Friday, January 07, 2005

Zen and the art of Amazon recommendations

Chris Anderson at Wired interviews Jeff Bezos. Jeff had some interesting comments on Amazon's personalization:
    We not only help readers find books, we also help books find readers, with personalized recommendations based on the patterns we see.

    I remember one of the first times this struck me. The main book on the page was on Zen. There were other suggestions for Zen books, and in the middle of those was a book on how to have a clutter-free desk.

    That's not something that a human editor would have ever picked. But statistically, the people who were interested in the Zen books also wanted clutter-free desks. The computer is blind to the fact that these things are dissimilar in some way that's important to humans. It looks right through that and says yes, try this. And it works.
What makes the recommendations so effective is that they're non-obvious but still relevant. Recommending other books with Zen in the title is obvious and not particularly useful. Readers could easily find those books themselves.

Recommending a book on simplifying your work environment is surprising and interesting. It's a book that would have been difficult to discover on your own.

But Jeff's comments might overemphasize the difference between humans selecting the recommendations and a computer selecting the recommendations. The computer merely does an analysis of what humans are doing. It's going out to the community and saying, "What other books do you like?"

It's as if everyone who bought that Zen book e-mailed you recommendations on other books to buy, but all the wisdom of this crowd is gathered automatically and with no effort from the community. It's easy to use. It requires no effort. It just works.

No comments: