Quote:Original post by Oluseyi
This goes back to the problem of interpreting rating as a measure of expertise. boolean is the tenth highest rated non-mod, but I don't put much weight on his technical skills, personally, so why should his rating of a post as helpful mean more than S1CA?
I'm not rejecting the idea. I'm just not convinced yet.
That's precisely what I'm hoping can be cured, although I didn't say it too well.
Like I said, I have no real ideas for how to actually implement such a thing; certainly a simple calculation based on, say, current user rating isn't going to work well.
I think the question partly comes down to purpose. Are we interested in flagging technically valuable content, or just "nice posts" from "nice people"? Judging from the suggestions of archiving highly rated posts, I assume that the intent here is to focus more on the technical value side of things. (I mean specifically that the per-post rating is meant to have this focus; not necessarily the site/community as a whole.)
As such, if a per-post rating (or at least some part of the per-post rating system) is specifically emphasized as a metric of technical merit, it should be possible to aggregate a user's post-rankings to construct the weighting value. However, as long as there is any component of mere "helpful" or "friendly" or "funny" or whatever, that function is clearly disrupted.
The current rating system was intended to be a self-moderation tool; it's designed to measure how well people play nice and get along. For that purpose I think it's worked quite well.
However, we have a separate need in the community, and that is to gauge technical value. This is, by and large, completely orthogonal to measuring community participation skill - hence the common problem of figuring out if someone's high rank is because they're making an effort to not be a dick (like me), or because they actually know something and are worth listening to.
I think both metrics are needed, but clearly any solution that tries to represent both concepts in a single number is not going to be optimal. I'm not sure I like the idea of muddying the waters by attaching
two numbers to everybody, but it may be unavoidable, and with suitable interface-level abstraction it should be manageable.