The book in question can be found here, and the relevant chapter downloaded here.
In 9.1.1, it shows an example of what it calls a "utility" matrix. It posits some hypothetical user ratings for movies, then explains how an algorithm attempts to use them.
I can't quote the matrix itself effectively here, but it's worth downloading the chapter to have a look at it. Basically, it shows that in a recommendation system (like Amazon's, BN's, etc.), you have a massive database of user-book rating pairs, almost all of which are blank. The recommendation system looks at what the user has rated favorably, then looks around for similar users and tries to see what they've also rated favorably (it probably also takes other factors, like genre, length, etc. into account as well).
Notice that most user-movie pairs have blanks, meaning the user has not rated the movie. In practice, the matrix would be even sparser, with the typical user rating only a tiny fraction of all available movies. The goal of a recommendation system is to predict the blanks in the utility matrix.
Put pretty simply, that's where recommendations come from.
And that has driven home for me the importance of those star ratings in a new way. They're necessary for two things:
1.) They allow the algorithm to identify readers who haven't read your books but whose tastes are similar to those who have.
2.) They nudge the algorithm toward recommending other books by you to people who have given a high rating to one of your books in the past.
And that's on top of all the people-related reasons that star ratings matter.
Another salient quote:
In most applications, the recommendation system does not offer users a ranking of all items, but rather suggests a few that the user should value highly.The way I read this, it helps explain in a new way why it's so important to connect with highly active readers. Let's say you have a reader who's read every bestselling fantasy novel that came out in the last year and rated it on Amazon. They also read yours. They like it enough to give it 5 stars. The algorithm can match a very high number of other readers to that reader ("Oh, this person liked Republic of Thieves and Soulwoven? Maybe all these other people who liked Republic of Thieves will like Soulwoven too). That doesn't work nearly as well if the readers who rate your book only read, say, three books a year.
Final quote and thoughts:
An extreme example of how the long tail, together with a well designed recommendation system can influence events is the story told by Chris Anderson about a book called Touching the Void. This mountain-climbing book was not a big seller in its day, but many years after it was published, another book on the same topic, called Into Thin Air was published. Amazon’s recommendation system noticed a few people who bought both books, and started recommending Touching the Void to people who bought, or were considering, Into Thin Air. Had there been no on-line bookseller, Touching the Void might never have been seen by potential buyers, but in the on-line world, Touching the Void eventually became very popular in its own right, in fact, more so than Into Thin Air.I have spent a lot of time wondering what, exactly, happens when a book suddenly takes off on Amazon. It's a common enough story: "I spent three months or so seeing a few sales a week. Suddenly, in March, my sales took off. I have no idea why."
Maybe this is why.
.jpg)
.jpg)
