One of my most treasured cassettes from the days where I put a gross amount of effort into music collecting was that full b-side on the tape, with a Jon Spencer Blues Explosion bootleg collection on the other side.

Due to the complexity of song similarity, it would not surprise me if Spotify were using a machine learning-based algorithm. And if that's the case, it's essentially a black box, and not a single engineer at Spotify can tell you specifically why this is happening other than hunches and general ideas. It does make me wish that Spotify would let users manipulate the recommendations, though. Like a toggle for obscure vs mainstream, or limit to just billboard 100 hits or something.