Soul Mates Matching Algorithm

The algorithm is unknown to me, so this may already be implemented or thought of but... I have some very popular artists I listen to and some very obscure ones as well. I get tons of soul mate suggestions matching only the popular artists, several of less popular artists, and very few with both genres of artists. When finding soul mate matches, it may be useful to weigh matches inversely proportional to their popularity, kind of like TF/IDF. I.e., if user listening tendencies are represented as latent vectors maintaining information about preference, then perhaps scale the contribution from less-popular artists so that the resultant overall user vector is not dominated by popular artists, but highlights the uniqueness of the listeners choice.