Show HN: Personalized book recommendations with Librarian AI
librarian-ai.com
librarian-ai.com
The thing that I was trained to do when I was a librarian, which I never see in recommendation engines, is to ask the question "what did you like about this book?", or "why did you like this book?", or any number of related questions that get to what you really want. Understandably, an algorithmic approach usually assumes the book is a single unit that is liked (or not) by others, and can thus be linked to other things people like on a fancy graph. This is useful because you don't need a granular understanding or human-level comprehension of the books. But, it tends to make for more generic, less interesting recommendations.
Sometimes the answers people give to these questions are really interesting: "there was one scene that really reminded me of a canoe trip I took with my dad" not just "the user wants [books with canoes]". At present, you need to be human to actually understand that it's not the canoe that matters, it's something else, and the right recommendation may be a book that appears completely different on the surface.
- Plot
- World Building
- Character
- etc (There's a specific list mentioned, but I can't recall it)
And in which other sites you input a list of books and get a list of recommendations?
Like this one: anonymously, directly?
They're anonymous-ish (in that you don't have to be logged on, though I'm sure tracking cookies gunna track), and they're direct, but they're only for one book at a time.
Book Title search seemed like it could use some kind of "not found, but..." Helper, or suggestion functionality.
I searched "Moscow Calling," "Remembered Prisoners of a Forgotten War," and "Nixon at the Movies" with no title results. Huckleberry Finn worked OK.
Would use it again though, thanks for dropping a link in the discussion.
We don't have those books in our system yet as no author has recommended them is why nothing is showing up. We are still pretty limited as we have interviewed 8,000 authors in different areas, but that isn't a lot in the grand scheme of books. Slowly getting in more every week :)
For example, I like Bill Bryon and have read most everything he has written. He's good at making subjects I am not interested in captivating. For example medical history and anatomy in his book The Body.
I would love to read other books that make the uninteresting interesting but when I put The Body into Shepherd (or most recommendation engines) I mainly get other books about anatomy. This makes sense on the surface, but I've already read a book on anatomy. I'm ready for something else interesting.
I realize that might not be very actionable or useful, but when I am looking for something new to read I don't know what I am looking for. I just know the "vibes" books and authors I like, and often find that recommenders default to similar subjects only. Maybe this is difficult or impossible to capture, but it's been nagging me as something missing.
Yep, I totally understand what you are saying, and that is what I am working toward. It is a challenge.
Right now, I ask authors to share 5 books they love around a topic/theme/mood. So I am collecting a human grouping of books similar in some way that humans decided along with why they love it.
For example, if you like "Notes from a Small Island" or "Down Under" (two of my fav Bill Bryson books), here are books that authors have grouped with it and also loved:
https://shepherd.com/books-like/notes-from-a-small-island https://shepherd.com/books-like/down-under
So I am using human groupings to help discover books that humans associate with his work somehow. And above each book, their list title helps you understand how their mind grouped it.
For example, "Into The Wild" is associated with it based around a human grouping books that capture the spirit of Jack Kerouac's On The Road.
What do you think?
What is next to improve this?
I'll add filters to help people hone in on why they loved the book to help pull that thread out and show books along those lines. For example, if I liked Catch 22 for its absurd humor, I want to make that the common denominator on the page. Or if I like "From a Small Island" for the humor and history, picking those to try to find books with a similar book DNA.
In about 4 to 5 weeks I'll ship the first version of this on the bookshelf pages. And in maybe 12 to 16 weeks on the books-like pages, as that needs a lot more work.
And, I am hoping next year to launch Book DNA, where I pull in reader data and help line up books that are your all time favorites with others to map this out a bit further. Slowly getting there and a lot will dependon collecting more data from readers in 2024.
I like the idea of pulling as many threads out of a book to drill into why you might like something. Especially if those threads are not necessarily related to each other as that should widen the recommendation pool.
I still think there is something worth doing with author links as well. I played around with gnooks.com after someone else posted a link and within a few minutes found authors I had never heard of that had books that looked interesting.
Yep the author links is coming as well, I've been working on some basics there.
https://www.amazon.com/Financial-Shenanigans-Fourth-Accounti...
http://portal.som.yale.edu/sites/default/files/files/MGT%208...
Ya that is harder topic, I am working on more around fraud, but here are some that might be good: https://shepherd.com/bookshelf/fraud https://shepherd.com/bookshelf/confidence-trick https://shepherd.com/bookshelf/money-laundering
In about 5 weeks I'll have filters to help you hone in from those to a specific topic.
You also might like this list on money laundering from Monica, she is a world expert on the subject and did a fascinating list on the subject: https://shepherd.com/best-books/tackling-money-laundering-ri...
> Prompt: give me some books similar to "Three Body Problem", "Brave New World", and "Dune" - but dont give me the common suggestions, give me some more esoteric stuff i probably haven't heard of
> I understand that you're looking for lesser-known, yet thought-provoking science fiction novels similar to "Three Body Problem," "Brave New World," and "Dune." Here are some esoteric suggestions that you might not have heard of:
> "Blindsight" by Peter Watts - This hard sci-fi novel explores the nature of consciousness and humanity's first contact with a terrifyingly alien intelligence.
> "The Quantum Thief" by Hannu Rajaniemi - A post-human heist story set in a future Solar System, where advanced technology has given birth to an array of strange and wondrous societies.
> "The Left Hand of Darkness" by Ursula K. Le Guin - A groundbreaking work of feminist science fiction that explores gender and societal norms on a planet where inhabitants can change their sex at will.
> "The Stars My Destination" by Alfred Bester - A classic sci-fi novel that combines elements of revenge, teleportation, and corporate intrigue in a richly imagined future world.
> "The Dispossessed" by Ursula K. Le Guin - A thought-provoking tale of two societies, one anarchistic and the other capitalistic, and the physicist who attempts to bridge the divide between them.
I can import the same echo chamber anywhere I go
I want to hop dimensions, we dont have a shared reality and are coddled with a familiar soma just like in brave new world
Honestly the best I’ve seen for this are website that just let you browse YouTube and social media sites under a different users session
https://help.goodreads.com/s/article/How-do-I-import-or-expo...
I don't think its recommendation engine is anything sophisticated, it's just working with really really rich data.
I just checked, and it says it's recommending a book to me (Singularity Sky) based on five other books I own (Engine City, Spin State, A Deepness In The Sky, Recursion, The Skinner). But as it's always "based on these five books", I have to assume it's doing a lot more matching than that.
My point is that recommendations that are based on social networks become less and less useful as that network grows. And LT seems to have hit the perfect size for its recommendations to be useful. I'm actively dreading the day that they either get worse (i.e. the network gets too big), or that LT goes away entirely because it has no place in the modern internet anymore, multiple times a week, to the point where I'm thinking about ways to preserve its current form in some way or another.
If more people would be on there it wouldn't be as useful too anyone anymore, instead of being as useful to more people. So instead of gatekeeping it's more like preservation of something that would otherwise vanish. Librarything is basically one of the few national parks the internet has left.
Finally, and maybe most importantly, my original comment was really meant as a light-hearted comment on the magic of LT with some of the thoughts I have on the matter sprinkled in. But I should've known that HN would interpret it in the least favorable way.
Searched for 1.The flute and flute-playing : in acoustical, technical, and artistic aspects, 2. An essay on the construction of flutes, giving a history and description of the most recent improvements, with an explanation of the principles of acoustics applicable to the manufacture of wind instruments 3. A treatise on the flute
It suggests Beauty and the Beast and Other Fairy Tales The Flower of Paradise Songs for Bass in a Comfortable Range Philip Steele of the Royal Northwest Mounted Police and The Flute Book (this is the only relevant result).
The recommendations seem to get better with every author you rate.
I kind of expected that on my first sample set (Ishmael/Accelerando/A Gentleman in Moscow), but even the second one (Priory of the Orange Tree/Traci Lords:Beyond it all/Civilized Man's Eight Deadly Sins) is severely tilted to "everybody is reading it!"
A last try with a decided "60's Scifi" slant - Small, still voice of trumpets/The Tomorrow People/Deathworld. And I get Hitchhiker's Guide, War of the World's, Rainbows End.
I don't look for book recommendations to learn what everybody is reading, those books are inescapable anyways. I'm looking for thematically relevant, but might surprise and delight me.
Doing this well is hard even with curated lists and experts, but it's definitely possible to do better by just weighing popularity less heavily.
EDIT: The recommendations still feel less like AI and more like k-means clustering.
Interestingly, I asked ChatGPT for recommendations based on the same list and it gave me a very similar result, including matching the top two suggestions! Coincidence? (I added a note that I was interested in finding new authors and it gave me a much better list.)
To further prove my my point, I've just connected to a Singapore VPN, refreshed the page, and all book titles switched to English. When I disconnected from the VPN and refreshed again, the titles reverted to German.
No details in the page, but since this submission is a "Show HN", delta234 must be the bridge...
...so not too sure about it
What's the underlying algorithm?
A veteran for movie recommendations: https://movielens.org/
Though as always, the problem with book recommendations is not that it's hard to find books about a specific topic, the problem is finding books that are a) good and b) different enough from other books I've read about a topic so as not to be very repetitive.
If this could solve a) I'd be very happy, though I cannot see how it would accomplish that. The best attempt at solving this for myself was just going by book recommendations from actual people, but even that is hit or miss. Many people recommend books about a topic on which they've only read that one book and so they don't actually have anything to compare their book to. A book recommendation like that doesn't contain information about how good a book is, just if the one who read it liked it.
So in my opinion this is actually a pretty difficult problem and I'm not sure throwing an llm at it (I'm assuming that's what the ai in the title refers to) will solve it.
Still, it's a neat project.
A related project is https://shepherd.com - but it focuses on high quality human suggestions.
Like, I get that the type of guy who likes Infinite Jest would probably like the three James Joyce books recommended to me, but thats more of an Amazon recommendation than an informed AI one. Its also surely the case that an IJ fan is aware of Joyce already!
I tried to be more obscure and personalized: 2666 by Robert Bolaño, The Dissertation by RM Koster, and Notable American Women by Ben Marcus (which is a novel despite its title). I got back a David Foster Wallace book, another Bolaño book, a book of short stories that is maybe actually interesting but very different from my input, and a homemaking book about reducing waste in your home.
Kinda seems like a waste of energy hooking up AI to this, you could just scrape Goodreads and get the same quality output.
I don't think that everything needs to be monetised. On the contrary - there's too much stuff existing merely to just extract value from users. But, the recommendations I've seen here look really good and I'm already ordering one of the books (the other one I've already bought).
It seems like it needs an option to thumbs up/down the responses so that the system can continue to hone in on books that are new to me.
Immediately had the same experience: input a Theroux, Bill Bryson and Bruce Chatwin into the engine, as I've been longing for some new travel books to read, but naturally it just gave me... five or six other books by Paul Theroux and Bill Bryson.
Definitely needs more input parameters to be useful.
You begin by entering what you want to get recommended, and some examples of what you like and don’t like. After this it lets you swipe cards with suggestions to like/dislike them and further train the program.
https://apps.apple.com/nl/app/ai-suggester/id6447003073?l=en
The Road -> Same Author as Blood Meridian and I guess Mccarthys most mainstream successful book?
Barbarian Days -> ???
The World according to Garp -> Same author as Cider House
The Devil in the White City and The house on Mango Street are the two recommendations I don't know which also don't seem too obvious. But on first look both books don't appeal to me.
FWIW I've noticed I tend to gravitate more to authors than specific subjects, and finding a new author whose catalog I can mine is somewhat rare.
Thank you!
I tried asking chatgpt, but it’s too much data for it to digest. And from what it can digest it gives me some real obvious and generic recommendations.
I prefer Readow.AI for instance because it recommends a lot more books, it's quicker, and the books it recommends are occasionally ones I've never heard of (and thus more valuable to me).
Though the auto-complete kind of went off my phone screen, I knew I should scroll down to check.
Thanks for sharing it, will use again for sure.
I just found the botany of desire - next book I think.
And how would that work?
Is it like "Given preference for A, B and C, suggest more"? One would like to understand what makes (specifically to this task) such box a good engine.
(This is for GPT-3 and ChatGPT. Haven't tested GPT-4)
I think these will be the first books most people think of.
Your idea is great and I hope it works out. I could only get errors when trying.
For me, I don't need any more book recommendations. Once I started reading more a few years ago, each book referenced or even recommended other books, and I'm still riding that wave.