[sldev] Research on flagging/rating systems? (Re: Your Feedback Wanted on Search Flagging !)

Celierra Darling Celierra at gmail.com
Sun May 4 16:27:03 PDT 2008


On Sun, May 4, 2008 at 5:46 PM, Jason Giglio <gigstaggart at gmail.com> wrote:
>  ... Also it wouldn't be too hard for
>  someone to implicitly approve a few thousand messages with a bot, since
>  most of them won't be incorrect (most entries are legit), and boost a
>  bad weight that way.
>

If you bot-approve a few thousand messages and none of them are
controversial, then you don't get any extra weight because the
Lindens/volunteers will never get around to those.

Alternatively, you can also explicitly forgive more/less based upon
how controvercial the case was, with a max gain of 1/r down to a
factor of only 1 for no-controversy cases.

On Sun, May 4, 2008 at 5:46 PM, Jason Giglio <gigstaggart at gmail.com> wrote:
> Argent Stonecutter wrote:
>  > Something else: acceptance or mere inaction should not imply approval.
>  >   * A resident may disapprove of the whole system, or only use it to
>  > flag spam.
>  >   * A resident may merely be reluctant to use the system.
>  >   * A reluctant user's flag would seem to be more important.
>
>  I'm not sure that's a bad thing.  If you hate the system so much that
>  you never use it, yet you still find an entry objectionable enough to
>  flag, that entry must be pretty bad.
>
>  Really I'd be more worried about there being no server-side way to
>  determine this implicit approval.

I think this is generally a problem about how to define an approval
vote...  My first instinct was to just define it such that the most
utilized entries are harder to flag, but yes, then an implicit
anti-flag shouldn't have a major penalty for those not very willing to
participate in the system....  But whatever definition is eventually
used, there needs to be *some* penalty for incorrect approval votes,
because otherwise bots would be able to continually approve of bad
things to keep things them away from the top-n list.

That reminds me, I forgot something... as I had defined things,
occasionally one would need to check a list that's something like the
top-n by min(weight of 0s, weight of 1s), to watch for abuses of bots
trying to keep something out of suspicion (by flooding with
approvals).

-- 
~Alex and Cel


More information about the SLDev mailing list