Tuesday, November 23, 2004

Artist similarities in music

Brian Dennis points to an interesting paper by Brian Whitman and Steve Lawrence, "Inferring Descriptions and Similarity for Music from Community Metadata".

If that title didn't turn you off completely, the paper does have an interesting idea. Basically, they mine text in web pages, discussion groups, and blogs (which they call "community metadata") to discover information about music artists. They extract phrases from the community metadata and use it to find relationships. Because they analyze the web pages and discussion groups continuously, they claim to be able to capture short-term trends, like a groundswell of buzz around a particular song or artist.

This idea of extracting data and relationships from community metadata is clever. AllConsuming.net is an interesting example of this for books. It "watches weblogs for books that they're talking about". Memeorandum is an interesting example for news. It watches blogs to see what news articles they are talking about.

By the way, one of the authors of this paper, Steve Lawrence, is now at Google.

No comments: