I would like a sentiment analysis of the comments posted beside each article link. A kind of temperature reading. So you can guage if the comments are generally positive or negative and use that as a proxy for whether the article is considered valuable enough to click on.
It's not perfect as conversations evolve away from the articles. But it would be useful I think.
In natural language processing and Hacker News in particular, it's often incorrect that a submission with mostly positive comments imply an article is valuable, and conversely a submission with mostly negative comments imply an article is not valuable.
That rationale just leads to even more groupthink than usual for an online community.
Maybe the quality of submission is not what's being judged, but whether or not the discussion itself is valuable. I for one prefer to read discussions in a positive tone...
more pointedly, comment sentiment is a naive and misleading signal of either submission or discussion quality. most simply because it's a poor proxy for why, but also because quality is an amorphous concept that defies direct rationalization (i.e., being numericalized), among other factors.
naive heuristics are also gamed more easily, especially because they have a tenous relationship to the desired signal in the first place.
slightly more interesting would be absolute value of comment sentiment, which would be a (still naive) measure of controversy/engagement. to really get at quality and value, you'd have to consider comment semiotics/semantics (the symbolic meaning and content), which complicates the effort exponentially (perhaps impossibly so).
That's not what I'm saying. I'm saying an article that is positive sentiment OR negative sentiment is INTERESTING enough to click on. An article with luke warm sentiment either way is less interesting.
It's not perfect as conversations evolve away from the articles. But it would be useful I think.