Tuesday, August 08, 2006
End of the E-mail World, Nigh
by Tom Bozzo
I don't know if ISP-level filter parameters have been twiddled to reduce false positives or what, but I've spent a lot more time in the last few months training my Bayesian last line of defense. On the plus side, Apple's Mail does seem to eventually get the hint.
That's more than I can say for the work filtration, which like Ken's has been less than less than impressive — though I think I've bludgeoned it into accepting mail from my home accounts and from frequently e-mailed client contacts without applying a "SUSPECT:" tag or worse.
What's also fascinating is the cycles in the stuff that gets past the filters. At one point, as many of you may have seen, certain classes of spams entered territories of incomprehensibility not normally associated with measurable advertising response rates — you could take those as a sign of just how near zero the message costs were. Now, I have p3n1s-enhancing drug come-ons and penny stock advice arriving in nearly clear text. I'd suppose this might exploit some cyclical dynamics of adaptive filters, such that the gobbledygook will return before I knew I missed it. Were I not a bit overwhelmed, I'd go looking for papers.
A couple years ago, it had seemed that spam filters had briefly obtained an upper hand, albeit at the cost (in my experience) of occasionally ensnaring e-mails near the dividing line between mere junk and spam. It may in part reflect good net hygiene, but time was I'd almost never see a spam in my actual inbox. That has changed, and not for any user-behavior reason I can identify.
I don't know if ISP-level filter parameters have been twiddled to reduce false positives or what, but I've spent a lot more time in the last few months training my Bayesian last line of defense. On the plus side, Apple's Mail does seem to eventually get the hint.
That's more than I can say for the work filtration, which like Ken's has been less than less than impressive — though I think I've bludgeoned it into accepting mail from my home accounts and from frequently e-mailed client contacts without applying a "SUSPECT:" tag or worse.
What's also fascinating is the cycles in the stuff that gets past the filters. At one point, as many of you may have seen, certain classes of spams entered territories of incomprehensibility not normally associated with measurable advertising response rates — you could take those as a sign of just how near zero the message costs were. Now, I have p3n1s-enhancing drug come-ons and penny stock advice arriving in nearly clear text. I'd suppose this might exploit some cyclical dynamics of adaptive filters, such that the gobbledygook will return before I knew I missed it. Were I not a bit overwhelmed, I'd go looking for papers.
Comments:
<< Home
with the proliferation of spyware it doesn't matter how discrete you are with your email address. one of your friends has spyware on their computer running Windows Me. the spyware sends his address book out, and voila, you're sunk.
# posted by 
: Tue Aug 08, 08:12:00 PM
: Tue Aug 08, 08:12:00 PM
True enough, though last time I checked out my mother's XP system, it was remarkably free of unwanted code -- the possible state of my grandmother's Win98 system frightens me, though.
Amazingly, though, I get almost no spam at the gmail account, even though most of my correspondence has shifted there, and my address is not that well hidden in the sidebar.
Amazingly, though, I get almost no spam at the gmail account, even though most of my correspondence has shifted there, and my address is not that well hidden in the sidebar.
I've notice that most of the SPAM that slips through my filters these days consists entirely (or mostly) of attachments, which means that there is no text for the filters to act on.
Mrs. C.: Upon closer inspection, the cleartext penny stock come-ons are graphics attached to the spam message, hence as you say no text for conventional filters to analyze (though some may decide that an attachment w/o body is suspect, which has false positive problems). The obvious solution is to set these second CPU cores (or third or fourth, in the Mac Pro [drool]) all the new computers have to the problem of OCR'ing image attachments and analyzing any resulting text for spam. It's not that hard of a computing problem for the amount of CPU horsepower that now goes idle...
Post a Comment
<< Home
