That paper describes what they are doing.

http://en.wikipedia.org/wiki/Netflix_Prize

The class of algorithms that it's dealing with is called collaborative filtering and the Bell Labs team has done some impressive stuff there. The contest is more academic than it would seem on the surface, since as was mentioned in a recently posted article, the data set is rather narrow in scope. In other words, it's more of a collaborative filtering algorithms contest than a "find the best recommendation algorithm" contest.