On Thu, 17 Nov 2016 20:49:49 -0500 Alex wrote: > Hi all, > > We have a lot of users who use email to share photos. Empty body, 2M > JPG attachment, nothing in the subject. > > How do you train bayes for these?
Same as anything else, they still have header tokens which may help, particularly for ham from low spam servers. > Is 3.0 a suitable score for BODY_EMPTY? Note that this shouldn't apply to the image-only emails: meta BODY_EMPTY __EMPTY_BODY && ... && !__MIME_ATTACHMENT There are several other exceptions.
