I'd seen various HN trends tools over the years ([1] [2]), but they all used strict keyword (n-gram) matching. That limited a) how sophisticated any trend-surfacing could be and b) the depth with which you could explore the full discussion around any topic. How can you tell what people are saying about, e.g., agentic coding when you have to search for a dozen applicable keywords at once?
So I built p-Hacker, which uses a ML pipeline to cluster the HN corpus into distinct topics and spot current trends beyond simple keyword frequencies (though it does that, too). You can search for full topics or simple keywords, and compare them side-by-side.
I started off running it on my laptop just to work out the mechanics, but after lots of iteration thought maybe I should put it online publicly. There's lots of room for improvement (the data only goes back about two years, nightly updates take a couple of hours, item topic membership could be improved).
Feedback welcome.
[1] https://news.ycombinator.com/item?id=22233457 [2] https://hntrends.net/
Comments URL: https://news.ycombinator.com/item?id=48211212
Points: 2
# Comments: 0