Has anyone done sentiment analysis on HN?
I ask as comments have really started to feel more reddit-like knee jerk thoughtless narrative (in volume). Still great commentary on here but some almost thoughtless bot like comments showing up diluting quality maybe as a result of reddit tanking.
The trouble w/ sentiment analysis is that you really need a model trained for the domain. A model that is good for product reviews is not going to be good for stock market discussion boards or angry toots. I’d badly like an angry toot filter but I’d have to find about 5000 angry toots and 5000 non-angry toots and that is probably about a week’s worth of psychologically difficult work. (The whole point is that I don’t want to see the angry toots)
“Thoughtless”, “low-effort” comments on HN might not be so debilitating to look at though but given that “bad” is in the eye of the evaluator you probably would have to look at a lot of them to make a model.
I done sentiment analysis of tweets. It literally was as simple as looking for positive words like 'good', 'amazing', 'awesome', 'brilliant', etc and maybe looking for a smiley-face emoji alongside the tweet, and I found positive sentiment for brands and interacted with the tweets. I was working for a social media company at the time that done brand engagement. Now Twitter is dead we're looking at Bluesky, Nostr, Lemmy, Mastodon, Threads, etc and do our analysis there. It's very astro-turfey, but it works.