After reading and heavily agreeing with this post, I came to the conclusion that either goodreads is not really trying, or -more likely- the data is just not good enough to make decent recommendations. There are so many biases in the review data that are impossible to fix in any kind of sparse matrix recommendation algorithm. For those who want to try anyway, it might be worth downloading an existing dataset (1) (104 million reviews) and try, before worrying about scraping and api limits.

The only solution (in my experience) is to get some other way of quantifying content, like Spotify does by manually labelling tracks. After some ddg I found storygraph (2), which does this. Its search engine is quite impressive, might be worth trying.

[1] https://sites.google.com/eng.ucsd.edu/ucsdbookgraph/home [2] https://beta.thestorygraph.com