On Thu, 12 Feb 2004, The awesome and feared Glenn Johnson commented thusly,

> On Fri, Feb 13, 2004 at 08:12:22AM +0600, Grendel wrote:
> 
> > I didnt have to even bother to add you, with your last post you
> > automatically triggered my killfilter. I really am considering
> > releasing the source for this killfilter as it learns quite well,
> > as it would have studied your posts and marked you out as a likely
> > candidate long before your current post.
> 
> I hate to interject here, but how does a killfilter work?
 
Well its like this, in the good old days you had a killfile, ie a file 
that you put all the email addreses that you dont want mail from. Now when 
say you receive a email a program checks the killfile and if the senders 
email is present it takes appropriate action, it can save it to a seperate 
folder like mail/pests or send it to /dev/null.

One method this was used for was to remove spam, ie put the list of known 
spamers email addreses and you will not get mail from those addreses. Also 
if you dont like email from someone in this group you add his/her email 
address to the kill file and you wont be troubled by him/her anymore.

Now of course the trend for spam is to use a bayesian filter like bmf or 
bogofilter. How these filters work is that at the start you train them by 
feeding them a database of emails and tell them that this is what spam 
looks like. Then you feed them your normal email and tell then this is 
what my normal mail is. Now you run the program. 

Now when thee bayesian filter receives a email it searches its database 
for strings in the email and calculates a probability whether this mail is 
spam or not. Then it adds the latest information from the email to its 
database. So as you can see the more spam it receives the better it 
becomes at detecting spam. Of course if it flags a normal mail as spam, 
then you can instruct it that this mail is not spam, its a genuine 
message. So it will learn and not make the mistake again. trust me that 
bmf and bogofilter are better than spamassasin. I tested that with the 
last 1000 emails I recevied 200 were spam, and only 1  got through. 
Freshmeat has more about this in the spam-filter review on the main page.


So the killfilter which i use is a bayesian filter, ie it has been trained
to recognise people in email lists who have a annoying potential, for
example people who use bad words, it also analyses the replies to my posts
and ranks a persons reply as hostile or friendly. It also analyses my
replies to users and decides if my reply is hostile or friendly and thus
grades people in the list. 

All this is done through bayesian studies so the more I correspond the
better it gets at detecting people. So based on things like the above it
gueses in advance whether a user should be added to the killfile and his
email being sent to /dev/null and does it.

Grendel

 

-- 
Hi, I'm a signature virus. plz set me as your signature and help me spread
:)

--
[EMAIL PROTECTED] mailing list

Reply via email to