Petronia Radio

[petronia]

• 14:10 transparency and trust: Pandora as "smart friend" #
• 14:11 the rickroll/hacking problem: Paris Hilton tagged as "brutal death metal" on last.fm #
• 14:12 the last.fm tag cloud for Hilary Rodham Clinton orz|||| #
• 14:13 somanyMP3s.com: shilling via recommender (the high-tech version of paying to get a single on the radio playlist) #
• 14:17 transparency/steerability: why this rec? (list common tags), delete or resize tags to change their relative importance to the rec #
• 14:17 music.tastekeeper.com #
• 14:19 reach (into the long tail): top 200 tracks account for 17% of track sales, 95% of indie tracks sold less than 100 dls (22% sales) #
• 14:21 1% of tracks account for 80% of sales; 13% of sales are from American Idol or Disney artists (2007) #
• 14:23 Oscar Celma (UPF) analysis of last.fm popularity: sure enough, a power distribution (also, audience bias) #
• 14:23 the top 8 artists represent 3.5% of total plays: Beatles, Radiohead, Chili Peppers, Death Cab...... Death Cab??? #
• 14:24 668 artists represent 50% of total plays #
• 14:26 recs starting from head (Beatles): 18.5% head (82 artists), 80.4% midrange (6569 artists), 1.1% long tail (239798) #
• 14:27 48.5% of top 20 in head, 51.5% in midrange, none in tail #
• 14:27 starting from mid-tail: 5% head, 71% midrange, 24% tail #
• 14:28 starting from far tail: 0% head, 18% midrange, 72% tail #
• 14:30 context: purpose (jogging vs romantic dinner), re-purpose (playlists and mixtapes), self-investment, shifting taste #
• 14:34 user types: savants (7%, Hype Machine), enthusiasts (21%, last.fm), casuals (32%, Pandora), indifferents (40%, iTunes Genius) #
• 14:36 random music discovery via shuffle play: 1) doesn't scale, 2) iPod whiplash, 3) music graveyard (64% of songs on iPod never played) #
• 14:39 Ishkur's Guide to Electronic Music (tinyurl.com/8ez53o: how to build it automatically? #
• 14:40 "the celestial jukebox": shades of Doctor Who ahaha #
• 15:26 was just told in class that blip.fm had gone bankrupt over the last week but can't find a thing about it online so hold that thought :/ #

Automatically shipped by LoudTwitter

Comments

[petronia.livejournal.com]

Doesn't scale = you can only calculate relationships between tracks for a shuffle-play algorithm on a limited number of tracks. Discussion only briefly touched this but my intuitive suspicion is that the problem is NC-hard. Corroboration being the fact that (apparently!) Apple's shuffle algorithm is designed for a number of tracks that's less than the number storable by an iPod Classic, so each time you set it to shuffle it takes a subset of ~20 albums.

Basically, even if all the music in the world were freely available in a gigantic archive, you wouldn't be able to listen to it on shuffle. XD Which leads to:

Music graveyard = "your iPod is where your music goes to die." People pile music on their iPods then never listen to it again. They surveyed 5,000 iPod users and found that on average, 64% of their iTunes tracks had 0 listens.

iPod whiplash = that feeling you get when shuffle pulls a loud rock track right after a string quartet movement. XD; Actually, I think Apple have just about eliminated this: I'm really impressed by the shuffle function on my nano, as it captures a lot of the same stuff I listen for when judging mixtape transitions. (That would include - to various degrees - similarities in key, melody, instrumentation, and stuff I don't have words for because I'm not a trained musician or sound engineer. I don't know if they use composer metadata, although I can't imagine why they wouldn't.)

[worldserpent.livejournal.com]

Ah, thanks for the info! I had no idea the apple shuffling algorithms were that complex. If people have all of this music and never listen to it again... Interesting. I had thought people only bought music that they liked, but if people have a lot of music but don't use over half of it, that does seem to show something about the music industry (though I'm not sure what).