It’s not really though - if it was known in advance that you did not like a track then there would be no reason to recommend it. It’s the classic precision vs r

It’s not really though - if it was known in advance that you did not like a track then there would be no reason to recommend it. It’s the classic precision vs recall trade off: I can create a recommendation algorithm that only recommends your favourite song, forever, and that will have perfect precision but miserable recall. To increase recall we have to accept a drop in precision.