Fuzzy co-clustering considering site-wise confidence of vertically partitioned cooccurrence data
Toshiya Oda, Katsuhiro Honda, Seiki Ubukata, Akira Notsu · 2015
In many real world data analysis tasks, it is expected to get much more useful knowledge by utilizing multiple databases stored in different organizations, such as cooperation groups, state organs and allied countries. However, in many such organizations, they often hesitate to publish their databases because of privacy and security issues although they believe the advantages of collaborative analysis. To utilize such multiple databases without fear of information leaks, a privacy preserving procedure was introduced to fuzzy clustering for categorical multivariate data (FCCM). This method deals with vertically partitioned databases, and several experimental results demonstrated that the method can contribute to revealing global intrinsic co-cluster structure rather than individual site-wise analysis. In this paper, site-wise confidence is introduced to the method to weaken harmful effects of insignificant sites for estimating stable intrinsic co-cluster structures. Site-wise confidence is measured with a degree of coincidence between site-wise object memberships and the global ones. Several experimental results show that site-wise confidence is useful for estimating reliable co-cluster structures.