24 Comments »

RSS feed for comments on this post. TrackBack URI

1. Todd, clearly what’s needed here is a robot music recommendations recommendation robot. Thanks for being a lesser, albeit human, proxy in a pinch, but I’ll wait for the upgrade.

   Rob

   Comment by roblord — February 10, 2006 #

2. I first started using launchCast about two years ago. It wasn’t more than a week before I wanted to upgrade my account. Since then, I have purchased at least 15 albums because of launchCast’s recommendations that I never would have without the service. I think that’s saying something. I have enjoyed the service so much over the past 2 years, but I can’t enjoy it anymore.

   In November I switched to Mac. I know you are probably sick of hearing it by now, but in order to keep this service popular, in my opinion, it needs to be able to run on several browsers, which also means several operating systems. I knew getting a Mac that I would loose the ability to listen to launchCast, and it was actually a major hesitation in my decision to make the switch.

   I also see launchCast as a gateway. This free personalized radio service was so good it got me to pay for it monthly, and because I was allready paying for that, I didn’t mind upping the subscription a few bucks when the Yahoo Music Unlimited started. I went from a $0/month customer to a $7/month customer just because of launchCast. I know that’s not a lot, but by supporting multiple browsers I really think you’re missing out on a big market of people who love music and are willing to pay for it.

   If there are any improvements in launchCast at all, I think that Firefox support should be at the top of the list, anything else is useless to a lot of people.

   Thanks for listening, sorry to write a novel,
   Mark

   Comment by mhusson — February 10, 2006 #

3. As a music fan, I’d like to point to a couple of then/now experiences I’ve had with Yahoo! Music’s recommendations:

   THEN: When I was in high school (graduated 1990, just to put it on the time continuum for your) I used to beg my mom to drive me from Goshen to South Bend, Indiana to go to Trax records and buy Maximum Rock N Roll Magazine. Then I’d go home, read the reviews, have my mom write checks, and order albums like the first OpIvy and Isocracy 7″s and The Victim’s Family’s “Things I Hate To Admit”.
   NOW: I open Yahoo! Music Engine from anywhere in the world and it tells me there’s a new Jello Biafra and The Melvins album in the same place on the page that iTunes would have showed me some lame “iTunes becomes eclectic” bullshit.

   THEN: In college I ordered all the Can (amazing German experimental rock) CDs from some european online store that was only available via telnet. They were $30 each, plus international shipping and a three week wait.
   NOW: When the Can albums came available in Y! Music Unlimited, there they were on the home page, saying, “HEY COME LISTEN TO ME!”

   It’s hard to deny the difference in experience an accessibility this creates. Not to needlessly evoke The Long Tail meme, but clearly items in the tail not only become much easier to find, but it also becomes much more likely that they’ll find you, and with much less effort.

   Robby, you’re right that this is merely a best guess and a proxy for the human approach. Clearly we believe in the human sharing approach, note the Email/IM/Link buttons on every album page, and the ability to do the same thing with playlists.

   But remember that these recommendations *are* created with human input. We aren’t recommending bands that “sound like” Radiohead to Radiohead fans. As a Radiohead fan that would be the *last* thing I’d want, some lame sound-alike. We are, however, recommended Beck to Radiohead fans. Even though they sound nothing alike, we can easily recognize the tendency of those fan bases to overlap.

   Finally (talk about writing a novel), I really haven’t found that Pandora recommends much obscure stuff as you (and Steve Clark) have argued. I put in Bob Dylan and got a station with Donovan, Townes Van Zandt, Springsteen, Bob Seger, The Doors, and CCR. Nothing too obscure there. In fact I’d say Springsteen, Seger, The Doors, and CCR are borderline bad matches for what I’d expect from a Dylan station. Even if it had been playing Dave Von Ronk and Ramblin’ Jack Elliot I’d have given it a pass.

   Also, I think it’s incredibly ironic that Pandora has gotten all this “Web 2.0″ buzz when IT DOESN’T LEVERAGE THE POWER OF THE NETWORK. A bunch of monkeys at keyboards classifying music is about as closed a system as you can get.

   ian

   Comment by iancr — February 11, 2006 #

4. I’ve been using all three services for varying lengths of time and love them all [especially the communities and groups in last.fm (formerly audioscrobbler)like it so much I subscribed for a paid account and it got better]. I’ve only recently started using Pandora. It seems okay. I need more time to develop an opinion though. Launchcast has always rocked with the expansion of genres in Plus and its wicked cool rating system. The new features which link to YME are gresat too.
   Big thumbs up to Ian for his customer support and taste in groups like The Can.

   Comment by enoch — February 11, 2006 #

5. One feature that would be nice is to build a playlist within YME based on several songs (i.e. up to 10?) to help target the selections as opposed to the feature that builds a playlist from just one song. Also – it would be nice to “Wall off” the play-list ratings (if one wanted to) to not affect the overall profile…Also on the moods in LaunchCast – does not seem you can create a playlist from a “Mood” station – just from the primary? Perhaps I am missing the “how to” here….Overall the YME is better than any other engine I have used.

   Comment by EB_Baller — February 11, 2006 #

6. Very interesting post, thanks. My only reservation is how fair and robust your “Black Eyed Peas test” is. There’s one issue about how replicable your judgements of ‘not-so-great recommendations for Black Eyed Peas fans’ would be among other experts. But secondly, and I think more importantly, there’s the question of whether doing a test based on just one band name (rather than an extended listening session) is properly representative.

   I think you could argue that a good personalisation engine would start off by making a broad set of recommendations in order to avoid making too rigid and blinkered assumptions about a listener based on limited evidence. A more rigorous and interesting test of a recommendation engine might be how well it responds to extended ‘training’, and whether it gets into all the corners of someone’s musical taste, rather than just the middle of the room.

   Comment by davidjennings — February 12, 2006 #

7. Ian,

   Huh? Don’t front for robots. And your arguments entirely miss my fundamental point.

   “But remember that these recommendations *are* created with human input.”

   Yeah, remembered. Absurdly arbitrary inputs human; aburdly arbitrary outputs robot.

   “We aren’t recommending bands that “sound like” Radiohead to Radiohead fans.”

   First, don’t associate “we” btwn humans and robots, besmirching both.

   Secondly, you’re wrong, sounds-like definitely enters the decision matrix at the human or robot level, but that’s moot.

   Lastly, my point remains that a robot music recommendation makes as much sense as robot music recommendation recommendation robot, that is, that only human (and I use the term loosely wrt Todd) perception avers a difference — no effective secondary robot might exist.

   To engage my point, argue that music affinity is deterministic, not indeterministic. You’ll lose. Argue that there is an intrinsic, universal human response and cultural sinew propagated by, say, Bowie’s Low. That’d be excruciating.

   Launchcast is doomed. Pandora is doomed. Robot music recommendations and the hard AI swamp from which it crept is doomed. Stop hurting music.

   Webjay shall soundly trounce Launchcast and none too soon. Go Lucas go.

   Comment by roblord — February 13, 2006 #

8. Rob,

   I like this argument as much now as I did at Pac Dining Car more than two years ago. It’s academic and looking for absolutes rather than useful solutions in the short-term. To say algorithm-generated recommendations and affinities are useless is to say that my THEN/NOW examples provided no value to my music experience (which would be fun-hating horseshit and you know it). It’s to say that “people who bought also bought” provides no value on Amazon (ditto).

   Sure, sounds-like is a factor in the human input, but I’m amazed how good the recommendations can be when it’s hardly possible. Check the album recommendations for Alice Coltrane’s “Journey in Satchidananda” or recommended artists for Captain Beefheart. You telling me there’s a lot of “sounds-alike” there?

   You act like the argument is *instead* of human-to-human sharing. Amazon has algorithmic recommendations AND list-mania. Should they remove the recommendations? That’s what you’re arguing, no?

   Clearly we’re in agreement on Webjay’s value.

   ian

   Comment by iancr — February 13, 2006 #

9. I think all of these services are valuable. The only question is which are the most valuable?

   Some despise “robot recommendations”, as I do, but I’m far from your average music consumer. My dad on the other hand LOVES Launchcast. He doesn’t know a lot about music and wants to click once to start the music and forget it.

   Many fall into this category.

   Pandora isn’t that great when compared to Launchcast recommendations. But it rocks when you enter a song name. Try entering an obscure song you love and seeing what it kicks back. I have only done this twice, but was impressed with the results. Not because I found new bands, but because I could hear the similarities in songs across genres. Pandora can tie moods and instrumentations together well, as long as the initial song is not that complicated.

   Now, if I could enter a producer, sideman, instrument, time signature, key, chord progression, subject, situation, etc and have it kick back the best songs with the parameters I set you’ve got something.

   SELECT song
   FROM YMU
   WHERE genre = funk or soul
   AND year = 1977
   AND quality > 8
   AND subject = girls
   AND guitar_effect = wahwah

   Comment by Robert of the Radish — February 13, 2006 #

10. I have heard a couple comments about the bake-off not being valid because I’ve only compared one artist. I actually compared quite a few artists, and only included one primarily due to space considerations. I invite you all to check out other artists and post your comparisons. I found Yahoo! Music’s fan stations to be more compelling than the others’, but I would like to know what you think as well.

    Comment by tbeaupre — February 13, 2006 #

11. Great post Todd. To me, Last.fm and Soundflavor are social networking sites that are built around music. Sure they recommend music, but that’s a by-product, not the primary focus. Pandora is far more interesting to me. It’s all about recommending music. The results are often too obscure for my tastes, but it’s an interesting service. What makes YME/LC+ perfect for me is that my rating effort leads to music discovery. I hear things, I rate them, YME learns from that, and sends me new music I usually like. The more you can tie LC’s ratings with MusicMatch’s music discovery engine, the happier I’ll be.

    I’d love to see MM’s AutoDJ make it to YME. Or at least let me create a Similar Artist Playlist from the artist page AND the More Mainstream/More Obscure pages (for that artist). The playlist generator that made it in the new release for device syncing gives me hope AutoDJ will come soon…

    Comment by buchananmb — February 14, 2006 #

12. Ian,

    I’m not suggesting robots are useless; I’m insisting they are pernicious. The traditional radio model at least joined people in herded mediocrity. Robotcasts are reductivistic, arbitrary dead-end singularities. Robotcasts hurts music in ways broadcast and muzak couldn’t attempt to achieve.

    b0th4t3r4l!f3,
    r0b

    Comment by roblord — February 14, 2006 #

13. The Yahoo! Music Blog Rocks…

    I’ve recently realized that the Yahoo! Music Blog is my new favorite Yahoo blog–official or not. As noted by DJ Alchemy: This isn’t the way we are used to billion-dollar corporations talking to their customers. In these and other examples, Yahoo emb…

    Trackback by Jeremy Zawodny's blog — February 14, 2006 #

14. Rob,

    “The traditional radio model at least joined people in herded mediocrity.” You seem to be implying that traditional radio mediocrity is better that what we offer? Based on what value?

    LAUNCHcast and recommendations leverage herds for their advantages, but do not reduce everyone’s experience to the lowest common denominator. They enable more diversified herds, each of which is more tuned to its members’ tastes than the traditional radio model.

    Care to provide any evidence to support your theory that our recommendations and customized radio are described accurately by ANY of the $2 words you used?

    * pernicious (causing great harm; destructive… of what?)
    * reductivisitic (minimizing… our product does not focus on a few characteristics, but millions of music fans’ opinions, which are far from simplistic)
    * arbitrary (seems like a filler adjective)
    * dead-end (based on comments I’ve seen from many users, LAUNCHcast broadens music horizons both in terms of consumptions and relationships with other music fans. personally, the best man at my wedding was someone I met through LAUNCHcast 6 years ago because we shared music interest in an obscure artist. he’s still a great friend, and we would not have met through broadcast radio)
    * singular (while there is individual control, recommendations are largely based on opinions of users with tastes in common)

    Also, recommendation systems are not predicated on affinities being deterministic or music generating a universally singular response. One person can like a certain peanut butter and boysenberry jelly sandwich because of the chunky peanut butter. Another person can like it for the raspberry jelly. A good recommendation system can deduce chunky peanut butter tastes from boysenberry jelly tastes, and use that knowledge to recommend great foods that might taste nothing like either, but are nonetheless delicious to you and your idiosyncratic human art-loving herd.

    Comment by tbeaupre — February 14, 2006 #

15. OK here’s a seed I’d like to plant.

    OpenTaste.org.

    End-users want to be able to move their personalization and behavior patterns between systems.

    I’ll leave it up to you nerds to figure out how to do it – but let’s not let these ‘tastes’ be used as a form of lock-in!

    Comment by marccanter — February 14, 2006 #

16. [...] One particularly interesting post is on various personalization systems. [...]

    Pingback by Marc’s Voice » Blog Archive » Yahoo Music Blog — February 14, 2006 #

17. I’ve not tried Yahoo Music but Jeremy’s blog post about you guys got me interested. I buy more music a month than is probably good for me and I’ve tried Pandora and Last.fm both for sources of new music.

    You know what those two services have over Yahoo Music? They work on my Mac and with Firefox.

    I would really truly like to try your stations – they look Great. But I can’t.

    Drop the MS Monopoly nonsense and gain some more customers (we’ve already proven we have open wallets – we bought a Mac, didn’t we?).

    Comment by danielcole — February 15, 2006 #

18. Having used Pandora some, I agree, it is pretty much novelty only. While it is nice to run into some obscure artists, I’ve found that you constantly have to add in new songs and artists to your station or you end up with the exact same sound over and over.

    I’m serious. It’s a total dead end for someone with eclectic tastes like mine. Even after I based my station on a rock song, a rap song, a trance song, and throw in maybe Hootie and the Blow Fish for a few suprises, within a week I ended up with 10 drum-and-bass songs in a row. Who’s running this thing?

    LaunchCAST has much more leniency especially with it’s moods. Sometimes I’m good for anything, other times, like during Calc homework, I need a something more ambient and electronic.

    I love Y!ME so far. It has it’s kinks, but every other program in the world goes through the same. I’m gonna hang out with you guys for a while and watch what comes out.

    Also, one feature I really would like to see you guys adopt: I would love it if I could with a simple click snatch the first 10 or 20 songs off of my recomended list and drop it into my MP3 player. Right now, I list all songs, drag the first 20-ish to a playlist, then drag the playlist to my MP3 player. I could do the automatic synch, but I’m still manually working with my playlists. Not a big deal, but it would really rock.

    Comment by krinderlin — February 15, 2006 #

19. Thanks Todd — this posting was very timely for me, as I’ve been going Last FM crazy the past couple of weeks (http://www.last.fm/user/indiesoc). A long-time Launchcast listener, I’ve been intrigued by the options that Last FM presents. First, it is much easier to set up new “stations” via the tag process. To sample new artists, just pick 15 that are available in the database and . . . you’re off (with subscription)! It is also a great way to explore less popular genres, such as psychedelic pop, krautrock, or post-rock, via (free) global tag radio. These genres and many others simply have no presence on Launch. Last FM has 2 other advantages worth mentioning: the free service allows unlimited skips (unlike Pandora and Launch), and access to music from the Beatles and Beastie Boys.

    I’ve also gone back to Pandora, and, taking Robert’s suggestion, entered a song title, with some pretty great results. I’m well versed in obscure bands, yet Pandora gave me artists I had never heard of that were also, in fact, similar to the song I had entered. I understand Pandora/Music Genome Project is a song-based recommendation system, so it might be that it works better when you enter a song rather than an artist. The obvious downside of Pandora is the user’s lack of control, including not knowing what is in the database.

    While both Last and Pandora have done a better job than Launch of introducing me to new music in the recent past, this has a lot to do with the fact that they provide, for me, untapped databases of music. I would still recommend Launch as THE place to start exploring for newbies. But, I do wish it were easier to find the hidden gems in the Launch database. The algorithm prioritizes new releases, but most everything else has to be searched out artist by artist. Fan stations are useless, since they play the most popular songs in the artist’s given genre and little else. (I was going to include my rant about the Zombies fan station here, but I’ll spare you the gory details.)

    In the end, I’m glad all of these sites exist. As far as music discovery options are concerned, I say the more, the merrier!

    Comment by indiesoc — February 15, 2006 #

20. Discover New Music…

    Inspired by the latest entry at Yahoo! Music Blog, I registered at Last.fm and started feeding my playlist information to them via the AudioScrobbler iTunes plugin. My profile there is slowly building and I am looking forward to checking out what their…

    Trackback by Andrei Zmievski — February 17, 2006 #

21. You know, I finally started using last.fm just before I read this post after coming over from Jeremy Zawodny’s site. So I figure hey, I should give Yahoo! a chance, right? But because I finally collected enough taste to switch to a Mac last year, you know how this story ends. Shame and heartbreak, baby. Why must this be?

    P.S. What, I’m supposed to use Netscape Navigator 4? Hi-ho, I’m surfing Arpanet! Sure, I guess I could download it from Decrepit Browser Depot, but then I could also turn on the radio.

    Comment by zachbaker — March 4, 2006 #

22. [...] Play It! A recent post at the Y! Music Blog got me thinking about music recommendation software like LAUNCHcast, Pandora and Last.fm. I definitely think they have their value, but I still feel recommendations by real people are superior. These services are great for folks without a lot of time or knowledge and are also good tools for playlisters to augment their work. Let me know what you think? This playlist features songs recommended by Pandora as sounding like “Glōsōli” by Sigur Ros. I had to disregard a bunch of dogs by Chicago, Celine Dion and Peter Cetera to come up with this list. [...]

    Pingback by » Sounds Like “Glōsōli” by Sigur Ros - Yahoo! Radish — May 8, 2006 #

23. [...] Sounds Like “Glōsōli” by Sigur Ros Play It! A recent post at the Y! Music Blog got me thinking about music recommendation software like LAUNCHcast, Pandora and Last.fm. I definitely think they have their value, but I still feel recommendations by real people are superior. These services are great for folks without a lot of time or knowledge and are also good tools for playlisters to augment their work. Let me know what you think? This playlist features songs recommended by Pandora as sounding like “Glōsōli” by Sigur Ros. I disregarded a bunch of dogs by Chicago, Celine Dion and Peter Cetera to come up with this list. [...]

    Pingback by » Sounds Like “Glōsōli” by Sigur Ros - The Rhapsody Radish - Music Playlist Archive — January 25, 2007 #

24. 
    Why are you roaming here and there if you can Download (100% Free) , Listen , Watch Videos (of related songs ) all the English , Hindi , Punjabi , pakistani songs , albums , live shows , concerts
    of all the singers , artists whom you like to listen

    This is an most user friendly Music Portal of WEB

    Comment by pz10 — July 7, 2010 #

Leave a comment

Name (required)

Mail (will not be published) (required)

Website

XHTML: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>