Bayesian training [Was: Spam From Idiots]

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 08:45:34



Me too. But you just got me thinking about my SpamAssassin bayesian
training.


I think the above type of emails, at least the ones I have been getting
where the body is just a bunch of random phrases, are meant to throw off
learning/training type spam filters like bayesian ones.

I'm pretty sure you don't use SA, and I know you use a C/R system, but
please indulge me.

Those emails themselves are not spam (at least by my definition since
they are not selling anything, although they are certainly unsolicited
bulk emails).

So my question is this - should I pass those type of emails on to SA's
bayesian training system, or just delete them?

[snip]

<Sorry to hijack your post>

--
Troy Piggins
,-o Ubuntu v5.10 (Breezy Badger): kernel 2.6.12-9-386,
o ) postfix 2.2.4, procmail 3.22, mutt 1.5.11i,
`-o slrn 0.9.8.1/rt (score_color patch), vim 7.0
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Sam » Thu, 11 May 2006 10:03:51

This is a MIME GnuPG-signed message. If you see this text, it means that
your E-mail or Usenet software does not support MIME signed messages.
The Internet standard for MIME PGP messages, RFC 2015, was published in 1996.
To open this message correctly you will need to install E-mail or Usenet
software that supports modern Internet standards.

Troy Piggins writes:


No, he doesn't. Beavis makes a lot of noise about C/R, but in
http://www.yqcomputer.com/ , suffering cat out of the bag:

Beavis> Note that the filter I use for Usenet mail (see below) is
Beavis> not a C-R system.

That's the best testament as to the effectiveness of a C/R system.


Beavis is the last person you want to ask for any technical advice. Here's
a brief capsule summary of Beavis's previous technical tips:

http://www.yqcomputer.com/
Explorer

http://www.yqcomputer.com/

http://www.yqcomputer.com/ "hack my C:\
drive"

http://www.yqcomputer.com/ , the virus and trojan expert

See the Beavis FAQ that's regularly posted to this newsfroup.



-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)

iD8DBQBEYTv3x9p3GYHlUOIRAvqzAJ9cWRi7rec2Zg9GRbQFOOGvLuMAiwCdGShv
+/O60oscbiuf2bVAQ89moc8=
=+ylo
-----END PGP SIGNATURE-----

 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 10:34:13


[snip]
[snip]

I wasn't asking him directly. I would've emailed him for that.
I posted on USENET which is a public forum. Yes I know some of the
phrases I posted referred to "his" system etc, but anyone can
read and followup here as you have done.

I was hoping for someone with some knowledge about bayesian systems to
reply. That's why I changed the subject heading. In hindsight maybe I
should've started a new thread but I got lazy.

Do you know anything about bayesian training in relation to my question?


I haven't seen those deliberately provocative and antagonistic posts of
yours for some time since they have nothing to do with this group.

--
Troy Piggins
,-o Ubuntu v5.10 (Breezy Badger): kernel 2.6.12-9-386,
o ) postfix 2.2.4, procmail 3.22, mutt 1.5.11i,
`-o slrn 0.9.8.1/rt (score_color patch), vim 7.0
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Kari Hurtt » Thu, 11 May 2006 12:56:56

Troy Piggins < XXXX@XXXXX.COM > writes:



If SpamAssasin is automatically learned them as non-spam ('ham'), then
at least you want learn them to spamassasin as spam.
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 13:22:22


They are not learned as ham. I have passed them to sa-learn as spam to
date, but I am not sure I should be doing this. They are not really
spam, they are just random phrases/words set to throw off spam trainers.

Sample body:
-----8<-----
Tomorrow begins the last week of Autumn, said Thorin one day.
And winter comes after autumn, said Bifur.
And next year after that, said Dwalin, and our beards will grow
till they hang down the cliff to the valley before anything happens
here. What is our burglar doing for us?
Since he has got an invisible ring, and ought to be a specially
excellent performer now, I am beginning to think he might go through the
Front Gate and spy things out a bit!
Bilbo heard this-the dwarves were on the rocks just : above the
enclosure where he was sitting-and Good Gracious! he thought, so that
is what they are beginning to think, is it? It is always poor me that
has to get them out : of their difficulties, at least since the wizard
left. Whatever am I going to do? I might have known that something
-----8<-----

--
Troy Piggins RLU#415538 < http://www.yqcomputer.com/ >
My dogma just ran over my karma
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Kari Hurtt » Thu, 11 May 2006 14:04:11

Troy Piggins < XXXX@XXXXX.COM > writes:



SpamAssasin learns also from mail headers, so it is still possiblity that
SpamAsasin learns something usefull.

/ Kari Hurtta
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 15:36:12


Ok, thanks Garen and Kari. Will do some reading/research.

--
Troy Piggins RLU#415538 < http://www.yqcomputer.com/ >
My dogma just ran over my karma
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Alan Conno » Thu, 11 May 2006 15:50:33

On comp.mail.misc, in < XXXX@XXXXX.COM >, "Troy


Hey Troy!



Got it. Thanks.


Correct.


Sure.


Couldn't tell you, Troy, but someone will.

My hunch would be to delete them. If they are sent to throw off
your filter, why give them a chance?


No prob.

[Note: I don't read the articles of "Sam" or his numerous
sockpuppets or his 'friends', nor any responses to them, and
haven't for years. He follows me all over the Usenet, and I
still don't read his articles. This _really_ pisses him off.
<snicker>.]

Alan

--
http://www.yqcomputer.com/ ~alanconnor/contact.html
Other URLs of possible interest in my headers.
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Alan Conno » Thu, 11 May 2006 16:13:09

Postscript:


<snip>

I should note that a lot of C/R users _do_ use SA as the
conventional spam filter stage in their Challenge-Response
system:

http://www.yqcomputer.com/ ~alanconnor/elrav1/cr.html

The 'graymail' gets the challenge.

It's very easy, as you well know, to use procmail to pipe
non-passlisted mail through SA and return it to procmail, scored
by SA, for the C/R stage, with everything but the 'graymail'
dumped by a simple recipe.

I prefer a straightforward procmail ruleset for that
stage.

<snip>

[Note: I don't read the articles of "Sam" or his numerous
sockpuppets or his 'friends', nor any responses to them, and
haven't for years. He follows me all over the Usenet, and I
still don't read his articles. This _really_ pisses him off.
<snicker>.]

Alan

--
http://www.yqcomputer.com/ ~alanconnor/contact.html
Other URLs of possible interest in my headers.
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Sam » Thu, 11 May 2006 19:52:53

This is a MIME GnuPG-signed message. If you see this text, it means that
your E-mail or Usenet software does not support MIME signed messages.
The Internet standard for MIME PGP messages, RFC 2015, was published in 1996.
To open this message correctly you will need to install E-mail or Usenet
software that supports modern Internet standards.

Usenet Beavis writes:



Hey Beavis!


You don't use a C/R system, Beavis.


But that won't stop you from flapping your gums, right Beavis?


The golden rule of comp.mail.misc is: for every Beavis "hunch" or technical
advice, the correct answer always lies 180 degrees to the opposite.





-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)

iD8DBQBEYcYEx9p3GYHlUOIRAt0KAKCBN7QAAzsdlHfEet9O/ZDo7+PLdACfSERT
5dwDLdOW/gOph+w6O+R46FI=
=2bsq
-----END PGP SIGNATURE-----
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Sam » Thu, 11 May 2006 19:55:56

This is a MIME GnuPG-signed message. If you see this text, it means that
your E-mail or Usenet software does not support MIME signed messages.
The Internet standard for MIME PGP messages, RFC 2015, was published in 1996.
To open this message correctly you will need to install E-mail or Usenet
software that supports modern Internet standards.

Troy Piggins writes:


It depends on the individual implementation.

Although something may seem random to you, there are patterns even to random
number generators.

A smart bayesian filter might pick up on the fact these spams -- for example
-- consist of two MIME sections, the text/html section containing a single
<HREF> with an <IMG> inside it that references a CID: of the second,
image/gif MIME section; as I suspect you'll find many of these spams
contain.

You can't really answer this without knowing the particulars of the bayesian
implementation.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)

iD8DBQBEYca8x9p3GYHlUOIRAmIxAJ9/T+8LR2Sps+9L8fPJA9gM2uLAYwCfTkwl
O9KP/p7tTrcFVwdc8VRZqWc=
=FwOt
-----END PGP SIGNATURE-----
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Sam » Thu, 11 May 2006 19:57:32

This is a MIME GnuPG-signed message. If you see this text, it means that
your E-mail or Usenet software does not support MIME signed messages.
The Internet standard for MIME PGP messages, RFC 2015, was published in 1996.
To open this message correctly you will need to install E-mail or Usenet
software that supports modern Internet standards.

Usenet Beavis writes:



How about a deodorant, Beavis? Do you use that?


You don't say?




-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)

iD8DBQBEYcccx9p3GYHlUOIRAq4fAKCAFGSj1mRJtUmV3sT3syEDEDcVNgCfVwHV
l7wkX/Dwsnj3TXcnzAhCh0E=
=40kM
-----END PGP SIGNATURE-----
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 20:38:07


G'day Alan.


[snip]

Based on some advice from others here, and also some reading I'm doing
(or rather started), SA also learns from headers so it is probably
worthwhile using them for training as well.

Peace.

--
Troy Piggins RLU#415538 < http://www.yqcomputer.com/ >
My dogma just ran over my karma
 
 
 

Bayesian training [Was: Spam From Idiots]

Post by Troy Piggi » Thu, 11 May 2006 20:41:59


Fair enough. Thanks for the pointers. I'll do some more reading
hopefully this weekend.

Peace.

--
Troy Piggins RLU#415538 < http://www.yqcomputer.com/ >
My dogma just ran over my karma