The very best submissions get virtually no discussion.
One measurement to make might be the votes divided by comments.
Let V = votes
Let C = comments
Let IT = 1.0
Let R = V/C
If R > IT then the article is "worth IT"
Let "the paper it's written on" be 0.1.
If R < "the paper it's written on" then it's not worth the paper it's written on (not that it's written on paper, but you get the idea).
But all this is tinkering around the edges, and doesn't feel like it's significant. Perhaps a new model of how a site like this works at all is necessary.
"The very best submissions get virtually no discussion."
I'm not sure I agree with you on that. The best stories produce a high volume of high-quality discussion. Your proposed algo punishes submissions that generate discussion, which seems to be the opposite of what's desirable.
I have to agree with RiderOfGiraffes -- an article with a high number of comments relative to points usually indicates that the article is more opinionated than informative, since most everyone who upvoted it posted their personal reaction to it. Although they are popular, I feel these are usually low-value articles and discussions.
Additionally, it's very hard to have a high-quality discussion that is also high-volume. Once an article's comments get over several pages long, you can be sure that most commenters are not reading everything before commenting, so you get a lot of similar comments. Sometimes this is valuable (e.g. "Ask HN"), but most of the time it just worsens the signal-to-noise ratio.
Maybe we just need a semantic distinction between editorials (opinion pieces) and articles (investigative/fact-laden pieces)? Perhaps make editorials count for less, in the same way self-posts currently do?
I think those good stories that generate good discussion still get more points than comments. With the really good stories, no matter how many people comment, more people upvote, because there are lurkers.
It's not perfect, but I'm finding it a pretty good predictor.
One measurement to make might be the votes divided by comments.
Let V = votes
Let C = comments
Let IT = 1.0
Let R = V/C
If R > IT then the article is "worth IT"
Let "the paper it's written on" be 0.1.
If R < "the paper it's written on" then it's not worth the paper it's written on (not that it's written on paper, but you get the idea).
But all this is tinkering around the edges, and doesn't feel like it's significant. Perhaps a new model of how a site like this works at all is necessary.