Finally got around to watching Dan Frankowski’s “You Are What You Say: Privacy Risks of Public Mentions” Google Tech Talk the other day. (I had the link set aside for two years. There’s a bit of a backlog.) In the talk, he takes an “anonymized” set of movie ratings and removes the anonymity by matching them to public mentions of movies in user profiles on the same site.
Interestingly, the ratings themselves weren’t as informative as the actual choice of movies to talk about. In the case of a site for movie buffs — ahem, film aficionados — I couldn’t help but think about participants in discussions using obscure film references as colored tail feathers as they try to out-strut one another. Of course this has significant impact on such a method, making the point that individual uniqueness is only a signature for identification: what makes you different just makes you more visible to a data mining algorithm.
The other interesting bit from the talk is about 20 minutes through, where starts to address ways to defeat such methods. There aren’t many good ideas, because of the tradeoffs involved in each, but it’s interesting to think about.