Going Head To Head With Genius On Playlists 174
brownerthanu writes "Engineers at the University of California, San Diego are developing a system to include an ignored sector of music, dubbed the 'long tail,' in music recommendations. It's well known that radio suffers from a popularity bias, where the most popular songs receive an inordinate amount of exposure. In Apple's music recommender system, iTunes' Genius, this bias is magnified. An underground artist will never be recommended in a playlist due to insufficient data. It's an artifact of the popular collaborative filtering recommender algorithm, which Genius is based on. In order to establish a more holistic model of the music world, Luke Barrington and researchers at the Computer Audition Laboratory have created a machine learning system which classifies songs in an automated, Pandora-like, fashion. Instead of using humans to explicitly categorize individual songs, they capture the wisdom of the crowds via a Facebook game, Herd It, and use the data to train statistical models. The machine can then 'listen to,' describe and recommend any song, popular or not. As more people play the game, the machines get smarter. Their experiments show that automatic recommendations work at least as well as Genius for recommending undiscovered music."
Re:Wow. (Score:5, Insightful)
I know you say you've tried all possible music sites... but on Pandora if you create a new station from an artist or song, they'll give you the criteria they use to populate the playlist.
Set up stations based on enough songs, and it's pretty easy to understand at least part of how their algorithm works. A big problem, of course, is that some of the criteria are somewhat subjective, which is why you may disagree with them. I find this especially true when creating stations based on artists, not songs.
I just wish I could tweak the individual conditions to see where it'd get me... like having all criteria match except genre.
Re:Sponsored herd-it advertisment? (Score:5, Insightful)
Anyone actually use "Genius"? (Score:4, Insightful)
I tried Genius for awhile, but I recently disabled it for two reasons:
1) The "recommendations" were not very good nor did they maintain a "common theme", by which I mean, I chose a rather edgy electronica/punk song by Crystal Castles... three songs down we get something by The Nationals... which is very mellow rock. If I choose a song that is edgy, electric, and with a faster pace, I want ALL the songs in that 25 song playlist to be at least within a similar genre.
2) It takes up too much time when importing vast libraries to new machines. I recently centralized my 300+ GB music library on a Mac Mini Server, iTunes was unusable due to genius choking on the sheer volume of data it had to deal with.
In the end, it's really nothing more than a way for Apple to try to get you to buy more crap from the Apple store.
They lost my wallet years ago to Amazon MP3 store who had no DRM. I see no reason to go back to iTMS even now that their DRM is gone. Especially seeing what dicks Apple has been with their conduct around ACC, "fair play", and App Store lock-down.
for some, 'good' and 'popular' are the same (Score:4, Insightful)
The Academic meets Capitalism (Score:4, Insightful)
Capitalism: Hello there Academic. How are you?
Academic: Hi... what they heck are you? You look so strange to me...
Capitalism: I'm Capitalism. Oh, I'm really not all that strange, but I might be a little complicated to understand.
Academic: Complicated!?! I am the master of complicated, I am an Academic for crying out loud.
Capitalism: Ok then. Let me try to explain myself. I am a system that provides stuff via supply and demand.
Academic: Nonsense! I dont hear music that deserves to be heard on the radio or on popular websites!
Capitalism: Deciding who deserves what really isnt my thing... see... its about supply and demand...
Academic: But who decides whats in demand!?! Certainly it cannot be the uneducated "masses", they... just aren't qualified!
Capitalism: No no... its about what many individuals, smart or otherwise, want based on need or dozens of different other factors.
Academic: Preposterous! How could they possibly know what they want or need if they havent been exposed to it?!? Foolish Capitalism!
Capitalism: Well, there are a lot of musicians out there and only so many different ways to get them heard, and, well, there are people out there who spend their lives learning what people like and dislike, and even they arent always right... so the best at determining who does best succeeds...
Academic: Rubbish! What we really need, is for the qualified, with a broad base of tastes to make an application for people to give them a view of all the music that is out there!
Capitalism: I guess you can try, no one can stop you, but you might not succeed.
Academic: Your so short sighted. I don't need to worry about succeeding, I receive public money to pursue my higher realm of thinking.
Capitalism: Right on... so I guess you will compete and regardless if your product sucks, you dont have to worry about it because your really just spending someone elses money.
Academic: Its progress my dear boy. Progress.
Re:Bias exists for a reason (Score:5, Insightful)
Are you generally this obtuse?
The idea is not to popularize shitty bands. Given perfect AI, this program is supposed to do the following:
1) Listen to all popular music (for various classes of popular).
2) Figure out why that music is popular (for its class).
3) Listen to any *new* track and figure out if it is like those popular tracks (and any popular class).
Now of course we don't have that kind of AI and hence all this research.
The idea is to promote good bands that would have been popular except for the fact that they are not already popular and hence might go unnoticed.
Re:Bias exists for a reason (Score:0, Insightful)
Um, no. Popular bands are popular cause they pay a lot for coverage advertising and record labels.
*Some are very good, some are crap. My point is that being popular does not necessarily correlate with being good.*
I listen to music, I play music, I'm not one of the "all new music sucks" idiots, I also listen at least something from each genre so it's not that I don't like X genre and while people do.
I know a lot of people who listen to good music, most of my friends spend a lot of time showing each other songs and bands--we don't always like eachothers music, but we admit that it's at least good. A lot of people do this, but there are also a great majority of people who only really have 20 songs on their iPod that they bought because they thought it was cool as they heard it on hot 99.5 or such.
Now, I'm talking about highschool kids so maybe my observations are biased.
Re:Bias exists for a reason (Score:0, Insightful)
There's two flaws in your argument. The first is the idea that exposure has no training effect on a listener's likes and dislikes. I don't have any data on this but I'd bet that's untrue. The second flaw is assuming that unpopular bands are unpopular because they are essentially inferior products in the eyes of the market. This might be true in an efficient market but if the market itself is dominated by a few small outlets (Clearchannel Communications for instance) popularity doesn't indicate a better product or user preference so much as it indicates monopoly power over the industry. Microsoft and McDonalds would be similar examples that spring to mind.
Re:Mathematics != human preference (Score:5, Insightful)
But that's not the point of the algorithm. The point is to generate a playlist of songs that share some characteristics with criteria you have specified via examples (seeds for a station, and up/down votes on songs in the playlist). It's not about your personal likes and dislikes, it;s about songs that are similar.
My personal experience: Don't use downvotes until the station is somewhat mature. Use upvotes only, so the algorithm can find the common ground basis for the station. Then, after 10-20 hours of playtime, use downvotes to start eliminating unwanted characteristics.
One last point: Pandora is good for general tastes. As your wants get very specific within a genre, as you point out, it starts to fail. My general advice for you is to not try to use Pandora to create a pseudo-random playlist of only songs you know you like. After tailoring your station, buy the songs you like. Then you can create a playlist in your preferred audio-file management software, and listen to only the songs you like. Revisit Pandora or elsewhere to expand your collection as needed. I think this is the only way you'll be really satisified.
Re:Wow. (Score:3, Insightful)
I'm still skeptical about these algorithms for music recommendation.
You should be skeptical, but not dismissive. In its current state, this type of service is more like directed browsing than a true recommendation. But it still yields the an occasional gem, and with continued participation and increased competition it will get better. Skepticism makes it a useful tool, if you can live with having to wade through some misses along with the hits. Blind acceptance will, of course, be mercilessly exploited and the unwashed masses will still end up listening to the likes of [insert music you despise].
Re:Mathematics != human preference (Score:4, Insightful)
Re:The Academic meets Capitalism (Score:3, Insightful)
Re:Bias exists for a reason (Score:3, Insightful)
3) Listen to any *new* track and figure out if it is like those popular tracks.
The idea is to promote good bands that would have been popular except for the fact that they are not already popular and hence might go unnoticed.
In this scheme, the no-name band that is most successful in cloning the big-band sound will score the highest.
You might as well be hosting the Fat Elvis competition at the state fair grounds.
At least as well (Score:2, Insightful)
An underground artist will never be recommended in a playlist due to insufficient data. It's an artifact of the popular collaborative filtering recommender algorithm, which Genius is based on.
Their experiments show that automatic recommendations work at least as well as Genius for recommending undiscovered music.
At least as well as never recommending? That is astounding.
Re:Anyone actually use "Genius"? (Score:3, Insightful)
I've made three observations with Genius:
1) It used to be much, much better. Whatever Apple's doing to incorporate new data is having an adverse effect on the quality of the results.
2) It doesn't work particularly well with large libraries. When I upgraded my hard drive, and merged my "archival" collection with the "everyday" music I carry around with me, I found that the quality of the genius results seems to have deteriorated, even though it doesn't necessarily choose any songs from the huge pile of jazz and classical that I added.
3) Genius seems to ignore album tags. If I have two copies of the same song (as happens sometimes, as I like collecting live recordings and radio sessions), Genius seems to pick whichever song is alphabetically first. This can be annoying, as it prevents certain songs from *ever* appearing.
Doesn't work for me (Score:3, Insightful)
I've played with Pandora radio a bit before, but it doesn't tend to work for my taste in music. I like songs with funny lyrics... Monty Python, King Missile, Nellie McKay... the algorithms that focus strictly on musical styles generate pretty hilarious results, but not what I'm looking for :P
I've been pretty happy with some of the dj internet radio stations, though, like somafm.com and some of the ones on di.fm . Before I found some of those stations, I didn't really think there was much of any music that I liked. Certainly not on broadcast or satellite radio.
Meaningful data first play, bs on repeats (Score:2, Insightful)
It's nice to be important but it's more impo...etc (Score:3, Insightful)
Please read the GP. He's talking about the difference between Miss Kittin and Scooter. Scooter! There's a difference between "broadening your taste" and licking out a septic tank.