Ian A B Eiloart wrote:
> --On Wednesday, March 24, 2004 10:21 am +0000 Bruce Richardson
> <itsbruce@???> wrote:
>
>> On Wed, Mar 24, 2004 at 09:36:16AM +0000, Ian A B Eiloart wrote:
>>
>>> > If greylisting becomes common, it stops being a low-scale problem and
>>> > they deal with it. Shouldn't be hard.
>>>
>>> Yes, but that is an argument that you could use against ANY anti-spam
>>> proposal. Its a little like telling a medieval knight that body
>>> armour is
>>> useless because one day someone will invent guns. This is an arms race -
>>> we can't win or lose, we can only stay ahead or drop behind.
>>
>>
>> No, it is not the same. Neither SpamAssassin nor DSPAM is going to
>> become obsolete overnight. Greylisting may well - it's a simple
>> one-trick pony.
>
>
> But Baysian filters have their limitations. Researchers already have lists
> of words that these filters tend to like, and techniques for refining those
> lists on a per filter basis.
>
Do you have any references for that?
The only reports I've read have been for a very specific set of words. The
researcher used their own filter and repeatedly refined the search based on
the results of that filter. I've not seen anything involving a set of words
which could be got through a generic Bayesian filter - after all, everyone's
mail is different and the number of words and occurence of each in the
spam/ham list will be different.
The only time I could see this approach working at all is for a system which
used a standard Bayesian database. No good spam filter would do that, and
with subsequent training the statistics/database would alter anyway.
--
Nigel Wade, System Administrator, Space Plasma Physics Group,
University of Leicester, Leicester, LE1 7RH, UK
E-mail : nmw@???
Phone : +44 (0)116 2523548, Fax : +44 (0)116 2523555
