4 ms·

Maybe a better approach would be to silently gather statistics and find out the degree of correlation between any given user's "inappropriate" rating and the ac

by icky 19y ago

Maybe a better approach would be to silently gather statistics and find out the degree of correlation between any given user's "inappropriate" rating and the actual (as determined by you) inappropriateness of any given item. Then, if someone has, say, a NEGATIVE correlation between their inappropriateness rating and the actual inappropriateness of said item, you can still use that person's ratings as a useful source of information... just not the way they intended ;-)

rzwitserloot 19y ago
Extra points for creativity but that assumes that 'content is inappropriate' abusers do not use the 'mark inappropriate' feature for actual inappropriate content. I don't have any quotes or statistics here, but common sense suggests to me that abusers will mark real inappropriate stuff and stuff they just don't agree with. By keeping mum, you also lose the scare effect, though OTOH that might keep the "I'll just create a bunch of accounts" folks down. A third risk is that legitimate marks no longer get made because the potential marker doesn't want to risk the ban. However, if the ban is ONLY on the use of the mark as inappropriate button and not for anything else, then I doubt it would actively scare users.
- icky 19y ago
  > By keeping mum, you also lose the scare effect, though OTOH that might keep the "I'll just create a bunch of accounts" folks down. Exactly: if someone is making consistently bad flaggings on one account, you don't want to tip them off and have them flee to a new account. > A third risk is that legitimate marks no longer get made because the potential marker doesn't want to risk the ban. However, if the ban is ONLY on the use of the mark as inappropriate button and not for anything else, then I doubt it would actively scare users. You'll note that my proposal does not include a ban of anything, not even the "mark as inappropriate" button. I just suggest letting the bad users keep putting in their bad ratings, and using that information in the opposite manner as they intended. You could even have some sort of Bayesian (naive or Markovian, depending on your computing resources and scaling needs) classifier set up to see what words or phrases in a post make any given user more likely to mis-rate. Then you could separate out a user's probably-bad ratings from their probably-good ones.

by icky 19y ago

rzwitserloot 19y ago

Extra points for creativity but that assumes that 'content is inappropriate' abusers do not use the 'mark inappropriate' feature for actual inappropriate content. I don't have any quotes or statistics here, but common sense suggests to me that abusers will mark real inappropriate stuff and stuff they just don't agree with. By keeping mum, you also lose the scare effect, though OTOH that might keep the "I'll just create a bunch of accounts" folks down. A third risk is that legitimate marks no longer get made because the potential marker doesn't want to risk the ban. However, if the ban is ONLY on the use of the mark as inappropriate button and not for anything else, then I doubt it would actively scare users.

icky 19y ago
> By keeping mum, you also lose the scare effect, though OTOH that might keep the "I'll just create a bunch of accounts" folks down. Exactly: if someone is making consistently bad flaggings on one account, you don't want to tip them off and have them flee to a new account. > A third risk is that legitimate marks no longer get made because the potential marker doesn't want to risk the ban. However, if the ban is ONLY on the use of the mark as inappropriate button and not for anything else, then I doubt it would actively scare users. You'll note that my proposal does not include a ban of anything, not even the "mark as inappropriate" button. I just suggest letting the bad users keep putting in their bad ratings, and using that information in the opposite manner as they intended. You could even have some sort of Bayesian (naive or Markovian, depending on your computing resources and scaling needs) classifier set up to see what words or phrases in a post make any given user more likely to mis-rate. Then you could separate out a user's probably-bad ratings from their probably-good ones.