I think there's another variable the model can't really account for (and I'm surprised wasn't mentioned much in the article, though perhaps it was already covered previously). People actually do like some songs better than others.
I expect that what friendships actually do is strongly modulate this like/dislike for particular songs. Based on the kind of music favoured in Europe that would probably go a long way explain why Sweden did so well and Norway did so terribly.
In earlier articles the "quality" of the song was in fact accounted for - if a song did better in the semi finals than its position within the network of countries giving each other points by default warranted, it was considered to be a better song.
That said, while the predictions had Sweden on top, it also had Norway in third place. Obviously the effect of a song being good isn't as strong as this model thinks. The top three had two countries that have lots of friends, Russia and Serbia, that had not-too-terrible songs that therefore managed to scrape away points even from countries that aren't friends.
I expect that what friendships actually do is strongly modulate this like/dislike for particular songs. Based on the kind of music favoured in Europe that would probably go a long way explain why Sweden did so well and Norway did so terribly.