It’s an interesting challenge. Modern recommendation systems grew powerful because of enormous amounts of instant feedback. You can capture clicks and view time on the web. You don’t get that in books.
I see three possible solutions:
1. Google approach: scrape the web for book recommendations and somehow create an ML recommendation system that’s better than Goodread’s
2. Pandora Radio approach: (semi-)manually create classifiers for books (genre, tone, character traits, etc.) and build a recommendation system with that.
3. Practical approach: find book reviewers whose opinions you trust and follow their recommendations.
I thought https://book.sv was pretty good. It was on HN recently: https://news.ycombinator.com/item?id=45825733. When I inputted 5 books I liked, the recommendations were a combination of:
1. books I had already read and enjoyed before
2. books that were already on my list (either from friends or other recommendations)
3. books I hadn't heard of
That said, I haven't read a book from #3 yet, so I can't fully vouch for it, but #1 and #2 are positive signals to me.
It’s an interesting challenge. Modern recommendation systems grew powerful because of enormous amounts of instant feedback. You can capture clicks and view time on the web. You don’t get that in books.
I see three possible solutions:
1. Google approach: scrape the web for book recommendations and somehow create an ML recommendation system that’s better than Goodread’s 2. Pandora Radio approach: (semi-)manually create classifiers for books (genre, tone, character traits, etc.) and build a recommendation system with that. 3. Practical approach: find book reviewers whose opinions you trust and follow their recommendations.
Paste in last 10 reviews to Gemini or gpt and ask for 20 "rare-gems, unique and exquisite," with descriptions. Works well
Once you cross 100 books its all repetition. Just like HN comments.