Chart Arcs: Visualizing Music Listening Behavior

Martin Dittus · 2006-11-09 · data mining · write a comment

A little while ago I implemented another visualization roughly based on Martin Wattenberg's arc diagrams. It's a visualization of a person's weekly Last.fm charts, designed to convey how your listening behaviors change over time, and also (just because it's really easy to determine) the mainstream-ness of your taste. The previous data visualization I had called IRC Arcs, so it's only natural to call this one Chart Arcs.

I'm currently thinking about meaningful representations of a person's music listening behavior, e.g. visualizations that show aspects of your musical taste and habits, and I think that this particular visualization is quite successful in conveying a small part of that.

All Word Extinguisher


joanofarctan	martind	spencerhyman

I made graphs of the Last.fm office crew, and it's interesting to compare different people's graphs -- joanofarctan has quite an obscure taste when compared to the rest (his graph has very slim arcs), my own graph shows how much I'm a fame whore in my musical taste (top chart positions are usually held by more popular artists, as the dot sizes of top chart positions show), and I quite like the graphs of of people who have a less-populated user profile, as you can see in e.g. the minimalist appeal of spencerhyman's graph.

Truffle? No, Shuffle!


skr	muesli	Jonty

I was surprised to see that some of the graphs were just hard to read random piles of data upon data, and I didn't really find an obvious explanation. See e.g. skr's big splotch. The listening behavior that resulted in those graphs didn't seem that much different from everybody else -- those were usually really active listeners, but others with a very active profile or a similar number of unique artists had very well-defined graphs.

I then started to get PMs from Last.fm users who wanted a chart arc of their profile, and it was then that a pattern emerged -- one user from Buenos Aires was wondering why his graph was fuller than some of the others, and we started to chat about his listening behavior, and it turned out that he's listening a lot in shuffle mode. I asked around in the office, and it turns out that our Last.fm colleagues with big splotches all listen in shuffle mode a lot of the time!

The resulting graphs now made sense -- "shufflers" will have a higher number of frequently recurring artists in their weekly charts when compared to "album listeners", who might might listen to a completely different set of artists from one week to the next. As we don't draw arcs when an artist leaves or enters your charts (it will only influence the coloring of the respective chart position's circle) graphs of shufflers will show much more local movement than those of album listeners.

Last.fm station of the day: Raum247's "feier-abend" tag radio. (And sorry for the very cheesy headlines...)

dekstop: weblog