Amazon is the great mystery of our time. They are so secretive about their algorithms and inner workings that it's very difficult to tell why certain books launch to the top while other possibly better books languish in the nether regions of their rankings. I'm here today to shed some light on at least one part of the mystery, but bear with me... this is going to get technical. VERY technical.
If you're not sure what I'm talking about, you should stop reading right now, go to Amazon, and look at any page, even their home page. Everywhere you look on Amazon, you'll see "More Items to Consider", "Customers Who Bought This Item, Also Bought...", "Inspired By Your Wishlist", or any number of other recommendation lists that they feed you while browsing. These lists are where you want your book to be.
"Why?" you ask? Because these are going to be some of the best recommendation lists that you will ever encounter. Seriously. The technology that goes into creating these lists is unmatched by any other site in the industry, except maybe Netflix. These lists will sell your books, I guarantee it. You just need to get your book there.
Oh, sure. Is that all?
Thing is, Amazon wants to sell your book. They want to sell everyone's books. They want to sell millions and millions of everyone's books. Know why? Because they make lots and lots of money. How do they make sure they sell the maximum number of books possible? By creating really, really good recommendation lists. Don't believe me? They're not the #1 book seller in the world for no reason.
Item-to-item Collaborative Filtering
Holy crap. I hit you right over the head with those great big words. Here's where the technical stuff starts. Put on your hard hat, and watch for falling math.
Amazon has some of the biggest data stores in the world. They keep track of everything. Every item that you've ever purchased, when it was purchased and for how much. Every item you've ever clicked, what you clicked after that, how often you visit certain pages, how often you purchase an item after visiting the page, and how many times you visited the page before you bought it. Thing is, Amazon is greedy. They don't like to share data, because that's their competitive edge over everyone else.
I don't blame them.
Their recommendation system is based on one thing and one thing only, and it's really very simple when you tear it down to just this: who bought what items. But they don't even care about who bought what, all they really care about is how many people bought the same two items.
Here is the basic algorithm, in all it's glory:
For any Item (Item_X), compute a list of items that have the most in common, according to sales.
This list is computed based on something called a Commonality Index (CI). The CI of any two items (Item_X,Item_Y) is calculated using the following equation:
CI(Item_X,Item_Y) = Ncommon / √(Nx * Ny)
Take 3 items:
Item_X = purchased 300 times (Nx)
Item_Y = purchased 200 times (Ny)
Item_Z = purchased 30000 times (Nz)
Find which item (Y, or Z) has the most in common with Item_X, where:
Ncommon(Item_X, Item_Y) = 20
Ncommon(Item_X, Item_Z) = 25
CI(Item_X,Item_Y) = 20 / √(300 * 200) = 0.08165
CI(Item_X,Item_Z) = 25 / √(300 * 30000) = 0.00833
Item_X has MORE in common with Item_Y than Item_Z, even though Item_X and Item_Z have more common purchasers.
This is important. It means that you have no immediate hope of being paired with Item_Z until you have a fair number more purchasers in common. Want to be recommended to people who buy the big guns at the top of the list? You need to get MORE people who buy the top books to also buy your books. In this case, Item_X and Item_Z would have to have almost 250 common purchasers in order to top CI(Item_X, Item_Y), and get recommended over it.
But Wait, There's More!
It's still not as simple as just a single score, and you had to know it wasn't. So what's the catch?
Your sales have an expiry date. That's right. When working out the recommendation list, Amazon only considers the last six months' worth in sales data. By doing this, they keep the recommendations fresh. They don't sell books by recommending what was popular last year. They want to show you what's popular now.
The rest of the recommendation lists are figured out based on what you've looked at recently, what you've bought recently, what's currently in your cart, what's currently in your wishlist, what products you've marked as "liked", and what products you've reviewed. Each of these factors produces a different list of items based on the same equations above.
Lighting the Fire
So how do you take this information and use it to light the Amazon Recommendation fire?
Two Words: Targeted Marketing.
Most people don't know what that means. Do some research, and get started today. I'll do that as well, and will bring you another post later on, all about it.
I hope this helps at least some of you out there. If you're looking for a way to help me out and spread the word, clicking the Tweet and +1 buttons below are quick and simple ways that I would greatly appreciate!
Thanks for reading!
I'm always interested in hearing what you have to say. Contact Me, I'd love to hear from you.
Don't forget to join in on the conversation in the comments section below.
17 Comments(103 Pending Approval):
Back To Top