The Trouble with Automatic Recommendations

Last week I was chatting with a coworker who said that she loved the Harry Potter books, but otherwise didn't read much fantasy, and she asked me what she should read. It got me thinking: there ought to be a Harry Potter read-alike list for adults. I'm hoping to make that the subject of my next post: a list of books for the adult Harry Potter fan.

I'd love to get suggestions, but I warn you: this will not be a list of books about kids learning magic. I want a list of books that capture the wonder and adventure, the likable characters, the easy readability, and the sense that it's important to do what's right.

Which brings me to the main topic of this post: automated book recommendations. They are terrible. I find the "find similar books" features on NoveList and What Do I Read Next to be utterly useless. (Amazon does a little better, but not much.) And I think I know why.

NoveList and What do I read next make decisions based on topic. They say "Oh, you liked this story because it's set in the deep south. Would you like to read some more books about the deep south?" Or "women detectives," or "coming-of-age stories." The problem is that topic rarely has anything to do with why I like a book. (Rarely, but not never. For instance, I like dragons. I'll read almost anything with dragons in it. And many people have similar affinities. But these affinities aren't a factor in a majority of my reading decisions.)

Amazon, of course, makes decisions based purely on "people who bought...also bought...", meaning that for any genre bestseller, the recommendations will consist of the rest of the bestseller list. Not helpful.

Our reading decisions are based on intangibles: tone, character, moral compass, etc. We need a system that reflects these intangibles. And Amazon's failure to produce useful recommendations from its one-dimensional rating system tells me that only a multi-dimensional system will work.

Harry Potter read-alikes for adults will make an appearance next time. In the meantime, another list: the factors that I think define what we like in what we read. (I envision these as sliding scales, not multiple choice.)

Setting and circumstances: realistic or fanciful?

Moral stance: clearly defined vs. shades of grey?

How many POV characters?

Pace: quick and easy, or slower?

"Fun" book vs. "serious" book?

Characters are motivated by: duty, revenge, survival, logic, emotions?

Characters are: likable and easy to identify with, or complex and flawed?

I know that's not everything, and I suspect I'll be adding to this list as time goes on. Feel free to offer suggestions!

What I wonder for the long term is, would this be workable as a rating system? Could Innovative, or NextReads, implement a rating system like this and create an automated system that would offer true read-alikes?