CmdrTaco's Journal: IRC Forums, Legaleze, The Multi User Moderation Problem

Allright I got another one. I have a huge pet peeve: I friggin' HATE it when you get emails that contain these gigantic disclaimers at the end. Like 3 paragraphs of legalese that essentially says "You can never repeat this information to anyone".

Except they put it at the end of the email.

The really bad part of these things is that often we are dealing with a legitimate problem where legal issues need to be addressed. But in order to deal with the issue, I immediately must concede on their terms, not to talk about it. Even more annoying is that I don't get to find out about that little agreement until AFTER I have already read the message and been "Contaminated" by its contents. Bah. I should add a procmail rule to bounce all such mail.

We're planning an IRC forum for thursday at 8pm (Eastern) on irc.slashnet.org. The channel will be announced in a Slashdot article in the next day or so. We don't plan to discuss any biz related issues, but we're open for moderated discussion on moderation, story selection, code plans, etc. SlashNET forums are always fun, and thanks to the nets Opers for hosting it for us.

Ya know a very common problem on Slashdot is users who use multiple accounts and multiple AC posts to inflate the size of their viewpoint. We've seen discussions with a dozen user accounts participating, and many AC posts, but all coming from two or three IPs. Certainly these could be proxy servers or corporate firewalls, but I've seen discussions where it was quite obvious that a dozen "Personalities" involved in a discussion were actually the same person.

I've long struggled with a way to retain anonymity for our legitimate posters, but also to simultaneously discourage users from taking advantage of the system in this way. It's not fair to users who fairly use a single account and try to represent their views using a single account name, and never resort to anonymity to inflate their viewpoint.

The bad way to solve the problem would be to simply show the relationship between comments... not telling you the IP, but telling you that users in fact shared the same IP. Unfortunately this sorta defeats the whole anonymity thing, and I think that it's very important.

Another possibility would be to tie karma up to your IP somehow. We already do that to a certain extent: a single IP that is host to a dozen user accounts has a single counter to prevent crapfloods. So if a single user were to post a dozen comments and get moderated down from Score:1 to Score:-1 a dozen times (24 down mods) that IP would have posting restrictions automoatically put in place upon it (No ACs, Slow Down Cowboy Warnings, and in extreme cases, an all out ban).

But what I was thinking is that we could tie karma into the equation. Essentially, have a small chance that M1 could affect other users sharing your IP.

Now hold on you say, isn't this unfair? But I remind you that karma doesn't matter- it's a tool for us to make the system better, not your value as a human being. More so, I'm not saying "We will do this", I'm thinking out loud.

Let's say there's a number N which is the number of comments from a given IP participating in the discussion. Now lets say there's another number, U, which represents the number of unique users sharing this IP. AC could count totally independantly, or be summed up as a single user. More likely we could use the Logged in to AC post ratios that exist on the site as a whole, and say something like every 3 AC posts count as an additional user, so U++

What I need to do is figure out a way, using these numbers, to decide arbitrariliy if an IP is Suspicious or not.

Let's take a hypothetical discussion where 10 user accounts are sharing an IP, and each posts 3 comments. Plus for good measure, 15 AC posts. Using the numbers I describe above, that would make U=15 and C=45... and the ratio of posts to user is 3:1

As U gets larger, my suspicion needs to be raised. There are very few proxy servers with 15 active users. So my guess is that at U=15, we already have an issue. I'd need to crunch some numbers to come up with some average post ratios... but I bet 3:1 is pretty typical. If things go to 2:1 or 4:1 maybe that means that raises suspicion. Maybe its 5:1.

Again, none of this will be 100% accurate, and before I'd consider implementign it, I'd crunch some hard numbers to see what scenarios it catches.

The other side to this is what to do with IPs that we regard as suspicious. For the purposes of this discussion, I'd need to come up with some sort of suspicion rating (Let's say S). For example, if U=30, Let's say S=90%... I'm pretty sure that if 30 users are sharing an IP, and all participating in a discussion, that something fishy is going on.

Now what we could do with this is essentially to start unifying karma for all the users from this IP. If S=90%, perhaps all 30 users take a -1 karma hit for being moderated offtopic. Maybe we fudge it out so 90% of the 30 users take a -1 karma hit. The Point of all of this is that we start having the ability to treat many users sharing an IP as a single user when they start acting up. Then we can inact the Slow Down Cowboy post delays... or enough down mods and suddenly a user impersonating a dozen user accounts finds himself with a dozen user accounts with terrible karma, posting at -1, and not really bothering anyone any more excpet those that read at -1... and honestly those folks are suicidal anyway ;)

The hard part is to work out algorithms that catch the dozen guys who actively do this, but without hurting the legitimate proxy servers used in hundreds of businesses. All while retaining anonymity where possible.

Anyway, those are just some ideas. I don't know if we'll ever implement anything based on them... I just like to think out loud about ideas so people know what we're dealing with on this end. These ideas might be entirely for shit, and they will be forgotten along with countless other ideas that are either to complicated, computationally infeasible, or just stupid.

We could chat more about this at thursdays IRC forum if people are interested.