The World of Music, originally uploaded by densitydesign.
We began using the ratings from 150,000 users from Yahoo! Music LAUNCHcast. Users rate music on LAUNCHcast using an intuitive interface. The ratings then influence what music their customized radio station plays.
Afterwards, we used the user ratings to compute the similarity between the artists. Essentially, for each pair of artists, we counted the number of users that rated both artists highly to determine the similarity. This operation gave us the artist-artist similarity graph. In the image below, we show the clustered adjacency matrix of the graph. The blocks you see in the middle of the picture are similar groups of artists. (In fact, this is how we got the color for each artist in the final picture.)
To determine the layout, we used an optimization algorithm based on semi-definite programming to compute a layout of the graph on a sphere. The layout algorithm tries to find good ways to split the data and naturally separates different groups of music. (In this case, the most natural separations are musical genres!)
Once we had the layout, it was constrained to lie on a sphere. We then “unrolled” the layout to get the image above. (That’s why some of the “routes” look like airplane routes — they are great circles!)
