music recommendations: about

the data

Based on playout data for Radio1, 1Xtra, Radio2 and 6Music from September 2007 until mid-June 2008.

See BBC Audio & Music Interactive at Mashed 2008 for more details.

the recommendations

Based on Latent Semantic Analysis, a technique used in Information Retrieval.

Build a term-document matrix based on the artist play per brand: artists are terms, brands are documents.

With latent semantic indexing, we map this data into a n-dimensional space that let's us:

• recommend brands/artists based on another brand (example)
• recommend brands/artists based on another artist (example)
• recommend brands/artists based on a last.fm profile (example)

the recommendation engine

Uses the Semantic Space engine developed at the University of Southampton by Jonathon Hare.

the code

Hosted on github: http://github.com/metade/music-recommendations