Markov Playlists

Why is nobody using markov chains to generate music playlists?

Playing music on shuffle is shit, full of jarring transitions, incongruous switches of tempo and topic. I masssively prefer listening to a carefully-mixed playlist, either via spotify or youtube, or an old-fasioned radio station like Radio Paradise.

So why not build a semi-shuffle? Start with one song. Look for all the tracks that get put onto playlists immediately after it. Play one at random. Then repeat the process — follow your second track with a random selection from everything DJs decided to play after it.

This is basically a Markov Chain.

It’s an idea blindingly obvious, relatively simple to implement, and in a domain that must appeal to thousands of CS students. Markov chains are immensely popular for toy projects online, since they generate fun output for very little coding.

But poking around online, I’ve not been able to find anybody using markov chains for playlist generation.

Leave a Reply

Your email address will not be published. Required fields are marked *