Steven W. Orr wrote: > On Thursday, Apr 26th 2007 at 01:45 -0400, quoth Matt Kettler: > > =>ram wrote: > =>> Are the spammers testing some new spamtool > =>> I am getting mails with just a single word like "gushes" "using" etc > =>> > =>> what is this about now ? > =>> > =>Read the archives for more details, however the general consensus is > =>it's due to: > => > =>1) a mass run of short-emails to a broader-range of randomly generated > =>addresses in an attempt to > =>disover new ones. (aka Rumpelstiltskin attack) > => > =>- OR - > => > =>2) some spammer screwed up their template when they last pushed one out > =>to their botnet, and as a result the bots are generating emails with no > =>useful payload. > => > =>Both are quite plausible. > > Ok. I have questions: > > 1. Should I run these through sa-learn --spam or are these not to be > considered as spam? > Why wouldn't you?
Don't over-think your bayes training. If it's undesirable to you, train it as spam. If it's desirable to you, train it as nonspam. A lot of folks get caught in the trap of only trying to avoiding training "bayes poison" or "moderate" spam for fear these less obvious cases will confuse SA when it gets nonspam mail. Don't over-worry about that. The chi-squared combining algorithm is really quite good at not being fooled by tokens that appear in both kinds of mail. > 2. And also, maybe OT, should these messages be reported to SpamCop? > We all know they're spam, but to be fair, they're not trying to *sell* us > anything, thus providing a basis for not calling them spam. > > Spamcop uses UBE as their definition of spam. That's Unsolicited Bulk Email, as opposed to Unsolicited Commercial Email. http://www.spamcop.net/fom-serve/cache/125.html So, while these messages are not selling anything, thus not commercial, they are still unsolicited bulk mail. Based on that, I say they fit the critera so go ahead and report them if you like.
