“If you like X, you might like Y.”
So say today’s recommendation algorithms, which guide our movie rentals, the music we buy or stream, the books we read, and so on.
Recommendation is big business, because it takes the data trail we leave behind as we wend our ways through the world of media, and turns that into something decidedly less abstract: cold hard cash. The better these algorithms understand us, the more we stick to a given service — and the more we tend to buy from it, which is why Netflix considered it a bargain to offer $1 million for the best improvement to its movie recommendation algorithm.
Paul Lamere, director of developer platform for The Echo Nest (publisher of Evolver.fm) put Apple Genius and Google Instant Mix through a number of empirical tests, comparing their results with The Echo Nest’s own playlist engine, to see which one provided the best recommendations based on his own music collection.
Lamere’s analysis (referenced on Techmeme) takes a novel approach: looking for “wtf” inclusions — those recommendations that appear to have no basis in reality. In a nutshell, he found that iTunes Genius and The Echo Nest’s playlist engine (.pdf) performed admirably, while Google Instant Mix outpaced those two by a long shot, offering the least-sensical recommendations.
It’s an entertaining read, and makes a compelling case that Music Beta by Google needs to improve its new Instant Mix feature significantly, because right now, it’s spitting out baloney.