Recommendations and similarity calculation based on favourite songs' listens

As title says, I suggest to have second recommendation system which is based on favourite songs rather than groups.

There are groups which have only a small number of good songs (which are very-very-very-good), and they are falling behind in current scheme.

Also "listeners compatibility" and "groups compatibility" could be calculated based on this (the same way as now, but using favourite songs' listens instead of groups' listens).