Beta Testers required for my book-recommendation website
bookclub.ai
bookclub.ai
I couldn't get far enough with it to even see the complete loaded UI.
const user = app.logIn(credentials);
export const client = num_users===0 ?
// wait for 1 second before reloading the page
setTimeout(() => {
window.location.reload();
}, 350) : app.currentUser.mongoClient("mongodb-atlas")
This https://www.mongodb.com/docs/realm/web/authenticate/ shows how to login whilst waiting for a user to be created.
Edit: @aditiyadav Hope that doesn't come across wrong, just pointing out where the problem could be :)
When general users tell you about a problem, 80% of the time it's not that problem.
Much appreciated!
I have build an AI book-recommendation website this month & have 30k books in my database right now.
Bookclub.ai has based on genres, celebrity recommendations, popular lists and series.
I have applied AI to present a visual representation of "Books More like this" in the form of mind-map feature.
One of the best features is "Book-Genie" which recommends books real time based on your mood.
I am a sole member of this project right now and in bootstrap phase. Would love for you to checkout my website and any feedbacks/collaborations are welcomed.
Happy reading!
Link - bookclub.ai
My Email id (for feedbacks/collaborations) - 19aditiyadav@gmail.com
I'm working on different product that involves processing book content, and trying to plan for the computational requirements of millions of books and it's a lot, but I'm guessing running an language model over them would be several magnitudes more demanding.
Good luck, seems like a really promising use of AI!
Running a language model is challenging as it took ~22 hours for training on 30k books and I expect it to take at least a week on M1 GPU.
On Size, I expect 10Mn books to be taking ~16GB storage in my database.
Thanks for your comments and good luck to your data pipeline.
Initial feedback is that the UI on the webpage seems busted, or at the very least poorly-scaled. For example the "Celebrity Picks" photos are way too large and pixellated, and the "Recommended Lists" title text is hard to read against the background of Hillary Clinton's red jacket.
The UI also seems a bit too busy - hard to know where one should start the moment they land on the page. Particularly that it is recommending me a bunch of genres and celebrity picks before I've even given your model an indication of what I want. It'd be a lot more user friendly to either start with a simple explanation of "how it works" followed by a "start here" button, or simply ask the user for the last 3 books they read and then SHOW how it works by providing actual recommendations.
As someone who reads a bit (and who has a librarian partner who reads a tonne), this sort of app NEEDS integration with goodreads. We both maintain lists of what we've read, and what we want to read, on goodreads. If it was possible to just paste a link to my profile (i.e. give you the goodreads user id) and the page would say "of the books on your want to read list, you should read X next", or "why is Y on your to read-list? you wont like it", or even, "your 'want to read' list is missing Z!", I'd sign up in a heartbeat.
You could even scrape ratings from goodreads profiles to better train your model. I already rate books there (out of 5), so there is already more sentiment data there than you'd get by asking me to just list books as "like/hate".
As far as how book-genie operates, I put in "Sci-fi dogs" and it failed to recommend 'City' by Simak, but rather gave me a bunch of titles with 'Dog' in the title that aren't even sci-fi. To me this isn't really a good output, and tells me that your algorithm is quite shallow and isn't really recommending based on actual content.
Hope you keep going, as this is definitely a hole!
(I'm trying to get off Amazon completely. Tight Goodreads integration is therefore a nope for me.)
Key Action items: 1. Provide a sneak-peek to product(UX) 2. Integrate with goodreads or at minimum allow someone to enter their book list 3. Allow creating list from users and recommend NextUp accordingly 4. Improve Book-Genie, Have some plans to train the ML model on Reddit data
Really grateful here!
I also looked at your "genres" page. It's very weird that the lists don't match the bar at the top; are these things not generated from the same source? e.g. the bar starts with romance, fantasy; but the first categories are fantasy, magic. (also what is the difference between fantasy & magic)
Is there a way to add books you've already read & enjoyed to get similar recs? This is how Storygraph operates, and I've heard fantastic things about their recommendation engine, although tbh my TBR list is quite long enough and I honestly have no need for a product like this.
quick edit - also, the text of "Jurassic Park" seems to be in Spanish, while everything else is in English. Some i18n issue?
<<Genre Page>> Actually the "Bar" at the top is clickable and opens up set of carousels in below pane, e.g clicking on "Technology" will show carousels for "Programming","Technical" and "Computers". I plan to automate this process and personalize it based on user history.
<<Is there a way to add books you've already read & enjoyed to get similar recs?>> Yes, definite plans there and will pick it up after Book-Genie, Making the page responsive and improving search.
<<Jurassic Park>>- Yes, I need to add language detection filter.
Really thankful here and hopeful that this becomes a useful product for you in future.
Keeps continuously refreshing. Please fix.
What I really want with any recommendation system is more control with how wide or narrow the net is cast. Some genres/subjects I have completely over fished and need a really wide net. Something like Russian literature though you could just recommend me Dostoevsky because I don't know much of anything about the subject.
Edit: I asked it again, and got: a book about space, another kind of book about space (spacetime and geometry), a sci-fi book that definitely wasn't "hard" sci-fi by my definition, and a "Cat in the Hat" book about space. Certainly amusing!
Have touched upon approach here- https://medium.com/@19aditiyadav/how-i-built-bookclub-ai-ai-...
Thanks for your feedbacks and these queries X Results would definitely act as sniff test going forward.
Happy Reading!
Is the AI used to choose the recommendations or only for the graphic?
> recommends books real time based on your mood.
How can the website detect my mood???
You can check out my blog, it will answer all your questions -
https://medium.com/@19aditiyadav/how-i-built-bookclub-ai-ai-...
I understand its buggy currently and I am actively working to resolve these blank results issues.
Instead at the moment most of recommendation engine keeps recommending book in the same niche.
"intelligent fiction" - four non-fiction books.
"fiction by women" - two memoirs, two non-fiction....
So then I just tried "fiction" - guess what, four non-fiction books...
- Recommendations for books on crystals and healing
"Books like Name of the Wind"
- Recommends "Name of the Wind" and "Wind in the Willows"
I think this might not be right for my use case.
Also on Firefox it keeps reloading the page.
I understand you're the sole developer of the project and that you need to get help in any way you can.
But there are some really basic problems.
Many have already commented about the continuous reload thing. It can be fixed, apparently, by closing the tab and reopening. Apparently. That's what I did and it worked. But it's a big enough problem that needs a deeper look. That's why someone specifically trained to test would help you much better.
There seem to also be a bunch of bad-loading problems. Sometimes stuff doesn't load. This is so bad that you even have a message asking the user to reload or re-type the query until it works. Yes, I understand about the money restrictions, but throwing that against a whole community will hardly be productive.
I mentioned "bad-loading" and not simply not loading because sometimes the wrong cover loads for a certain book in the results and you hover over it and it's a different book, click it, and it gets the correct cover.
Again, I understand that you want help and are alone. But these problems are big enough to be showstoppers. In the sense that it will be much harder to get anyone interested if thing don't even load correctly or at all, or reload infinitely.
Now...
The mind-map seems like a nice feature. But it lacks links between the bubbles (at least in my browser) which feels weird. It makes it look more like a bunch of bubbles randomly thrown in there than an actual mind-map with a structure or relationship between nodes.
Also, it frequently features duplication. Just use the given example of The Great Gatsby and you'll get a second "The Great Gatsby" bubble as if it was a different book. I tried randomly "Neuromancer" and there were 2 extra Neuromancer bubbles. It doesn't look like they are duplicate entries in the dB but more like a problem building the mind-map.
I also get some random bubbles missing their title.
As for the "Book-genie" recommendations I can say very little. I tried the "Inspiring books with aliens" example and I got "Seven Brief Lessons on Physics" along Cixin Liu. Are these results good? I really can't say, but I don't feel like a physics book matches the mood. But in any case, one would have to test this much more deeply to actually evaluate if the recommendations are any good, or even simply any better than a keyword based search.
There's also more general design and UX comments that I could say. But again these are hardly specific details. It feels like you'd need someone to dedicate some time exclusively to that part and re-do things in general.
----
In any case, I do hope you get the help you need to make this work. I just don't think putting this up in its current state against a large audience will do much good.