The Magical Mystery of Recommendation Engines June 21, 2006Posted by David Card in Media.
Today Amazon recommended that I buy a Kanye West album because, according to the email, I bought the latest Franz Ferdinand, Kaiser Chiefs, and The Go Team. It also recommended the latest from Yeah Yeah Yeahs, based on Franz, the Go Team, and M.I.A. I am totally puzzled by trying to read between these lines on the connections.
Okay. Franz and Kaiser Chiefs were both top alternative hits that had some mainstream crossover. MIA and the Go Team are hits among the fairly trendy. But I bought both the latter as gifts — I bought them for myself previously, but not at Amazon, so Amazon doesn’t know that. (I rarely update my recommendations with feedback.)
I like the new Yeah Yeah Yeahs a lot, but I bought it weeks ago, though not at Amazon. It’s perfectly logical that it was recommended based on my buying habits — YYY is also an alternative hit with crossover. I am distinctly not a Kanye West fan. Also, roughly the same collection of purchases resulted in recommending The Strokes, another alterna-crossover, while an Elvis Costello/Franz combo produced the unlikely recommendation of the kind of old, but latest, and distinctly untrendy Rolling Stones album. Neither of which I liked much, or bought.
Sigh. Spare the robots and spoil the editors. And how ’bout something I wouldn’t have thought of?
In this particular case, collaborative filtering comes up small for my slightly hip tastes and frequent purchases. You’d think Amazon would have enough customers to do a little better by me. But it’s partly because I buy across physical retail, online, and digital, and they don’t have a clue what I actually listen to.
You can do the links yourself — I’m on a laptop today, which makes blogging oh so much harder. In order of preference, for these albums if not artists: MIA, Kaiser Chiefs, YYY, Costello, Franz, Go Team.