[spambayes-dev] [ spambayes-Feature Requests-762783 ] TMDA capabilities

Tue Jul 1 09:36:11 EDT 2003

Feature Requests item #762783, was opened at 2003-06-29 12:33
Message generated for change (Comment added) made by montanaro
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=762783&group_id=61702

Category: None
Group: None
Status: Open
Priority: 5
Submitted By: Why do you need this (beyond-thoughts)
Assigned to: Nobody/Anonymous (nobody)
Summary: TMDA capabilities

Initial Comment:
I just switched from TMDA (http://tmda.net/) to Spambayes.
Watching emails that were neither in black- or
whitelist was to annoying. Of course those people get a
reply (please see at TMDA-Homepage to see how it works)
but about 50% don't understand the reply-mail. (I don't
know why -- the text is very clear!).
Spambayes doesn't queue emails that can't be classified
-- you receive them as "unsure".

I think combining both techniques would cut down wrong
positive and wrong negative and the "unsure" numbers.
Additionaly it might be useful for training having a
definite black- and whitelist.

So what are good points (files, classes, methods) to
add such a feature?
Is it dependent whether pop3proxy, hammie, ... is used?

Thank You,
   Christoph Becker-Freyseng

----------------------------------------------------------------------

>Comment By: Skip Montanaro (montanaro)
Date: 2003-07-01 10:36

Message:
Logged In: YES 
user_id=44345

Whitelist functionality really isn't needed.  Spambayes already
tokenizes email addresses, so over time it effectively builds up
a whitelist for you.  Here are some examples from my current
training database (the tuple values are (nspam, nham)):

>>> db["email name:itineraries"]
(71, 10)
>>> db["email name:guido"]
(0, 8)
>>> db["email name:webmaster"]
(58, 45)
>>> db["email name:skip"]
(365, 314)
>>> db["email name:support"]
(136, 13)

Note that itineraries at mojam.com & support at microsoft.com
are frequently forged in mail I receive.  webmaster at mojam.com
and skip at pobox.com are forged a fair amount, but are also
frequently correct.  On the other hand, nobody has so far taken
Guido's name in vain in my incoming email.  (I rarely train on
Python-related email, so there are only a few messages from
Guido in my training database.)

Even if you implemented such a feature it would probably not
be as sensitive as the current tokenizing scheme.  In addition, you
would still have to scan your spam.  You will eventually get a
valid email message from someone not on your whitelist.

Regarding:

    Submitted By:
    Why do you need this (beyond-thoughts)

it's because (in general) too many people submit incomplete bug
reports anonymously and then can't be contacted to complete
their report.  This was a significant problem with the Python
project and sort of carried over to the Spambayes project.

----------------------------------------------------------------------

Comment By: Why do you need this (beyond-thoughts)
Date: 2003-07-01 09:29

Message:
Logged In: YES 
user_id=186848

I know that TMDA and Spambayes have different approaches,
but this is what makes it useful combining them.
With Spambayes till now I had no false classified emails
(just few "unsure") so I'm quite satisfied with it. However
I still have to check the emails in "Spam-Folder" because I
don't want to risk loosing some false-classified important
email.
When Spambayes had some TMDA capabilities it could just sent
a confirmation-email to the asumed spammer. If he doesn't
reply then it's really his fault. So I won't have to check
emails in the "Spam-Folder" at all.
On the other side people I send emails to could be
automatically added to a white-list so they surely won't
have trouble with Spamfilters.
I have thought of some more configurable rules that could be
added making Spambayes an interactive-AI-Spamfilter.

I'd like to make a demo-implementation but I need some
starting points. Especially I don't know how the Outlook
stuff works. But maybe I should just try implementing such a
thing for the pop3- and smtp-proxy.

Thanks,
   Christoph Becker-Freyseng

----------------------------------------------------------------------

Comment By: Anthony Baxter (anthonybaxter)
Date: 2003-06-30 20:38

Message:
Logged In: YES 
user_id=29957

I can't imagine that this feature would ever be added to
spambayes. It's a completely different approach to
spam-filtering, with almost nothing in common with the
existing approach. Spambayes will gradually improve as you
train it further - the initial flurry of unsures is probably
just insufficient training. Note also that you can adjust
the cutoffs to end up with more or less unsures.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=762783&group_id=61702