Islands of Music

RSS
Recomendar

Mai 20 2008, 14h26

The islands of music playground demonstration is something like a tag cloud where similar tags are located close to each together. The map was created using clustering algorithms.

These algorithms group similar music on islands. Similar islands are placed close to each other. For example, various flavours of metal are located close to each other in the upper right of the map. The map also suggests several more or less continuous transitions. For example, there is a path from folk to doom metal (via psychedelic, progressive rock, and progressive metal).

Another somewhat curious example is the sea of mistagged artist where various flavours of non-English world music can be found. Generally, not all clusters make sense and part of the explanation is that there is plenty of noise in the data.


In more technical terms:

The map is a self-organizing map of 13,000 randomly sampled Last.fm users labelled with tags and artists associated with each user.

Each of these 13,000 users is described with a tag cloud which is extracted from the music the user listens to. This data is normalized in a similar way as described here . One consequence of this is that a large part of alternative indie rock pop is averaged out.

After all the normalization and pre-processing the 13k users are represented by 2000 distinct tags resulting in 13k sparse vectors in a 2k-dimensional tag vector space.

Using singular value decomposition (SVD) the dimensionality is reduced to 120 dimensions. This 120-dimensional space is a latent semantic space in which no distinction is made, for example, between brazil and brasil.

Using k-means clustering 400 prototypical users are computed. Users very close to the zero-vector are not considered for further analysis. (Given the normalization and the latent space mapping, these zero-vector users can be interpreted as either very average users, or so unique that they can’t be described within the 120-dimensional space.)

Using a self-organizing map (SOM) the latent space is mapped to a 2-dimensional visualization space. The SOM has a size of 20 rows and 40 columns. A smoothed data histogram of the SOM is computed and visualized so that clusters show up as islands.
Envios aceitos
Playground

Comentários

  • eltitoskate2

    theres gonna be updates for this? seems quite cool. It would be cool to "zoom" in the islands

    Mai 25 2008, 21h32
  • whym

    totally agree to eltitoskate2. Or, it would be nice if there is a bit more detailed (high resolution) version of the map available.

    Mai 29 2008, 13h56
  • jellevc

    Very interesting work!!

    Mai 29 2008, 15h42
  • bernardoprieto

    It's a shame I didn't see this thing a couple of yeras ago when I was looking for a subject for my thesis... This is really cool. But I just had an idea that might make it more appealing. Imagine the islands animated as the tags and tastes of users change. I guess it would be a slight change between iterations, but in time, it would look like the evolution of the continents or something like that. Cheers!

    Mai 29 2008, 23h06
  • E1i45

    @ eltitoskate2 & whym: a higher resolution map is definitely an option. @ bernardoprieto: maybe in a few years from now we could give it a try :-)

    Mai 30 2008, 2h03
  • TripleF

    This is really awesome and interesting. One thing I was looking for in the demo was the ability to map a given artist on the map to see how it relates to other artists and genres. Great work.

    Mai 30 2008, 9h30
  • namnatulco

    Impressive. First thing I noticed: psychedelic and trance are WAY far way from eachother, but still, / are a mix of those. What I'm wondering is, is the map supposed to be infinite (ie. the bottom and the top connect, as well as the sides), or is this just co-incidence?

    Mai 30 2008, 21h57
  • E1i45

    @namnatulco: any connections between the bottom and the top, or the left and the right of the map are just co-incidences. However, it's one of the parameters of the SOM algorithm that allows generating maps where all sides are connected.

    Jun 2 2008, 10h13
  • Bisto2007

    Interesting stuff - Good work!

    Jun 2 2008, 18h43
  • xe54

    @E1i45 I'd like to get my grubby mits on the science and code behind this! We could pimp it out with some plasma / cellular automation :D What do you reckon?

    Jun 6 2008, 11h52
  • cococinelle

    @jarnik, that already exist... Some kinf users have developped a map when you can see how far or how close you are from your friends! I would like apologise for making difficult the work of algorithms with my non sense tag ^^ I'm really sorry but I won't change! Shame on me!

    Jun 13 2008, 20h04
  • vivrin

    Jun 14 2008, 13h52
  • joshkouri

    Damn, Last.fm is turning me into such a data & statistics geek. I love this stuff.

    Jun 14 2008, 19h31
  • wraith79

    Hi, this is quite interesting, however the map might be better with a third dimension. This could be hard to visualize, but a third dimension might offer better insight about cross-genre music. Or something like that.

    Jun 15 2008, 20h10
  • Andrew2119

    @ namnatulco - most of the music tagged as psychedelic is psychedlic rock, not psytrance.

    Jun 15 2008, 23h56
  • pandoras_locks

    very interesting would love to see this as a permanent fixture!

    Jun 16 2008, 13h41
  • Selenaru_Negrea

    I didn't seen it so far. Anyway, I think it's cool to be able to establish cities on these islands and share their location...

    Jun 19 2008, 7h57
  • Tueksta

    hmm, why does hip-hop appear twice? is it a distinction like west-coast and east-coast, or old-school and modern? Would it be possible to show, which squares have data and which haven't, without having to hover with the mouse first? wouldn't it be smarter to completely exclude comedy, soundtracks, sampler and mistagged artists? what do you think about a 3-dimensional version? is it smart to completely rely on user's tags? just some thoughts... :)

    Jun 23 2008, 9h01
  • Tueksta

    i wonder what this map would look like, if you wouldn't use tags, but pandora.com's dimension-set for songs! :D

    Jun 23 2008, 9h05
  • vesselinux

    could you map this music map onto the geographical map? might be interesting to see what type of music is dominant in which part of the world.

    Jul 8 2008, 16h46
  • Star_waterfall

    Awesome stuff! 2 vesselinux: Very, very interesting idea!

    Jul 8 2008, 23h38
  • henry25

    Interesting :) But it should be bigger with more names of artists. Or maybe with other maps of continents, countries etc. :)

    Jul 9 2008, 18h18
  • deuxpi

    Would it be possible to project an arbitrary user's tag cloud onto the model vectors and obtain a "You are here" point? It would also be a nice addition to the taste-o-meter. I hope not everyone lands on the "rock" island!

    Jul 22 2008, 3h24
  • beeka_uk

    Any chance of creating a playlist by mapping a journey through the islands?

    Ago 8 2008, 1h35
  • Kimik0

    J-pop is between Reggae and Deathcore? O_o

    Ago 28 2008, 21h34
Ver todos os 36 comentários
Deixe um comentário. Faça login na Last.fm ou cadastre-se agora (é gratuito).