Their recommender system likely has multiple inputs that are under specific constraints and weightings. How well their recommendation algorithms work will probably never be known by the general public as we are force-fed a steady diet of whitelisted staff picks and promotional items.