Okay, big shot. You've collected a comprehensive list of everyone's favourite musicians, ever. You've even roughly clumped them by how popular they are, instead of using linear algebra to do it properly using a huge linear regression (a trick I'm still working on.) Now what?
First, do it properly by linear regression. (A trick I'm still working on.) This will give you a fairly solid metric. I recommend using the proportion of correlation of liking two bands to determine how close together they end up, then finding some way to reduce that data down to a comparatively small number of dimensions, albeit with expected imperfections (hence the regression.)
Second, get that thing into a two- or three- dimensional representation and draw spheres of interest on it for each person, in as NP-complete and precise a manner as possible. Even consider having negative spheres if it helps simplify representation. Remember to ignore musicians the person hasn't heard before.
Third, Determine their spread, epicentre, etc., and use the aggregate data of those n-dimensional spheres to establish rough names for each region of significance. Then, you can have a list of interest groups like "Electrohouse, centred on Justice, with an adventurousness of 15%, including the following artists: ..."
Fourth, make millions by stealing this idea. (Note to self: check if someone has already applied this entire plan to music.)
This answers the question of how to precisely describe known and common ranges of interest in just about anything. It can be applied to sports, kinks, movies, writers, high art, whatever, and it will help identify eclecticness in taste (one with many small isolated pockets of interest), and bring to light patterns that could lead to an explanation of the underlying causes of taste. Also, by focusing on coincidence of popularity rather than superficial similarity, we make sure that the underlying criteria are legitimate, and we avoid schemata that would put, say, Thomas Bangalter and the Pointer Sisters in the same cluster. All of this is pretty exciting.
As for "don't know" responses to the initial survey: these are extremely useful in practical application, because they can be given as recommendations that are likely to be within the user's range of interests.
And so, thanks once again to the power of linear algebra, the world became a lot less mushy and a lot more scientifically sound. This is probably not especially H-creative, but it's sufficiently P-creative for my tastes.
First, do it properly by linear regression. (A trick I'm still working on.) This will give you a fairly solid metric. I recommend using the proportion of correlation of liking two bands to determine how close together they end up, then finding some way to reduce that data down to a comparatively small number of dimensions, albeit with expected imperfections (hence the regression.)
Second, get that thing into a two- or three- dimensional representation and draw spheres of interest on it for each person, in as NP-complete and precise a manner as possible. Even consider having negative spheres if it helps simplify representation. Remember to ignore musicians the person hasn't heard before.
Third, Determine their spread, epicentre, etc., and use the aggregate data of those n-dimensional spheres to establish rough names for each region of significance. Then, you can have a list of interest groups like "Electrohouse, centred on Justice, with an adventurousness of 15%, including the following artists: ..."
Fourth, make millions by stealing this idea. (Note to self: check if someone has already applied this entire plan to music.)
This answers the question of how to precisely describe known and common ranges of interest in just about anything. It can be applied to sports, kinks, movies, writers, high art, whatever, and it will help identify eclecticness in taste (one with many small isolated pockets of interest), and bring to light patterns that could lead to an explanation of the underlying causes of taste. Also, by focusing on coincidence of popularity rather than superficial similarity, we make sure that the underlying criteria are legitimate, and we avoid schemata that would put, say, Thomas Bangalter and the Pointer Sisters in the same cluster. All of this is pretty exciting.
As for "don't know" responses to the initial survey: these are extremely useful in practical application, because they can be given as recommendations that are likely to be within the user's range of interests.
And so, thanks once again to the power of linear algebra, the world became a lot less mushy and a lot more scientifically sound. This is probably not especially H-creative, but it's sufficiently P-creative for my tastes.
