[Updated with reasons why you most likely would want to use the Compaq's data sets over Advogato's. This update came as a result of pvg's diary entry implying there was little reason to use Compaq's data sets.]
Ankh mentions using the correlation between evaluation from various people to find out what to present.
This is a field that has seen some study; it is known as "Collaborative Filtering" (or just "Personalization" when the marketeers have been there), have implementations available from Firefly (now purchased by Microsoft, IIRC - their website is down, so I can't check), Net Perceptions (the commericalized aspect of Grouplens, and probably a couple of other companies.
If you are going to experiment with this, picking up a book on multi-variate analysis is probably not a bad idea. You can get free datasets from Compaq for research use. These measure how well a bunch of people have liked films (from EachMovie, while that still ran). They are probably better to use than the Master/Journeyer/Apprentice ratings from Advogato, as I expect "How well did you like that film?" to be more likely to evaluate the same criterion in each person than the various Advogato ratings. Apart from that, I expect the data sets to be much larger - EachMovie was mass-targeted and seemed quite popular.
Not that this is terribly relevant as a form of diary for me, but given the restrictions of the communications medium... *grin*