home ¦ Archives ¦ Atom ¦ RSS

Whitman & Lawrence: Mining Music Metadata

Via plasticbag is an oldie but goodie that's been stashed in my aggregator for a bit. In the 2002 International Computer Music Conference, Brian Whitman and Steve Lawrence describe a scheme to determine artist similarity based upon community metadata.

In a nutshell they take an artist name, ship if off to a music search engine, mine the top 50 pages for features using NLP techniques, and then cluster based upon the features. Evaluation is done using a "ground truth" of human compiled similarity lists.

An interesting approach to constructing context without much explicit, machine readable information. Parts of this are probably applicable to analyzing the blogospheres.

© Brian M. Dennis. Built using Pelican. Theme by Giulio Fidente on github.