> That works well for about two weeks until users figure out how to exploit this. They form groups that agree to coordinate to report content. Now reports become lower signal than before, but still somewhat useful. We use the reports, but try to limit exploitation.
What about human employees reviewing users reports, especially randomly picked samples from those that resulted in successful takedowns. If the content was found to be sound under the rules, penalise the reporters. Suspend their ability to make reports (probably in the form of silently ignoring them).
Ie some human moderation should still be done at this scale, but using a combo of random sampling and system-flagged-suspicious.
This is Slashdot's meta-moderation system from 20(?) years ago? Logged in users would randomly be selected to check moderation, and accounts that abused their mod points were flagged. Granted, you probably couldn't do that today as you could see meta-moderation brigades out in the wild. Employees should be better, although there's still the chance that an employee might use their position to punish opinions they disagree with.
Yeah I remember slashdot days. Having employees do the meta moderation is key imo, and sounds like would need a second level of meta moderation to review samples of the employees actions.
I’m also very pro a pay for final review, where eg you pay $50-100 or something around that level of inconvenience to have a multi-employee review (each employee reviewing it independently, unknownst to each other, and taking the majority decision).
I remember Dota2 (video game) did this, people fake reporting had less reports and they counted for less, while people that were consistently reported players that got punished got more reporting power. Only one report per match "counted" too so a group of player couldn't gang on reports on one player.
Of course that could be abused as well,but you'd have to make a group of people that first got the good rating then reported same people and that would be significantly harder
What about human employees reviewing users reports, especially randomly picked samples from those that resulted in successful takedowns. If the content was found to be sound under the rules, penalise the reporters. Suspend their ability to make reports (probably in the form of silently ignoring them).
Ie some human moderation should still be done at this scale, but using a combo of random sampling and system-flagged-suspicious.