Conversational quality metrics are difficult to come by. Most tend to be expensive to calculate, especially in the absence of clearly parseable engagement information (e.g., likes, re-shares, comments). I've played with this concept in a few cases, most recently when trying to make assessments of activity / genuine community in Google+ Communities, prior to their shutdown.

Some of the more interesting research I'm aware of was based on Usenet, looking at thread dynamics, out of Microsoft Research, in the early/mid aughts. I think it was Marc A. Smith, see for example:

https://www.microsoft.com/en-us/research/publication/picturi...

https://www.computerweekly.com/news/2240051972/MIcrosoft-Res...

(There were more detailed discussions I've seen in the past, I don't have references handy.)

In particularly, post/response dynamics seemed to fit a number of distinct categories:

- Single post, many responses, no follow-up: question with a single obvious answer provided by many.

- Single post, extended discussion, "conversation-killing" response: hard question, with unclear answer, eventually obviously provided.

- Long descending "right-shifted" thread between two parties: personality conflict / flame thread.

- Post, much response, multiple parties: successful trolling.

I've played a bit with the notion of indicators of significant vs. insignificant discussion, based on sentinal keywords: https://old.reddit.com/r/dredmorbius/comments/3hp41w/trackin...

I know HN has an API and have considered looking at it though as yet haven't explored this.

Any ideas as to what would indicate desirable or undesirable trends, threads, and/or posts?