Explore Hacker News by topic: discover stories, follow trends, and browse the archive.

88 points•andrearitossa•5 days ago•41 comments•
I built a tool to explore HN topics and how they change over time.

Currently covers 2024 to today. Is it useful? What's missing?

41 comments

kmaitreys5 days ago
Just a heads up: Your classifier thinks Sun (the company) is actually Sun (the star) and hence it[1] shows up among the posts under Astronomy topic.

[1] https://news.ycombinator.com/item?id=49787436

andrearitossa2 days ago
I had the same problem with "Deep Seek" classified in "web search" :>

I'd list it as limitation of embedding models pipelines. Fetching the link would drastically help, but I wont for now.

Ideas on solving the issue?

DonHopkins4 days ago
It seems unlikely for a Sun worshiper to develop a philosophy of total brotherhood. Sun worship is usually a primitive superstition religion.

https://www.youtube.com/watch?v=WNfTwGu9FpU

andai5 days ago
Cool idea!

From the name, I was expecting it would be a spatial tool, i.e. some kind of semantic graph I could navigate. I think that is fruitful territory.

Edit: Oh hey! There is almost exactly what I was visualizing, on the Connections page! This is the coolest part, why isn't this on the front page? (Also, please add scroll to zoom!)

As for the grid, I don't see the value of pagination. I'd rather have it all on one page, so I can scroll, Ctrl+F, etc.

Also, like half of the first page are different kinds of AI? If we want to visualize what % of HN stuff is AI, maybe it would make sense to have a meta-category for that.

(Would also be cool if HN itself supported tags you could toggle... show/hide AI :)

I'm also very curious about the evolution of this data over time. The popularity stats seem to be limited to last 30 days. Would be cool to see stats for all time, and maybe filter by range. (Or graphs? Graphs are cool!)

andrearitossa2 days ago
Hi, great feedback!

I took the suggestions: now the connections are first page; cut pagination.

Because 5 of the first 8 topics are AI-related then these topics gets displayed on top when ordering per activity, it might look different in a couple of years :> for sure I could make a graph topic popularity by time and see the major tech trends by time, non trivial would be to compare it against the YC call for startups.

Giving this to YC would be super cool for me! Any suggestion on how to give this to them?

finger5 days ago
Can you add some statistics about comment lengths averages over time. I swear comments are substantially longer on average today than they were even just 1 year ago, but I might be seeing things. It's easy to assume why this is if it really is the case.
random35 days ago
I thought it's a collection of maps (what an atlas is). Instead it's a list, 9 items per page, 23 pages - it's an index. A map lets you see the full picture in a glance. It seems like a missed opportunity to actually make it an atlas.
livinglist5 days ago
You might be interested in this: https://hnheatmap.com/
random34 days ago
thank you! that's nice looking! one thing from OP that's useful is the topic rather than the post. a mix between the two :)
andrearitossa2 days ago
It is a map :> try again
johnisgood5 days ago
The count of posts are probably not a metric for quality though (not that you claimed it is), I would like to see at least the number of words accounted for! It might just be the case there are way too many more of two word sentences! :D (And it still would not be a good metric, not sure why I brought up quality exactly, I am just wondering what would actually be a good metric here for that. :P)
andrearitossa2 days ago
The formular righ now for selecting in "the conversation now" is:

(points-1)× e^{-age in days / decay}

I am testing right now with decay: it was 14, I am wondering between 7 and 3.

Do you have suggestions on a different way to do it? Comments?

Read the full thread on Hacker News →

Related stories