I was browsing through some books at Amazon today and saw a field I’d never seen before: statistically improbable phrases (SIPs). These are phrases that occur frequently in one particular book and infrequently in others, and thus likely to be "interesting" or "distinctive". As an example, the Amazon page for "Finding Your Own North Star: Claiming Your RIght to Live the Life You Were Meant to Live", by Martha Beck, includes the following SIPs: your emotional compass, your social self, your essential self, your internal compasses, your true path, turtle steps, emotional shrapnel, your emotional wounds, emotional compasses, divine decadence, catalytic event. Clicking on "your emotional wounds" brings up a page listing 17 other books in which that phrase occurs at least once.
I’ve always been fond of turns of phrases, and so I like this new SIPs feature (though it doesn’t appear to be available all too widely yet — seems to be limited to those books for which an exerpt is included). I suspect I’d be more likely to buy books in which the author has coined some evocative phrases than a book in which no phrases — or none that evoke anything intriguing (for me) — are listed. As far as I can tell, this is the first time that Amazon has expanded its recommender system beyond collaborative filtering — where recommendations are based on people who have bought similar items to what you have bought — and into (or at least toward) content-based filtering — where recommendations are based on the actual content of the items that you have bought. It will be interesting to see how this new feature affects customers’ experience (and Amazon’s sales). If nothing else, it adds yet another cool feature that distinguishes Amazon from its competitors.
Comments
One response to “Amazon’s Statistically Improbable Phrases (SIPs): Potentially Useful Curiosities”
SIPs may be like fingerprints.
They may also be useful for clustering related books, especially in the Long Tail of books purchased infrequently or by buyers who don’t buy many books from Amazon.