The Fault in Amazon's Stars

If, like me, you thought books' star ratings on KDP were pure mathematical averages of review scores, Amazon is here to correct that assumption.

Amazon calculates

You can find this message by going to a book's Amazon page, clicking on the number of reviews, shown next to its star rating, and hovering over the rating where it's spelled out in text.

Hobbit Rating 1

Hobbit Rating 2
Some have argued that Amazon's delegation of its star rating system to machine learning helps mitigate the effects of fake reviews.

First, SJWs have moved on from spamming hatemob victims' books with 1-star reviews--not because the algorithm blunted their effectiveness, but because the account-level penalties already in place for tampering reviews deterred them.

And from personal experience, the review algorithm isn't too good at spotting attack reviews; not even when the review itself admits the reviewer didn't finish the book--which KDP knows.

Second, anyone who's comfortable placing the public perception of his work in the hands of machine learning shouldn't be allowed outside without a leash.

Last but not least, people who think a company that censors authors for ideological reasons can be trusted to impartially adjudicate "factors that establish reviewer trustworthiness" should share their drugs. I hope they brought enough for the whole class.

A generation of authors--myself included--flocked to Amazon, drawn by the promise of liberation from tradpub's tyranny.

Now it appears we've fled a corrupt, decrepit aristocracy for the cruel arms of a tech Napoleon.

It's not just authors who suffer. Amazon has the world's third biggest search engine, but their algorithm actually makes it harder for customers to find the specific products they're looking for.

Amazon has drifted from building the Everything Store to building Skynet. Readers and authors need a better online marketplace. Hopefully the current round of antitrust investigations will clear the way for a web retailer that's primarily interested in making money.

It may be sub-optimal, but Amazon is all we have for now. The best way for readers to support authors they enjoy is still buying their books and leaving reviews.

Buy my latest mecha thriller Combat Frame XSeed: Coalition Year 40. Or if you've already bought it, leave an honest review. Every little bit helps indie authors like me keep bringing you the fast-paced action you love.


  1. This sounds like what they did with the Hugos. They created an algorithm that could supposedly read the minds of the voters (sounds like the plot of a Campbell era short story) to tell whether they were voting for the "right" reason. At least Amazon has a few valid inputs like did the reader buy/read the book. The tru-fan types are just going on the idea that a lot of people liking the same books (that they don't like) constitutes "wrong think" and should be discounted. In both cases the process is opaque enough that the user can manipulate it to their advantage with little fear of detection.

  2. Have a look at Malak, short story at the link, for an interesting (pessimistic, because Peter Watts, of course) tale of machine learning...