Privacy-Preserving Publishing Frequent Sequential Patterns

Releasing frequent sequential patterns can compromise individual privacy of underlying sequences. We propose two concrete objectives as a potential standard for privacy-preserving publishing sequential patterns: k-anonymity and α-dissociation. The first one, extended from k-anonymity model for data, addresses the problem of inferring patterns with very low support, say, in [1, k) where k is an anonymity threshold. These inferred patterns can act as quasi-identifiers in linking attacks. We show theoretically that, for all but one definition of support, it is impossible to reliably infer support values for patterns with two or more negative items (items do not occur in a pattern) based solely on frequent sequential patterns. We formulate possible privacy inference channels for the remaining definition of support. The This paper is an extension to the PAKDD’07 conference paper [1]. The paper gives comprehensive description and theoretical analyses. In particular, Theorems 2-6 and their proof are added, and related work, more examples and experimental results are included. The author would like to thank J. Chen, H. He, C. M. O’Keefe and his other colleagues in CSIRO, Australia for their comments and help for this work. Partial financial support for this work from NICTA is acknowledged. NICTA is funded by the Australian Government’s Backing Australia’s Ability initiative, in part through the Australian Research Council. NICTA Canberra Lab, Locked Bag 8001, ACT 2601 Australia. Email: [email protected]. Tel: +61 2 61259500. Fax: +61 2 61258660. The work was partially done when he was with CSIRO Mathematical and Information Sciences, Canberra ACT 2601, Australia. RSISE, the Australian National University, Canberra ACT, 0200, Australia.