Cluster Analysis on musical data from Last.Fm
| On the left a graph of 388 users of www.last.fm is shown. The users are selected by
recursively descending the list of neighbours that last.fm computes for each user.
A measure of similarity is computed by last.fm for each neighbour.
The algorithm of this similarity computation is unfortunately not disclosed.
The graph shows a traversal of 2 levels deep, starting from myself as root, but selecting
only neighbours with a similarity higher than 50 (similarities range from 0..100). The
result is a list of 388 users. The outgoing edges from each user in the graph point to the
3 most similar neighbours.
By computing the optimal modularity the graph has been split into 6 clusters. For each cluster
the table below shows the 10 most played artists/composers.
|