Noplace had already gone viral ahead of its public launch because of its feature that allows users to express themselves by customizing the colors of their profile.
You used to have (still have, I think) very simple friends-of-friends algos that are not ML, they are just graph connectivity and distance with some union-joins thrown in the middle. They work for recommendations like steam games of what people who like this game also liked.
It's potentially worse, because before you may have had an algorithm that a human could fine tune or whistleblow on. Now if it is trained on large datasets, the resulting algorithm is usually a blackbox weight matrix, they probably have no idea what pattern it is optimizing :S