This and the other top comments are right that sharing an artist is only the crudest (and very noisy) way to estimate actual influence. I suppose one way to improve it would be to actually looking at the songs themselves, which might be possible with some deep learning? But that would require to actually have all the songs, which might not be feasible.