e.g. I'm sometimes seeing several essentially identical tracks at the top of recommendations (also mentioned in a comment by rayshan). You probably want to penalise tracks like that so they're pushed well down the list, i.e. penalise matches by metadata similarity (artist, title, etc).
OTOH I think it should boost results from more popular songs/artists, so the top result is less likely to be an obscure result that happens to sound similar. Some might argue it's a good thing to discover/highlight obscure artists, but for most users, it's more practical to recommend results that are already "proven" to be appealing. More obscure results could still be blended in if highlighting them is seen as a goal of the project.