Michael Barnes wrote: > What you are asking about is manual learning in the event of an error > by SA. > > Unfortunately, once the mail gets to a user (depending on their > computer skills), its pretty much gone.
*Especially* if they're using Outlook. Ugh. :( > What I mean, is that to feed the mails back to the bayes learning > process once it has gotten to the user is that the user somehow has > to get _the whole message, headers and all_ somewhere to be fed to > sa-learn. > > I work with pretty bright people, many are in graduate school for > computer oriented, but I would never ask for an original mail back > from one of them because it would be too difficult with the multitude > of (usually broken) mail readers out there. > > Part of the email rfc (I guess 822, not sure if its actually in > another rfc) contains a feature called "resending a message" or > similar. My mailer, mutt, has this feature, and it describes it as: [snip] Pegasus Mail supports a "Bounce" feature that pretty much resends the message, including all existing headers and body content (MIME goop and all). The major disadvantage of this is that it produces a message with additional Received: headers, and depending on the email path other bits of the message headers may get changed. :/ > Unfortunetly, I have only heard of one other old obsolete mailer > that has this feature. I'm sure there are others, but its not too > common. *nix-based mailers are far more likely to have this than Windows-based ones. That said, I've asked (and asked, and asked, and PLEADED with) users to forward mail as an attachment - which (from just about anything except Eudora and Outlook) gets me *exactly* the message the user receives. It's a little more work to untangle the attached message, but I've had *no* trouble feeding these into sa-learn. "Right-click, forward as attachment" works pretty well. -kgd -- Get your mouse off of there! You don't know where that email has been!