The matching is done by finding similar genres. Each book has a genre-probability vector. I then measured the pairwise distance between all the books for each g

The matching is done by finding similar genres. Each book has a genre-probability vector. I then measured the pairwise distance between all the books for each genre-probability vector (~3 million books, so this takes some time). The book suggestions you see are the top closest matches to that particular genre-probability vector. The searching is done with elasticsearch and the backend is Flask.