Hand-picking has some advantages but it also produces problem because it induces political discussion about who deserves to be in the hand-picked group of high voting power users.
The hand-pickers can be anonymous to everyone except the site owners. The picking needn’t even be a continuous process; it can just be done once with no possibility of discussion. People would still yell at us for the abstract fact that we implemented such a scheme, but we’d have to weigh that against what I expect would be a substantial increase in voting quality. (Nobody would lose their vote and this would help make it palatable.)
I don’t think having a few anonymous amount of high voting power users and most users at normal voting power would get around the problem of Eugine’s sockpuppets. A page-rank like algorithm on the other hand would make the forum robust against attacks of that sort.
You can additionally seed the algorithm with giving specific individuals higher voting power.
Using a pagerank algorithm against sockpuppets feels like the wrong approach. Consider that people learn to exploit google’s algorithms, and then they make changes to fix that, and people learn to exploit them again. We don’t want to be playing that game. And as long as we have few enough trolls, I’m not sure we have to: if we have an effective way to detect and ban them manually, that should be fine.
Algorithms like this might be valuable for other reasons, but I don’t think we should be using them against trolls.
Yes, we’d need a separate solution to sockpuppet attacks, like disallowing downvotes from accounts below a karma threshold, or the one about moderator database access that’s currently in the pipeline.
Sockpuppet attacks work because accounts are allowed to downvote an unlimited amount of times without ever posting anything.
Change it so that you get 1 downvote per week if you have posted a few comments, and the sockpuppeter has to automate commenting. That’s harder for them to do than it is for the good guys to ban their puppets.
In addition, use the usual tactics against bots, such as email verification and ip banlists.
I think the suggestion in the middle paragraph would hit a lot of people that we do want voting, both up and down; lurkers with good taste make the karma signals better.
I’m not convinced that lurkers’ downvotes are good for the site, but I assume that with the current state of the database this is not a question we can answer empirically?
I think the suggestion in the middle paragraph would hit a lot of people that we do want voting, both up and down; lurkers with good taste make the karma signals better.
I think on average the votes of lurkers are less valuable than the votes of participating members. If there’s an AI that identifies lurkers with good taste because they vote similar to a moderator, their votes would be valuable.
Without a way to distinguish good taste from bad taste lurkers I think it’s more useful to count people who have a higher stake in a forum more strongly.
In general I think it’s useful to give votes of new posters little weight. I don’t think that a specific threshold number is optimal. I think a PageRank like system would produce a better result in fending of Eternal September voting issues.
Or, better yet, it should score votes by a number of hand-picked people higher. Karma is an indicator of voting quality, but an unreliable one.
Hand-picking has some advantages but it also produces problem because it induces political discussion about who deserves to be in the hand-picked group of high voting power users.
The hand-pickers can be anonymous to everyone except the site owners. The picking needn’t even be a continuous process; it can just be done once with no possibility of discussion. People would still yell at us for the abstract fact that we implemented such a scheme, but we’d have to weigh that against what I expect would be a substantial increase in voting quality. (Nobody would lose their vote and this would help make it palatable.)
I don’t think having a few anonymous amount of high voting power users and most users at normal voting power would get around the problem of Eugine’s sockpuppets. A page-rank like algorithm on the other hand would make the forum robust against attacks of that sort.
You can additionally seed the algorithm with giving specific individuals higher voting power.
Using a pagerank algorithm against sockpuppets feels like the wrong approach. Consider that people learn to exploit google’s algorithms, and then they make changes to fix that, and people learn to exploit them again. We don’t want to be playing that game. And as long as we have few enough trolls, I’m not sure we have to: if we have an effective way to detect and ban them manually, that should be fine.
Algorithms like this might be valuable for other reasons, but I don’t think we should be using them against trolls.
There are no completely effective ways for banning people. Accounts can be reregistered.
Yes, we’d need a separate solution to sockpuppet attacks, like disallowing downvotes from accounts below a karma threshold, or the one about moderator database access that’s currently in the pipeline.
Sockpuppets can vote each other up to reach any karma threshold that’s a minimum for downvoting.
Sockpuppet attacks work because accounts are allowed to downvote an unlimited amount of times without ever posting anything.
Change it so that you get 1 downvote per week if you have posted a few comments, and the sockpuppeter has to automate commenting. That’s harder for them to do than it is for the good guys to ban their puppets.
In addition, use the usual tactics against bots, such as email verification and ip banlists.
We do email verification already.
I think the suggestion in the middle paragraph would hit a lot of people that we do want voting, both up and down; lurkers with good taste make the karma signals better.
I’m not convinced that lurkers’ downvotes are good for the site, but I assume that with the current state of the database this is not a question we can answer empirically?
I think on average the votes of lurkers are less valuable than the votes of participating members. If there’s an AI that identifies lurkers with good taste because they vote similar to a moderator, their votes would be valuable.
Without a way to distinguish good taste from bad taste lurkers I think it’s more useful to count people who have a higher stake in a forum more strongly.
In general I think it’s useful to give votes of new posters little weight. I don’t think that a specific threshold number is optimal. I think a PageRank like system would produce a better result in fending of Eternal September voting issues.