Are On-Line Personae Really Unlinkable?

More and more personal information is available digitally, both collected by organisations and published by individuals. People may attempt to protect their privacy by avoiding to provide uniquely identifying information and by providing different information in different places; however, in many cases, such profiles can still be de-anonymised. Techniques from the record linkage literature can be used for pairwise linking of databases, and for cross-correlation based on these pairwise results. However, the privacy implications of these techniques in the on-line setting are not clear: existing experiments depend on quasiidentifiers and do not focus on cross-correlation. This paper studies the problem of de-anonymisation and, in particular, cross-correlation of multiple databases using only non-identifying information in an on-line setting.