Semantic analysis of the Federalist Papers[0] comes to mind. It originally was not known which individual author wrote which paper, but stylometric analysis (i.e. word-counting and matching word frequency distributions of the unlabelled papers against those of labelled papers (in which the author was known)) made it reasonably straight-forward to identify the original authors.

[0] A set of historical papers of great political importance. http://en.wikipedia.org/wiki/The_Federalist_Papers

Taxi data from NYC was deanonymized:

https://medium.com/@vijayp/of-taxis-and-rainbows-f6bc289679a...

And then used to identify Muslim drivers:

http://www.reddit.com/r/dataisbeautiful/comments/2t201h/iden...

And then used to track celebrities:

http://theiii.org/index.php/316/which-celebrity-is-taking-a-...

I can't recall the article, but there was a case where public data was de-anonymized based on DOB and zipcodes, and it was incredibly successful in a given state.