https://en.wikipedia.org/wiki/Clustering_illusion
https://en.wikipedia.org/wiki/Poisson_clumping
The best song randomizer I've seen described generates a permutation of songs, then "plots" them in n dimensional space with closeness being closeness being a weighted distance of closeness in the playlist and whatever metadata you have about them. (artist, genre, year, mood). Then you iteratively break the clumps until you can't find any or the number stops decreasing. If you still have clumps then you raise the bar for what is considered a clump and repeat until you hit zero.