Spam filtering

Carter Schonwald carter.schonwald at gmail.com
Sun Apr 17 13:34:32 UTC 2016


Question: are we talking captcha or REcaptcha?
My understanding is that REcaptcha is better than old school captcha

Have we evaluated it as an option?
http://www.google.com/recaptcha/intro/index.html

On Saturday, April 16, 2016, Ben Gamari <ben at well-typed.com> wrote:

> Niklas Hambüchen <mail at nh2.me <javascript:;>> writes:
>
> > Hi Ben,
> >
> > Could we not have a captcha instead of a reject, to avoid false
> > positives? That would require no training.
> >
> > Since I assume most Trac spammers are extremely unsophisticated, a
> > simple hardcoded question like "What programming language is GC all
> > about?" may be sufficient.
> >
> The CAPTCHAs being broken are the reason why this incident occurred.
> I have added some more CAPTCHAs to try to dilute the pool of answers
> that they already know, but they still seem to solve them easily
> enough regardless. I can only imagine they have some sentient beings
> sitting at computers solving CAPTCHAs.
>
> I don't really feel like we can make the CAPTCHAs themselves any more
> difficult without excluding real new users, which I really want to avoid.
>
> Regardless, my goal here is to error on the side of less filtering, not
> more, even if this does mean more manual maintenance. To this end, I've
> configured the filters such that the probability of legitimate activity
> being suppressed should be negligible,
>
>  * I've been careful to only train the Bayes filter on obvious spam;
>    I have tested it against various snippets from the wiki and mailing
>    list and have yet to see it score anything legitimate with a spam
>    likelihood > 5%.
>
>  * Even if the Bayes filter does deem your content to be spammy enough
>    to warrant further attention, you will merely be asked to solve a
>    CAPTCHA. Posts will not be outright rejected unless it is quite clear
>    that they are spam.
>
> I am optimistic that the filtering will have negligible effect on
> legitimate traffic. As a smoke test I managed to create a new account,
> open a new ticket, and start a new Wiki page without even needing to
> solve a CAPTCHA.
>
> Cheers,
>
> - Ben
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20160417/3ca1002a/attachment.html>


More information about the ghc-devs mailing list