[mostlylurks]

The presented trichotomy between no moderation, moderation, and federated moderation is false.

Moderation can also be accomplished via a user-level web-of-trust system, where each user can choose who to trust as a moderator, and this trust can propagate recursively to the people trusted by the people you trust, and at each level (even when manually choosing people to trust) this trust can be fuzzy (not full trust vs no trust, but potentially something in between those two), and rapidly decreasing the more distant you get from those you've manually chosen to trust. To solve the issues of spam, censorship, and convenience simultaneously, you simply assign to users some moderators on the trust list by default and allow users to opt out of that trust.

This approach is also applicable in the same manner to the similar problem of curation (i.e. choosing what to highlight instead of what to hide), where the same four approaches are also applicable with largely the same pros and cons.

[ctrw]

I tried building a reddit style baord using that. The main issue became that calculating weights was more expensive than everything else combined. At one extreme you have the trust matrix which you just multiply with itself to get the nth hop scores, at the other you had the linked list graph traversal. Neither were good solutions.

With how much matrix multiplication were doing for machine learning using the matrix approach now might be feasible.

[simcop2387]

That might work nicely with a lot of the sparse matrix research going on too, since I suspect that the trust matrix should be very sparse

[zizee]

If I am understanding correctly, did you consider the option of calculating weights in the client?

[ctrw]

I would be leaking all social graph data and all voting data to each person using the service. So it's a no go from the start.

[TeMPOraL]

The problem description sounds vaguely PageRank-y, so maybe similar trick can be applied?

[Mentlo]

For this to work you'd need both a high coverage of users providing feedback signal (not happening in real cases), a low penalty of initial no-moderation to allow the system to find equilibrium, and relatively high time-invariance of system to ensure the penalty doesn't recur.

The point of the vehicle in the park game is that complexity isn't always reducible to a tractable problem. Which is fine, and we should learn to engineer systems that embrace the fuzziness, rather than assuming the problem is tractable and solvable.

[pdonis]

Does the system you describe actually exist and work anywhere?

[lr4444lr]

I'd say it works quite well here. The author is assuming disagreement on HN is some kind of failure. I don't see it that way at all, and sometimes even make comments on opposite sides of an issue within the same issue just because I recognize compelling points in a controversial topic. It's the topics that get monopolar responses that are the least useful and interesting to me.

[pdonis]

> I'd say it works quite well here.

Huh? There are only two moderators here and they aren't chosen by users, nor do users have to make any choice about "trusting" them. The HN moderation system bears no resemblance at all that I can see to what the post I responded to was describing.

[CamelCaseName]

Who's the second moderator? Dang and?

[pdonis]

Oh, that's right, dang is the only moderator now. There used to be two.

[lmz]

Sounds somewhat like https://en.wikipedia.org/wiki/Advogato#Trust_metric - not sure how well it actually worked.