[geeklog-devtalk] Idea for another comments system

[Ray Heasman]

Hi Vinny,
On Fri, 2004-09-10 at 05:51, Vincent Furia wrote:

> Basically your talking comment/article moderation with a twist.  I like it...

> 

> There is probably a simpler way of implementing it.  For instance, for

> each comment/story have agree/disagree/neutral buttons.  By following

> who a user often agrees with and who a poster disagrees with you can

> do some algorithmic analysis (which could be as simple or complex as

> you link, i.e. look at who those a user agrees with agrees with).  I

> think this will get you to the same place, but in a simpler manner

> (from the user's standpoint).


Yes. I am not sure exactly how much information will be required. I'll
probably write some simulations before I try a full implementation.


> Another issue you have to be careful of is maintaining threads when

> you reorder comments based of a user's previous responses.  Reordering

> comments can cause a loss of the flow of conversation.  Also you'll

> have to normalize because the nature of moderation is that first posts

> get more heavily moderated (and often time more highly moderated) than

> later comments posted to a story.


Yes. I realise this will be tricky. I have some thoughts on this, but it
is by no means a problem I have solved. I don't intend to re-order
comments, so much as only show the ones that make a threshold.

The problem that I have to deal with is what do you do when someone
makes a sterling reply to a crappy message? Do I show the crappy
message? What about a thread of good comments with one flaming
monstrosity in the middle? I haven't figured out what will work best
here. I suspect that I'll need to do some live experimenting...


> 

> > 

> > Trolls will quickly cluster themselves away from the main groups, which

> > will also regularly rate them low. When looking at comments, trolls will

> > only be able to see people that agree with them, or other trolls. Other

> > groups consisting of people that would tend to rise to trollbait will

> > hardly ever see the troll's comments.

> > 

> 

> I don't think this will work.  Real trolls will lie, and if they are

> any good they'll find sneaky ways around the scoring.  No reason to

> give up on the idea, but I don't think it will be an effective

> anti-trolling measure.


Well, you _never_ score yourself. By lying, all a troll does is degrade
other peoples score, and only slightly as long as trolls are in the
minority (which by definition, they are, because if they are in the
majority then they aren't trolls, they're fine upstanding members).

They can try and lie to clustering questions, but they then have to
predict other people's answers to clustering questions, and any
deliberate errors will probably be identifiable automatically. For one
thing, scores should often be reflexive - if A rates B low, then B will
likely rate A low. Imagine a normal user flaming a troll and the troll
replying.

I think the only "banning" offense for a site would be lying about
questions, which can probably be caught automatically. And, far more
subtle than banning is just having the admin manually assign the troll
to the "troll" group.


> As with any moderation system you have to build in ways to prevent

> abuse. "Mod-storming" is common on sites that don't take this problem

> into account.


*nod* Yes. I'll put more thought into this, but I suspect that the
reflexive check above would be a good start.

Anyway, any thoughts about how I should implement this? Can I build this
on top of the existing comment code or are we talking major surgery?

Cheers,
Ray