[Spambayes] Randomized Spam Beating SpamBayes (SpamBayes: message 14 of 20)

[Quinn]

> > For me they can be anything from 0% to about 55%. 
> > It might have some effect that I only sent an
> > excerpt; they vary in length.
> 
> 90% of mine are stock spam with an attached image (which is blocked by
> Outlook). Most get filtered. Some of them (about 20 per day these last
> couple days) get less than 5% of a spam score in SB.

Yeah, the ones with images get snagged fairly easily; I think having an
embedded image has gotten a high weighting in SB; almost none of my ham has
images.  I haven't seen one make it through to the inbox, at least.  Some
end up in the suspects folder instead of the junk folder, though.

> This is the first time since installing spambayes (about 2 years ago?)
> that spam is making it into my inbox several times per day. All of the
> spam that makes it into my inbox is the gibberish crap spam that, I
> believe, is intended to damage bayesian filters. Most messages are
> determined to have a zero spam possibility percentile by SB, 
> which until about a month ago (even with SB 1.0.4) just never happened. 
> My spam was ALWAYS filtered right. Now, with these new gibberish spams
> I think they're doing quite well at damaging the legitimacy of the 
> database.

Do you think using SB to delete them makes it better or worse?  I have been
thinking about deleting them normally so they don't dilute my definitions,
since I don't think there's any useful pattern in them for SB to twig on.

Does SB count numbers of times phrases are repeated?  That might help for
the ones I'm getting; they usually use the same fairly long sentence or
chunk at least two or three times.

Quinn
~~~~~~~~~~
A person's maturity consists in having found again the seriousness one had
as a child, at play. - Friedrich Wilhelm Nietzsche
~~~~~~~~~~