My tastes are not necessarily similar to my friends. That's actually what makes the email feed cool. I am open to branching out and reading books that I normally wouldn't expect to read.

In terms of recommending what I am already likely interested in based on my previous reads, the "Readers also enjoyed" section seems decent to me. Is that what people think sucks? I've used that a fair amount and found it to be valuable. Or is it the whole "Browse" section that's bad? I never look under there.

I use goodreads every day, but that means spending 5 mins max on it (I hop in, add a book that I heard about somewhere to my to-read list, then I hop off). And, like I alluded to in my first reply, from what I can tell this behavior also holds true for my friends. I don't think I'm an outlier, though I totally understand that people use the site differently than me and they find it wanting.

Goodreads 'also enjoyed' is one of the best implementations of this feature. I always see related books that actually make sense to be shown there.

That's the simplest and most naive implementation of a recommender. It doesn't take into account what I liked about a book and whether the one it's recommending is similar in that way or if I will dislike it for some other reason.

A better way to do recommendations is to find readers who have rated books similarly to me and tell me what books they've rated highly that I haven't already read. This is what Netflix's Cinematch once did and it was good. It used singular value decomposition and other linear algebra techniques on a giant matrix of raters.

See: https://www.netflixprize.com/

I have no expectation of ever seeing useful recommendations from the sites I frequent. This might be heresy on HN, but by and large recommendation engines are pure trash. Have you ever been recommended a useful product on Amazon? One would think that a company with such large resources could occasionally point me to a product I might want.

> doesn't take into account what I liked about a book

I don't remember ever seeing a rating system (for books or tv) that let me do fine grain rating. And if it had it, most people[citation needed] might not use it anyway. Also, how would I say that what I liked the most about a fantasy book is the magic system, for example?

I don't care HOW the recommendation is implemented, but what shows on the recommendation. For me, 'Readers also enjoyed' shows things that are sufficiently similar, and which people that I follow - friends and bloggers - gave good ratings.

The 'lists with this book' section is also good for discovering books, as the lists are user curated, and the title helps me know what's the common theme. Ie: books with cool magic systems.

That's the beauty of this. You don't need a fine grained rating system. By comparing your preferences to those of others, the aspects of what you like and don't like become implicit.

A product has features in some high dimensional quality space. When you rate it, you provide information about your preference for those qualities without needing to explicitly rate them or even have a concept of what they are called.

I'm not sure if I'm the only one, but I don't find the Netflix algorithm useful. With a few exceptions everything is a 97-99% match

That's weird, for me I see 30% and over, although most fall into 60% and over.

I think it could be: you didn't 'dislike' many titles or; you are not browsing things that are too different from what you usually watch

The current Netflix algorithm is useless. Cinematch is what they used to have and they had a prize, which they awarded, for $1M to anyone who could improve it by 10%. They've apparently abandoned it.

See: https://www.netflixprize.com/