4 ms·

Am I missing something with the second formula? h = (f - v)/(s * t) The perfect flagger would have v = 0 and f = s. His total flag score is t = s + x, where

by eric-hu 15y ago

Am I missing something with the second formula? h = (f - v)/(s * t) The perfect flagger would have v = 0 and f = s. His total flag score is t = s + x, where x is the number of non-honeypots flagged. His score would be: h = s/(s * s + s * x) = 1/(s + x) As s => infinity, h => 0. This score would actually punish a good flagger over time, no? A perfect veteran flagger with 100 for 100 honeypots flagged would have a lower honeypot ratio than a perfect newbie flagger who's 5 for 5 honeypots flagged.

tansey 15y ago
Correct you are! I've updated the article with a fixed version of this formula. This version should in fact rank the 5 for 5 guy slightly lower than the 100 for 100 guy. It's still not perfect but I believe it does its job in theory. Thank you for pointing that out. :)
- eric-hu 15y ago
  You're welcome! The current version looks better. I would add that the flag adjustment should also factor in honeypots seen. This version will punish flaggers (good or bad) with a large flagging history who've seen no honeypots. -(1-(f/(t+1)) goes to zero as t goes to infinity--desirable when someone has seen and not flagged honeypots, but not as desirable when someone hasn't (which would be any user with many flags at the time of algorithm implementation). I went back and read your footnotes. Footnote [1] is indeed a linkbait-y article. To me, it demonstrates a behavior described in another comment here: upvote as a save function. The title looks interesting, and in the middle of a work day, one may not have time for a long article. There's even more incentive to use it as a save as the front page volume cycles more. Personally, I think that this 'noise' in upvote value can be mitigated by adding a separate save function and perhaps even eliminating an upvote history visible to the user (migrating current upvote history over to save history first so users can still access their clippings).

by eric-hu 15y ago

tansey 15y ago

Correct you are! I've updated the article with a fixed version of this formula. This version should in fact rank the 5 for 5 guy slightly lower than the 100 for 100 guy. It's still not perfect but I believe it does its job in theory. Thank you for pointing that out. :)

eric-hu 15y ago
You're welcome! The current version looks better. I would add that the flag adjustment should also factor in honeypots seen. This version will punish flaggers (good or bad) with a large flagging history who've seen no honeypots. -(1-(f/(t+1)) goes to zero as t goes to infinity--desirable when someone has seen and not flagged honeypots, but not as desirable when someone hasn't (which would be any user with many flags at the time of algorithm implementation). I went back and read your footnotes. Footnote [1] is indeed a linkbait-y article. To me, it demonstrates a behavior described in another comment here: upvote as a save function. The title looks interesting, and in the middle of a work day, one may not have time for a long article. There's even more incentive to use it as a save as the front page volume cycles more. Personally, I think that this 'noise' in upvote value can be mitigated by adding a separate save function and perhaps even eliminating an upvote history visible to the user (migrating current upvote history over to save history first so users can still access their clippings).