tag:blogger.com,1999:blog-11758224.post114921509110450675..comments2023-09-28T07:23:51.376-07:00Comments on ProjectPerko: Opening Pandora's Black BoxCraig Perkohttp://www.blogger.com/profile/13173752470581218239noreply@blogger.comBlogger8125tag:blogger.com,1999:blog-11758224.post-1149602055945378172006-06-06T06:54:00.000-07:002006-06-06T06:54:00.000-07:00"Just out of curiousity, what sort of math is invo..."Just out of curiousity, what sort of math is involved in that?"<BR/><BR/>Bayesian Probability, [probably :D ]<BR/>Data Mining techniques would do it without having to re-invent the wheel, mathematically...Anonymoushttps://www.blogger.com/profile/08852086658302034166noreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149282023329535232006-06-02T14:00:00.000-07:002006-06-02T14:00:00.000-07:00I'll look into last.fm, thanks.As to the database,...I'll look into last.fm, thanks.<BR/><BR/>As to the database, I find it hard to believe that any number of specialists could reasonably map out the ten million songs out there. That would be absurd.<BR/><BR/>However, I didn't investigate the database very carefully, so it may be more reliable than I think.<BR/><BR/>As you say, that doesn't make the measurement any more accurate.Craig Perkohttps://www.blogger.com/profile/13173752470581218239noreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149280914739601532006-06-02T13:41:00.000-07:002006-06-02T13:41:00.000-07:00You're trivializing what Pandora has done a bit......You're trivializing what Pandora has done a bit... the database is not built by fans or by some third-world country's cheap labor, it was built by professional, experienced musicians in the LA area who were paid to "describe" the genome of the music. Thus, they know what minor harmonics, etc., are.<BR/><BR/>That being said, I would agree that the objective grouping of the music somewhat misses the boat: I listen to music for subjective reasons!<BR/><BR/>That's why I personally have found Last.fm to be the best of breed in this category. It builds it's database of associations based purely on listening habits of specific tracks. Not whole album buying habits (Amazon) or objective musical similarities (Pandora), but literally what you listen to.<BR/><BR/>Using a tool called the Scrobbler (or something like that), it plugs in to your favorite music player and tracks every track you listen to, how often you listen to it, how long you listen to it. Using that, it builds a model of association between that given track and the rest of your listening profile.<BR/><BR/>It's very cool. On their website they also have the necessary social software tools so users can tag, blog, comments, group, etc. around the music. Very cool, very functional, and the music selections are almost always dead on.<BR/><BR/>The coolest thing is that if you donate to Last.fm you get an upgraded account ($3/month). The upgraded account allows you to give Last.fm a list of artists and it will create a custom station that reflects those artists. I find it's success rate to be well above 90%.<BR/><BR/>Interestingly, you'll find bands like Radiohead and the Beatles are practically in every user's listening profile, thus no matter what band you choose you'll find Radiohead or the Beatles as an associated track... which, I'd say, is fairly accurate reflection of most of the web-savy users (most likely users of Last.fm) that I know in real life.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149280520570702962006-06-02T13:35:00.000-07:002006-06-02T13:35:00.000-07:00Ah, I see. Pandora and the Music Genome database a...Ah, I see. Pandora and the Music Genome database are the same core group. I thought the Music Genome database was unaffiliated.<BR/><BR/>Thanks for the correction!Craig Perkohttps://www.blogger.com/profile/13173752470581218239noreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149280222668308402006-06-02T13:30:00.000-07:002006-06-02T13:30:00.000-07:00The musical attributes database is Pandora's own a...The musical attributes database is Pandora's own and is presumably "clean." Whether that results in good recommendations is another matter, of course...<BR/><BR/>Robert Gable<BR/>http://rgable.typepad.com/aworksAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149267639500152202006-06-02T10:00:00.000-07:002006-06-02T10:00:00.000-07:00Amazon (and probably Google) use a one-dimensional...Amazon (and probably Google) use a one-dimensional version of what I'm talking about.<BR/><BR/>Yowsa. This turned into a long-ass comment.<BR/><BR/>I'll post about it. :DCraig Perkohttps://www.blogger.com/profile/13173752470581218239noreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149259531578913272006-06-02T07:45:00.000-07:002006-06-02T07:45:00.000-07:00Google makes its whole buisiness off of doing this...Google makes its whole buisiness off of doing this well. I see what problems you have with Pandora, I just skip the songs I don't like and deal with the 60% hit rate. <BR/><BR/>Just out of curiousity, what sort of math is involved in that?Patrickhttps://www.blogger.com/profile/13614962832390315553noreply@blogger.comtag:blogger.com,1999:blog-11758224.post-1149258163507831562006-06-02T07:22:00.000-07:002006-06-02T07:22:00.000-07:00I think this is actually what Amazon does with it'...I think this is actually what Amazon does with it's user system. You'd be really supprised how acurate Amazon's "here's things you might like" page can be, and it does it entirely based off people using the service, not on genre.<BR/><BR/>As an experiment (on my friend convinced me to do), you can go on to Amazon and start adding CDs you own and you can rate them. About half way through, it will usually start recommending things you already own, as well as other stuff you've never heard of. Very nifty.Anonymousnoreply@blogger.com