Friday, March 27, 2009

in which we get down...to the unconscious!

So somebody reviewed 1,302 songs by the same number of bands, giving each one six words only.

But how to centrifuge this toxic dump? Clearly there was no possibility of scraping the page and wget-ing the lot; Sturgeon's Law (90% of everything is shit) applies to music as it does to few other things. I thought of trying to express my tastes in a set of criteria, that I might even implement in a python script, but on reflection this seemed to be too much like work, and anyway, it didn't really fit the aim. I wanted surprises, not confirmation.

Then I had an idea; what about applying some sort of statistical method? Yer man had given each song a rating between 1 and 5; as you know, Bob, if you ask people in a survey to rate something on a scale of 1 to 5, they will go for 3 far more often than you'd expect from a normal distribution, because it's the safe choice. But presumably the ones he gave a top rating to must have something.

And there were basically two ways a song could get into the bottom rank; either it was objectively arrant shite, or else it was incompatible with the other guy's tastes. Now, I have no idea what those are and no reason to assume they are anything like mine, so in fact, being one-starred could actually be a recommendation. Similarly, being top-rated could be either evidence of quality, or else just a matter of taste. And I had no reason to imagine either case was more likely. Further, the principle of management by exception was in my mind; the top and bottom 10% must be doing something right or wrong, so they're the ones to look at.

So I decided to ignore all the 2s and 3s and most of the 4s, and then make a selection from the ones that remained, based on unreason and hunch, and at least once on the basis that they came from Leeds.

And? I'm grinning with delight at the results, a pile of 31 MP3s of which 30 are by people I've literally never heard of and at least 28 are utterly great. Here's the really interesting bit, though: I can't tell which ones were 1s and which were 5s. Well, there is at least one exception to that, but as a rule, no, it is far from obvious. And why are so many fronted by women? This isn't something I'd noticed as a taste, although - horribly - I just remembered that my father owns a vast amount of vinyl by early 1970s hippy-chick singer-songwriters. Boxes and Nick Hornbyesque boxes of 'em. That's hardly characteristic of the list I came up with, but it is scary. Perhaps it's sampling bias - or maybe the quasi-automatic process got around my unconscious prejudices?

1 comment:

Douglas Kretzmann said...

so, inquiring (and lazy) minds would like to know which 31 ?

Your method seems to me an excellent protocol - 'he has great taste in music' means only 'he likes what I like'.

kostenloser Counter