They took some well-known methods from statistics, including something called matrix factorization, added bits of knowledge about movie recommenders (like the fact that movies seen long ago tend to be rated highly) and then threw all of it against the wall to see what stuck.

FORBES: Digital Tools