There was the 'netflix challenge'. They published a pseudonymous list of (User, movie, rating), triplets and two lists of (User, movie) tuples. The idea was th

There was the 'netflix challenge'. They published a pseudonymous list of (User, movie, rating), triplets and two lists of (User, movie) tuples. The idea was that people could train a model to predict the ratings that correspond to the two lists of (User, movie) tuples. You could hand in a guess every day, and would get back data a score on the first list of (User, movie) ratings. In the end, whoever got the best score on the second list got $1,000,000. It was a really interesting challenge, and some good research came of it, but it turned out that it was pretty easy to de-anonymize users by correlating to e.g. IMDB watch lists.