cluster structures so that each anonymized record is indistinguishable from at least other k - 1 records. In this paper, with the goal of utilizing cooccurrence information, a k-member co-clustering-based k-anonymization approach is proposed induced by the MMMs-induced fuzzy co-clustering concept. Co-clustering is a useful method for extracting mutually familiar object-item pairwise clusters from cooccurrence information such as document-keyword frequencies in document analysis. The proposed k-member co-clustering model sequentially extracts clusters one-by-one such that the aggregation degree of each k-objects cluster is maximized from the viewpoint of MMMs-induced co-clustering. The advantage of the proposed method is demonstrated through numerical experiments.