Open
Bug 181953
Opened 22 years ago
Updated 2 years ago
marking postmaster mail as junk should only train on words in the enclosed message
Categories
(MailNews Core :: Filters, enhancement)
MailNews Core
Filters
Tracking
(Not tracked)
NEW
People
(Reporter: endico, Unassigned)
References
(Blocks 1 open bug)
Details
Attachments
(1 file)
(deleted),
text/plain
|
Details |
A lot of spam is adressed to non-existant addresses at mozilla.org such as anonymous@cvs-mirror.mozilla.org. The mail then bounces, generating a 'Returned mail' message informing the sender of the bounce. Except with spam, mail to the sender generally bounces so the bounce mail bounces and the whole mess goes to the postmaster. Embedded in the original message somewhere is the original spam. I'd like to mark these bounce messages as junk but I am reluctant to do this because I don't want to train mozilla to treat phrases such as 'The following addresses had permanent fatal errors' as junk. Normally the number of legitimate bounced mails that i should pay attention to is pretty tiny but i get hundreds of bounce messages containing spam. I'd like for all the bounced spam to be marked as junk but I'm afraid if I train mozilla on bounce messages then normal text in bounce mail will also be treated as junk and some day something will really break and i'll miss the legitimate bounce messages because mozilla marked them all as junk. When i mark one of these bounce mails as junk, i'd like for mozilla to only use the text in the original enclosed message in its junk filters. Finding the original message shouldn't be hard. Sendmail mailer-daemon messages contain the original mail as 'Content-Type: message/rfc822' mime attachements. This kind of logic may apply to other forwarded mail as well.
Reporter | ||
Comment 1•22 years ago
|
||
Comment 3•22 years ago
|
||
> I'd like to mark these bounce messages as junk but I am reluctant to do this
> because I don't want to train mozilla to treat phrases such as 'The following
> addresses had permanent fatal errors' as junk
This should be easy to compensate for: in addition to training it on a bunch of
postmaster bounces of junk, train it on a bunch of postmaster bounces of non-junk.
Your proposed solution really seems wrong to me, and I'm inclined to WONTFIX it.
Any folks on the CC list have comments?
Blocks: 11035
Reporter | ||
Comment 4•22 years ago
|
||
I don't think i have any postmaster bounces of non-junk.
Comment 5•22 years ago
|
||
I agree. By forcing mozilla to parse the email to determine if it is a Postmaster bounce or not, you simply open a hole for spammers to exploit. The better method is to train it on spam and non-spam postmaster bounces. The Bayesian filter will learn that the words 'The following addresses had permanent fatal errors' are INSIGNIFICANT, and do not bias the mail towards junk or non junk.
Reporter | ||
Comment 6•22 years ago
|
||
I don't yet believe your assertion that the filter will learn that the words 'The following addresses had permanent fatal errors' are insignificant. I'm sure that would be true if i had roughly equal numbers of bounces with good and junk mail. This morning i had 159 postmaster mails containing spam and zero without. I think maybe about 20 of these weren't recognized as junk. If I mark these as junk and go for weeks marking postmaster mails at the rate of 20 to one (or less) then isn't the filter going to strongly associate 'postmaster words' with junk? I don't really need mozilla to determine if a mail is a postmaster bounce message. What i really need is the ability to be specific about which message i'm marking as junk. Right now i can only mark an entire imap message as junk. Some messages contain attachments which themselves are email messages. I'd like the ability to mark just these embedded messages as junk. Maybe we could do this with a context menu when clicking on the attachment name in the attachments list. Or, make a way to extract the embedded message to a folder where i could mark the copy as junk.
Comment 7•22 years ago
|
||
how about "only filter the message selection" so that you can select the parts you want to filter and have junk filter handle the parts only?
Comment 8•22 years ago
|
||
I realize that this is a request for enhancement for Mozilla, but isn't the original problem much easier solved by making the MX for cvs-mirror.mozilla.org reject invalid recipients at time of RCPT ? Some MTA software, like Exim (which I run on protocam.com), has a configuration option for this. I see Gila is running Sendmail, and Lounge doesn't identify its server in the banner. It looks like sendmail has a ruleset local_check_rcpt for this purpose. Unless there are other situations where the feature to mark attached messages as spam might be necessary, it seems there's already a viable technical fix.
Comment 9•22 years ago
|
||
Today I had a similar situation which could be solved in the same way. It was not a postmaster bounce. I am the moderator for a few mailinglists, and sometimes people send spam to the moderated list. The mailinglist manager (ezmlm in this case) sends the spam to me for moderation. The moderation requests ends up in my spam folder however, because it contains the spam. If I untrain the message, I untrain all kinds of spam-words. So this is another case where marking part of the mail as spam/nonspam could be used to train the filter.
Updated•20 years ago
|
Product: MailNews → Core
Comment 10•18 years ago
|
||
Assigning bugs that I'm not actively working on back to nobody; use SearchForThis as a search term if you want to delete all related bugmail at once.
Assignee: dmose → nobody
Assignee | ||
Updated•16 years ago
|
Product: Core → MailNews Core
Comment 12•12 years ago
|
||
(In reply to Dawn Endico from comment #4) > I don't think i have any postmaster bounces of non-junk. In this case then, would it not make more sense to simply filter bounce messages? (In reply to Daniel Wang from comment #7) > how about "only filter the message selection" so that you can select > the parts you want to filter and have junk filter handle the parts > only? Does this serve most people's needs?
Updated•2 years ago
|
Severity: normal → S3
You need to log in
before you can comment on or make changes to this bug.
Description
•