I’m about to go on a five hour evening drive. I’ve rented the cool convertible, stocked up on water, charted my route, and prepared myself for some pleasant alone time on the open road, aiming to arrive at my friends’ house at just around midnight.
But there was one last thing to take care of. Being an old fashioned sort, I decided not to leave my musical experience up to Spotify or some similar service. Instead, I went out to the nearest retail emporium and looked for some music to listen to as I wend my way.
And in the course of doing so, I realized that I really like comfort music. Not exactly music that I already own, but things I already know I like. I’m willing to be a little bit adventurous. For example, I got myself a copy of the new Daft Punk album. I may end up not liking it, but from what I’ve heard, I am fully expecting to get lucky.
I wonder in what ways the situation — for example, a long evening drive in a convertible, as opposed to a marathon work session — influences the choice of music.
Are we really accurate, when left to our own devices, in choosing the right music for each occasion? I am sure there are algorithms out there which map mood and situation to music, selecting the “best” music for each situation.
In practice, I wonder which performs better — the listener who relies on his/her own instincts, or the algorithm that aims to know your varying musical tastes even better than you do?
Netflix awarded a $1M prize for an algorithm aiming to know your movie tastes better than you do. I wonder if the same underlying mechanism applies to music?
Adding context to a system similar to Netflix sounds cool. You could start by segmenting out each corpus of song popularity by user reported context (mood, situation, etc). It might even work with a single user’s data over time to some extent.