feeding bayes

[ChupaCabra <mi...@topstexas.com>]

I just noticed that some of my users are attaching multiple spams in one 
email to my "spam" user for training.  Aside from having to have 200 
spam/ham mails and the possibiliy of the messages being too large, would 
this have an effect on training?

Thanks.

-- 
Michael H. Collins  Admiral, Penguinista Navy

http://linuxlink.com

/"\    	ASCII Ribbon Campaign
\ /    	No HTML/RTF in email
 x     	No Word docs in email
/ \    	Respect for open standards

In a related story, the IRS has recently ruled that 
the cost of Windows upgrades can NOT be deducted 
as a gambling loss.

Re: feeding bayes

[ChupaCabra <mi...@topstexas.com>]

Thanks.  Glad I havn't started feeding it yet.  So what I need to do is
look at the attachments and then save then to a file then train on that.
correct?  That makes sense.  Maybe that is why my old SA instance got
kinda weird in the end.

Thanks


Matt Kettler wrote:

>
> Well, first, if you aren't stripping the attachments and training 
> those, your whole system is flawed regardless of how many spams they 
> send.
>
> You cannot train on messages forwarded by or sent by one of your 
> users. SA's bayes engine learns message headers, so you MUST strip off 
> their message prior to training or SA will learn that your users are 
> sources of spam. Not a good thing.
>
> sa-learn really needs original messages, with original headers.
>
> Now, if you're stripping attachments and training the attachments, it 
> really shouldn't matter how many they attach, unless your script that 
> strips them is broken...
>

Re: feeding bayes

[Matt Kettler <mk...@evi-inc.com>]

At 11:22 AM 3/14/2005, ChupaCabra wrote:
>I just noticed that some of my users are attaching multiple spams in one 
>email to my "spam" user for training.  Aside from having to have 200 
>spam/ham mails and the possibiliy of the messages being too large, would 
>this have an effect on training?

Well, first, if you aren't stripping the attachments and training those, 
your whole system is flawed regardless of how many spams they send.

You cannot train on messages forwarded by or sent by one of your users. 
SA's bayes engine learns message headers, so you MUST strip off their 
message prior to training or SA will learn that your users are sources of 
spam. Not a good thing.

sa-learn really needs original messages, with original headers.

Now, if you're stripping attachments and training the attachments, it 
really shouldn't matter how many they attach, unless your script that 
strips them is broken...