Comment by Semaphor | Hacker News Reader

Comment by Semaphor | Hacker News Reader

The idea is cool, but I have some suggestions.

1. Use links, when there’s a list, I open links in new tabs that I want to check. This is some kind of JS-only app that doesn’t have links, only JS behavior. Please don’t break the browser for no good reason, this makes it really hard to actually use the site.

2. Get a better source for the book data:

> What is this book about?

> Read the award-winning, critically acclaimed, multi-million-copy-selling science-fiction phenomenon - soon to be a Netflix Original Series from the creators of Game of Thrones.

> 15+ pages of new, original content, including a glossary of terms, in-universe writings, and more!

> A USA Today Best-Selling Novel!

> "Unlike anything I've ever read. " --V.E. Schwab

> "Lesbian necromancers explore a haunted gothic palace in space!" --Charles Stross

> "Brilliantly original, messy and weird straight through." --NPR

Almost everything I checked had a bunch of absolutely useless marketing blurbs at the top, maybe one or two paragraphs of explaining what the book is about at the end, but not even always that. I don’t know how accessible it is, but goodreads tends to have descriptions without those blurbs.

> 1. Use links, when there’s a list

Yep, this is on my list to do for next year. I just ran out of time to do that this year (this is the first year I've done this).

I will also rethink what data I show for 2024 as I want to better show which authors picked which books visually.

> 2. Get a better source for the book data:

I wish there were a better source; book data is a nightmare. There is only one source of book data and that is the publishers through a format called ONIX. The publishers have no idea what they are doing, and so they stuff the book's "description" full of marketing crap as well as abuse the BISAC-defined genres :(. Goodreads has an army of unpaid people who clean this all up.

I am looking at rewriting the book's description using AI as it seems like it is the only choice to get rid of that nonsense without a ton of manual work (since I don’t have user accounts to crowdsource that).

Genre miscategorization is harder, I am playing with AI to help me clean that up, but also thinking about using crowdsourcing some of that.

Just one example: I found several editions of Dune that report they were published in the 1780s and 1880s... and this is a book selling millions of copies a year.

Have you tried archive.org’s OpenLibrary [0]? They have an API and at least sometimes decent data.

Great news for 1. ;)

[0]: https://openlibrary.org

Ya, I tested it in the first year and it was worse than publisher data by a wide margin. I will take another look soon to see if it improved. And maybe I can use their data to augment publisher data and improve the overall quality.

Ouch, okay, that sucks. At least the descriptions I checked were always decent.

Ya I didn't test that at the time (more book meta data), that I will look at. I also might be pulling the wrong ONIX data and need to up my license and take on the full HUGE database of books rather than the API I am using.

> Genre miscategorization is harder, I am playing with AI to help me clean that up, but also thinking about using crowdsourcing some of that.

AI could indeed help with that.

But if you're publishing a books-of-the-year list, you could also check the categorisation manually - especially if you adjust the inclusion criteria to give you a tractable number of books.

For some datasets, 4 hours of manual classification will do a better job than 40 hours of trying-to-get-the-AI-to-work :)

For 1, try this quick-and-dirty fix:

  document.querySelectorAll('.book-preview-body').forEach((e, i) => { let title = e.querySelector('h2').innerText.toLowerCase().replace(/ /g, '-'); let button = e.querySelector('button'); button.innerHTML = `<a href="/book/${title}">Learn more</a>`; })

I'm guessing the link from the title, which may not always be accurate (e.g. for non-unique titles). The actual link isn't loaded until the button's modal is displayed - a more foolproof way to do it would be to load the modal's data (it's in a nearby div's hx-get attribute) and get the proper URL from that.

For sites I use regularly that have problematic UIs like this, I write TamperMonkey userscripts (yes, I have quite a lot of them at this point).

Thanks, taking a look now!

btw all fixed now :)

Issue #1 irks me so much. Often I'll find myself having gone through many things in a list opening all in new tabs. Only to check my first tab 5 minutes later and find the same page reloaded.

That’s even worse, yeah, at least this site fails early by not reacting at all ;)

Yep, I am sorry, I also use tabs this way and I just ran out of time this year. I made the mistake of doing my user testing on a small non tech crowd who doesn't do that so I just missed it :(

> "Lesbian necromancers explore a haunted gothic palace in space!" --Charles Stross

This does properly describe Gideon The Ninth, by Tamsyn Muir.

Something I am playing with is using ChatGPT to do emoji descriptions as a joke :). The results are quite funny.

Reply on news.ycombinator.com