-------- Original-Nachricht --------
> Datum: Fri, 31 Jul 2009 09:38:50 +0200
> Von: Sebastian Toepfer <[email protected]>
> An: [email protected]
> Betreff: Re: [Dspam-user]     Dealing SPAM - extended feature request?

> On Fri, 31 Jul 2009 01:28:02 +0200, "Steve" <[email protected]> wrote:
> > -------- Original-Nachricht --------
> >> Datum: Thu, 30 Jul 2009 20:40:50 +0200
> >> Von: "Sebastian Toepfer" <[email protected]>
> >> An: [email protected]
> >> Betreff: Re: [Dspam-user] Dealing SPAM - extended feature request?
> > 
> >> Hallo,
> >> 
> > Hallo Sebastian,
> > 
> Hallo,
> > 
> >> I'm bad and make a top post.
> >> 
> > Ab in die Ecke und schäme Dich!!!
> > 
> erledigt.
> > 
Aha! Wireless keyboard in the corner and you type from there? Or long cable? :) 
:)


> > 
> >> I think it was nice when a user can do the following:
> >> low spam rate => tagging mail - not learn(toe)
> >>
> > dspam_admin change preference [email protected]
> > "trainingMode" "TOE"
> > 
> > 
> >> medium rate => quarantine - learn(teft?)
> >>
> > dspam_admin change preference [email protected]
> > "trainingMode" "TEFT"
> > 
> > 
> >> high rate => drop mail (deleted or so should configured by admin,
> because
> >>
> >> statutory provisions - think gobal quarantine was nice,  so admin can 
> >> recovery mails) - not learn
> >> 
> > The drop mail part is not build into DSPAM. But doing that with
> something
> > like procmail, maildrop, sieve or other tools is no big issue.
> > 
> No, not for all mails for the given spamrate only.
> 
> Nicht für alle emails, nur für die konfigurierte spamrate. Wenn ich
> micht
> recht entsinne hattest du mal erklärt das dein Setup an den Grenzen lernt
> und sowas ähnliches stelle ich mir hier auch vor.
>
Aha. Yes. My training script uses a thickness border in each class and I use 
that thickness border (or factor) to enforce training if a message is scoring 
into the correct class but did not score at least "thickness" above the 
decision threshold of the class.

It's rather something you only would use on a verified corpus but not in live 
processing of messages. The only situations where I could think to use 
something like that in live processing would be where we would have a neural 
network of other DSPAM instances and should a message score in a certain 
thickness border (ranging into both classes (HAM/SPAM)) then we could ask the 
other nodes in our neural network how they would classify the message in 
question and then use their combined result for making a decision. Or we could 
use a thickness border (again like above ranging in both classes) and should a 
message hit into that thickness border we could use semi statistical data to 
get a higher confidence/probability of our classification. Such an 
semi-statistical data could be RBL lookups, trusted path of delivery (aka: [IP 
of sender]->[IP of next hop from the viewpoint of sender]->[next hop]->[next 
hop]->[next hop etc...]->[IP of your MTA] and use that combination in such a 
way like we use now the automatic whitelisting), Whitelist Lookups (aka: WL), 
ASN BL/WL or doing things like recently mentioned in the study about SNARE 
(Spatio-temporal Network-level Automatic Reputation Engine) -> 
http://www.technologyreview.com/communications/23086/.


> High Rate spam ist
> immer
> wirklich spam so drop it (speichere ausserhalb der user-quarantine),
>
I would not confirm that. What is an external indicator (without reading the 
mail) that a mail is SPAM? The confidence factor? The probability? If that 
would be so easy then we would never have false positives and/or false 
negatives.


> Medium
> spam kann auch mal in selten Fällen Ham sein also soll der Benutzer
> entscheiden ob es richtig ist - aber nur wenn er Zeit hat (also
> quanrantine) und dspam sollte es sicherheitshalber als spam lernen. Low
> spam würde ich dann gern immer zum user leiten so das er sofort
> entscheiden kann was es ist und wenn ich es richtig verstanden habe wäre
> es in diesem fall besser dspam lern die mail nur wenn es falsch ist also
> toe? Natürlich sollte das alles der Benutzer selbst bestimmen können,
> aber das ist das was ich, als benutzer, gern hätte ... ausser es macht
> keinen sinn!
> 
> sorry for the german explanation. 
> 
> Think Julien has the better concept for this.
> 
> > 
> >> What low,medium or high rate is should be configurable by user.
> >> 
> > Open a feature request if you are really into something like that. For
> now
> > this functionality is not build into DSPAM.
> > 
> Will hear it's a feature was make sense first?
> > 
> >> What do you think, makes sense?
> >> 
> > To me? All is already there beside the last requirement.
> > 
> > 
> >> Sebastian
> >>
> > // Steve
> > 
> Sebastian
> 
Steve


> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008
> 30-Day 
> trial. Simplify your report design, integration and deployment - and focus
> on 
> what you do best, core application coding. Discover what's new with 
> Crystal Reports now.  http://p.sf.net/sfu/bobj-july
> _______________________________________________
> Dspam-user mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/dspam-user

-- 
Neu: GMX Doppel-FLAT mit Internet-Flatrate + Telefon-Flatrate
für nur 19,99 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Dspam-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dspam-user

Reply via email to