Proposals For Librelist Moderation Strategies

To understand the feature requirements for moderation we need some goals. Keep in mind that no moderation will be perfect, and you can easily come up with scenarios that will work around anything we come up with.

Therefore, we should focus on just some initial goals that will work right now, and keep in mind that these will need to be constantly tweaked and worked on as the spammers evade the measures.

• If given the choice between restricting free speech and preventing unwanted communication, free speech always wins.
• The system should increase the quality of discourse for any project, regardless of human language used.
• It should never give a small group the ability to hide communications from others.
• It should be implementable and not have high hosting costs.
• It should not rely on a dedicated person’s constant intervention.
• It never gates email through system before sending it, but rather allows initial emails with moderation after.
• It should use information from people’s rating habits to classify them as “ratings trolls” to prevent abuse.

With those goals in mind I’ve teased out two potential list quality strategies that might work.

via Proposals For Librelist Moderation Strategies.

Someone I work with turned me on to Librelist because they knew me interest and history with mailing list systems, and I find it interesting that some folks have decided it’s time to rethink the mailing list again.

They’re right. When I faded to black on the mailman project, it was at least in part because many of us felt that mailing lists were a technological dead end, and that deliverability issues because of anti-spam systems made the “personal mailing list” an increasingly difficult thing to accomplish.

Both are — for the most part — true. I certainly would never run my own mail server again, because the advantages of doing so are far outweighed by the time and hassle of trying to manage deliverability and reputation to make sure mail it sends gets accepted, and the constant onslaught of incoming spam turns them into a permanent infinite time sink. That’s why I either retired our lists or moved them to Yahoogroups (which I personally think is a pretty good system).

But there’s still room here to rethink the concepts and the Librelists seem interested in trying, and I think that’s great. Email and mailing lists are far from dead — but instead of stand alone delivery tools, they really shine as part of an integrated web strategy; Yahoo groups is a nice first generation of that, although there’s a lot more Yahoo could do if they decided to.

Message moderation really breaks down into two big problems:

• “Subscribe spam” where spammers sign up to the list to spam it.
• “Member warfare” where existing, approved members get into fights and they escalate into unacceptable territory.

The first is really simple to solve: new members are moderated, and messages aren’t posted until reviewed by someone to vet their content. Simple implementation; Yahoo Groups does it today, and on the lists I still manage, it works well to keep the spammers at bay. The way I manage it is all members are moderated until their first post. if their first post is acceptable, I turn off the moderation bit. To minimize delays in propogation of new member messages, simply choose a moderator pool large enough to guarantee held messages get reviewed and approved in a timely manner — you could even make that moderator pool all members in good standing if you want, because all you really need is someone you’ve trusted to post vetting that someone new is trusted to post.

Member warfare is trickier. I hesitate to call it trolling because the pure troll is a subset of the larger issue of two (or a small group of) people getting pissed off and going at it. A troll is simply one person going off on the rest of the list.

I’m more and more convinced the answer here are reputation systems, where over time a user’s membership in a group is used to define their abilities and restrictions. The longer a member is in the group in good standing, the more often they contribute material, the higher their reputation goes and the more the can do and the more sway they have on the decisions of the reputation engine. You can tweak the details of the algorithm almost any way you want, but if you define it in terms of “how long they’re a member” and “constructive contribution to the community”, you can come up with a metric on how valuable that member is to the community, and then use that to rank that member’s contributions and recommendations.

Here’s one rough view of how to build this. Please note that I firmly believe karma rankings are private and users have no way to see what their ranking is or compare it to others, except in really broad user categories (“member”, “senior member”, “top contributor”, “advisory board member”). As soon as you create a list of any form, you will attract people who see it as something they can game, and so they will.

User Karma is a value between 0 and 1, which starts at 0.5. Every time a user contributes to the system (a posting, a reply, a moderation recommendation, etc), the number gets bumped by some value. How much the value is incremented or decremented depends on how it’s rated by other users — so if User A posts a message, User B flags it as spam, but 80% of the membership feel that was a bad decision, User B’s karma is reduced in future decisions, they lose influence. Over time, the system self-corrects by giving increased influence of those who’s decisions match the community consensus and reduced influence to those who’s postings and recommendations don’t match up well.

The system can then choose whether to accept or flag for moderation a posting based on a poster’s karma score. You could potentially reject outright users that have karma scores below some value, or allow other members to choose not to see messages by users with karma scores below some value. Over time, users who are disruptive to the community will get karma’ed into the moderation queue (or out the door), and users who are seen as top contributors will have stronger influence.

My goals:

• A system like this can be built nicely with a good SQL backend and a bit of horsepower. I’ve actually done a detailed design and schema on this before, and it’s a fascinating thing I’ve always wanted to implement.
• It enables the power of individuals to police themselves.
• It limits the ability of an individual to harass or cause problems.
• It doesn’t lend itself to people playing the game of gaming the system by not exposing the details of the system (slashdot karma whores need not apply).
• Trolls get edited out of the system because the community will quickly recognize them for what they are and trash their karma, causing their postings to disappear to the bottom of the list.
• Cliques and Mafias have to be large to influence the results significantly. You don’t completely avoid the clique/mafia problem, but you can severely limit it’s ability to wreak havoc.
• It doesn’t require a lot of manual handholding or babysitting. Admins end up stepping in only in extreme cases.
• Because trolls tend to get edited out of the system quickly and automatically, they tend to go elsewhere because without feedback and controversy, they wither and die. And by editing them out of the system quickly, you avoid the whiplash and fighting that happens when people start fighting with the trolls and the wars break out.

Weaknesses:

• Any community tends to turn into an echo chamber. Automated systems like this encourage this because “different thinking” tends to get rated down.
• That’s usually a lesser evil to letting the trolls run wild.
• To my knowledge, nobody’s ever solved the problem of the conflict between the group-mind reinforcing the echo chamber and allowing the free thinkers to poke at the community’s comfort level by pushing them to think about things that make them uncomfortable. One person’s rebel is another person’s troll, and that’s not solvable in real life, much less in automated life like this…

These techniques are all based on (or stolen from) things that are in use around the net, with Amazon’s review feedback being one I really respect; while trying to avoid the pitfalls I’ve seen around the net (yes, I’m going to keep bashing on Slashdot’s karma system, it’s way too easy to game and always has been). It also (I believe) avoids the nasty politics that have made Digg a bit of a pesthole. And it’s also pretty lightweight and low-key, or at least it should be. The implementation details will be crucial, as will be tuning how the karma values adapt…

Related posts:

1. How Moderation can backfire A nice article on the side-effects of publishing reputation data...
2. Making Light: Moderation isn’t rocket science Making Light: Moderation isn’t rocket science: If you have a...
3. Evaporative cooling of group beliefs Evaporative cooling of group beliefs: Over on Overcoming Bias there...
4. Usenet mavens Fascinating piece on what they’re calling “mavens” in usenet —...
5. Group-forming Sébastien Paquet is part of a group that’s creating a...

Comments RSS Both comments and pings are currently closed.