You need to define "not scaling well" more explicitly. Where you
seem to be doing so "the signal to noise has been dropping... trolls...
number of comments... repetition of comments", I'm pretty sure I don't
agree with you.
There are some articles that generate a lot of discussion -- close to
400 comments in some cases. There are an awful lot that float around a
score or two comments. This is a broad range of interest levels, and is
a good thing: K5 can accommodate an intimate conversation, or a rollicking
debate society. Number of comments is not of itself a bad thing.
Repetition will naturally occur in long threads. You'd need something
akin to a Wiki to avoid this. That's not a K5 design goal.
There are more trolls. There was a rather interesting episode
earlier this evening as I write in which an author became unhappy about
a story that was dropped from queue (hey, it's happened to me too). But
the overall troll level is pretty damned low, and even now the hidden
comments display rarely shows more than one hidden post every day or
two. How does this compare to Slashdot again?
Signal to noise -- this is probably true, but with increased size,
the really good signal is increasing, and the moderation system tends to
pick it up. Not always, not perfectly, but pretty damned well for a
pretty simple system.
So, I have to ask, what's your specific goal. You're going to do
better describing your preferred system, and then designing controls to
help assure it. I don't see this here -- you've got an awful lot of
design, but not much goal.
There is a certain logic to having some resource limits on
specific system actions. How this gets tied together, I'm not really
sure, and any system is probably going to be relatively ineffective
against concerted attack -- you have to remember that there's no such
thing as an authenticated user on K5. A K5 userID is as close as the
nearest free email service. Still, it might make sense to have some
sensible limits per user, such as posts and submissions per day and/or
week. A set of descending limits -- say:
- 3 posts/minute
- 15 posts/hour
- 40 posts/day
...might make sense.
- 2 submissions/day
- 5 submissions/week
I am somewhat interested in ideas for making it worthwhile for people
to post high quality submissions, and not post low quality ones. I
haven't had any compelling ideas on this myself though. I'd also like
to encourage quality moderations, and discourage low quality
ones -- a key problem, of course, is working out what a quality
moderation is. A related problem is working out what K5 has of value
to provide its users. I suppose you're indirectly suggesting that
privileges to post, comment, and moderate, are those benefits.
Which gets to a few other issues.
Really abusive behavior only exists on the fringes. There
have been a handful of cases of flagrantly abusive moderation practices
(and a few more of people taking up vendettas or tit-for-tat campaigns,
but to little effect). We've had a couple of flameouts. There's Sig11,
thurler, and streetlawyer. And for the most part, K5 creaks and groans
around them. Limits sufficient to restrict comment postings or article
submissions would almost certainly have a more oppressive effect on the
white hats at K5, stifling conversation. It's better to have effective
monitoring systems to identify abusive accounts and cancel them than to
penalize everyone in the possibility that they might misbehave.
Unbalanced incentive schemes do more harm than good.
Case in point is Slashdot karma whoring. K5 Mojo is
mostly...worthless. I get to see some of the low-level crap that's
posted here. Hardly a compelling benefit. OK, so to sooth my ego, I
get a '0' rating to swat imbeciles with. Better than nothing, but as Rusty puts it, the thing that makes Mojo work so well is that it's really not worth anything.
Cumulative benefits promote distortions. This is
another element of /. karma, and is the reason why K5 moderation (and
Mojo) is on a 1-5 floating point scale, rather than an open-ended cumulative point system -- which now has caps hacked onto it. Mojo isn't something you can
bank. It's not something you can use to get ahead of the pack. You're
either approaching perfection (5). Quantity doesn't matter, quality does. And in the case of Mojo, old basis is aged out over time -- your score is present-weighted.
Where incentives and weights are used, they should be tailored
to goals. I'm pretty convinced that none of the specific
values for scores you've suggested are suitable. I'm also pretty
convinced that suitable values would most likely only emerge from a
predictive scoring model based on statistical analysis, for a system such
as you're presenting. By simplifying the system to a set of activities
and resource limitations on these activities, a direct correlation can
be determined. This simplifies cause and effect, and management. Not
that resource limits are necessarily required.
Growing the system is a goal. One criticism of K5 I've
seen expressed in several different ways it that growth is not
necessarily good, or that it's not a necessary goal, or that it
inevitably leads to failure. Larger communities make conversation more
difficult, granted. But the corollary is that all large communities
break down. I'd like to think this isn't necessary. As I put it
recently, I'd like to stop running away from where all the people are.
K5/Scoop are an experiment in scalable group discussion.
A sufficiently large community can't be fully
centralized. Bits and pieces of K5 have to be split off. I'd
like to see the article submission process decentralized such that
submissions go directly to some section -- Meta, MLP, News, etc., with a
quick "fix, kiss, or dump" assessment made. The submission would then
be floating either to the original section (if good enough), resubmitted
to a more appropriate section if miscategorized, posted to the author's
diary (if dumped), or to a morgue (spam, trolls). Moderation and
other activity including reads, comments, and comment moderation, on the
article in section would determine whether or not it got front page
placement, with rotation and retention times specified by section (and
yes, this really is simpler and better than what you're suggesting,
besides, I'm on a roll ;-). And good content from various areas,
including diaries, could be promoted forward. The centralized
submission queue would disappear. The apparent data load on K5 would be
K5/Scoop provides a wealth of data already. There's an
awful lot of data, some explicit, some implicit, being gathered by Scoop
all the time. Posts, moderations, reading patterns, submissions. I'd
like to see the existing data used in a form directly aimed at specific
goals (encouragement or discouragement of behavior) rather than layering
some Byzantine indirect scheme onto the system.
You make some good points. Coarse resource controls may be helpful.
Better incentive systems might help things out, but I don't feel this is
a currently critical issue for K5. Straightening out the moderation
queue is far more important. Tossing a scoring system willy-nilly isn't
necessarily going to fix things and may make them worse. Certain of
your goals appear to contradict K5's aims. But I don't agree with your
assessment that K5 isn't scaling well.
Karsten M. Self
SCO -- backgrounder on Caldera/SCO vs IBM
Support the EFF!!
There is no K5 cabal.