From: spambayes-bounces+rcoe=cambridgema.gov@python.org [mailto:spambayes-bounces+rcoe=cambridgema.gov@python.org] On Behalf Of skip@pobox.com Sent: Wednesday, August 02, 2006 11:55 PM To: Peter Barker Cc: spambayes@python.org Subject: Re: [Spambayes] Spam in Images
Peter> For my spam and non-spam, a good indicator is that I very seldom Peter> receive non-spam messages with a .gif image attached (attachments Peter> are usually .jpg or various document types). And if a wanted mail Peter> has a .gif attachment it has much more text than the usual Peter> gibberish in the spam messages (because it is usually just a Peter> company logo or similar, and not essential to the message). So if Peter> spambayes can score attachment type and text size it may help.
True, but for those people with correspondents who do send them mail with image attachments ("Subject: Cute pictures of my new granddaughter"), the presence or absence of images may fall around the middle and thus either not be used at all, or only provide a negligible bump in one direction or the other.
But Peter's point is that a picture of somebody's granddaughter won't be a GIF image; it will be a JPEG. Digital cameras don't use GIF, and hardly anybody else does either. Except spammers, who presumably care less about image quality than about file size. I suppose the result of scoring GIF images as spam will be that spammers will switch to JPEG. But anything that makes their lives more complicated is all to the good.
Bob> But Peter's point is that a picture of somebody's granddaughter Bob> won't be a GIF image; it will be a JPEG. Digital cameras don't use Bob> GIF, and hardly anybody else does either. Except spammers, who Bob> presumably care less about image quality than about file size. Sure, but just like everything else, the spamminess of specific image types is highly user-specific. I dont have any granddaughters (yet, thank God), so at best JPEGS are a wash. GIFs and PNGs look spammy though: % spamcounts -r image/ token,nspam,nham,spam prob content-type:image/gif,64,2,0.885801412431 content-type:image/jpeg,3,1,0.5 content-type:image/png,5,0,0.95871559633 Still, the image type is just one token. Almost no matter how strongly spammy it is, it's likely to be overwhelmed by whatever other tokens exist in the message. Skip
participants (2)
-
Coe, Bob -
skip@pobox.com