Abstract:

k-anonymity is a current hot spot for privacy preservation. The existing k-anonymous methods ignored the quasi-identifiers different influences on the sensitive attributes and clustered the tuples only, which caused much information loss while publishing the data. To cope with this problem, a novel k-anonymity via twice clustering and the concept of influence matrix to express the quasi-identifiers influences on different sensitive attributes are proposed. The clustering techniques over influence matrix are investigated and the tuples with near influences on the sensitive attributes are clustered to achieve k-anonymity. The experimental results show that the proposed methods are effective and feasible to privacy preservation.Compared with basic k-anonymity, the methods have less average equivalence class size and less run time.

Key words: k-anonymity, privacy preservation, data security, clustering