In a previous post about A Flat Hierarchy for Subjective mp3 Tags, I described the arduous and marginally rewarding task of tagging my entire library with as many ‘tones’ tags as AllMusic was able to provide. With foobar2000 0.9 final now less than a week away, these tags may prove useful soon enough. But a few weeks ago, impatient and curious, I decided to put them to another use:
By creating a tones/tones tree structure in foobar, I was able to count how often each ‘tone’ intersects with every other ‘tone.’ What you see above is the beginning of that data collection, which I ultimately planned to analyze in…some way.
After Googling around for ideas on tag clustering, I came across gCLUTO, a free piece of software that would, miraculously, do exactly what I needed â€” namely, magically figure out how best to cluster each tag with related tags. I figured four clusters would be a comfortable number, based on earlier reading I had done on a two-axis theory of musical emotion (intense/relaxed and positive/negative).
Unfortunately, my computer simply couldn’t handle even constructing and deconstructing the foobar tree without freezing up for about 45 minutes each time. Plus, collecting all this data would have meant hours and hours of work, for a goal whose benefits weren’t very clear to me at all, as well as a halt in incorporating new downloads into my library. It was a pretty exciting couple days while it lasted though.
3 Responses to “AllMusic’s Tone Intersections”
Cool idea! This looks like some of the AI clustering work i do for my day job, but more interesting. Could you send me the list of all of the tones/themes/moods etc. that exist in AMG? I’d like to give this a shot with the software (commercial and way expensive) i use at work.
Here’s a list of AllMusic’s moods:
Unfortunately this doesn’t show how often they intersect with each other; for that info, my tag data is necessary (I think — it might be fruitful to email some people at AllMusic and ask for that data).
Thanks for the interest.