Yeah, Amazon's seems to be based off of customer purchase history and some tagged metadata, but almost always just ends up recommending the most popular book by sales (which are not necessarily the best).

Short of some infeasible brute NLP of books, I can't really think of ways to do this in an automated fashion, outside of gamifying data input from users.

You can submit citation data for any written work (including books or scholarly articles) on Wikidata and view the citation, coautorship etc. graphs on the Scholia frontend. Works quite well in practice.