Erik Enge writes:
Anthony Ventimiglia anthony@ventimiglia.org writes:
How does mailman work, If it recieves mail the same way user mail is delivered, then it should be trivial to pipe it through a spam filter and procmail along the way. Then procmail could isolate spam and send normal messages along their merry way.
That is true but one of the things I do like with the solution I came up with for the place I work is that the spamchecking is done at SMTP-time and if it is judged to be spam it gets rejected there; no collateral spam via bounces etc.
The problem I have with that is the posibility of false positives. My filter is at a 98.17% success rate, but that still leaves the posibility that mail I want will be seen as spam. I've been thinkin about writing a script to send a reply to all spammers that their mail was rejected as spam, and possibly telling them a way to resend so it doesn't get rejected. While this seems a little silly, it's like giving the keys to a burglar, I don't think any spammers will actually read a response, but someone who is a false positive will read the response. Using this method, all spam could be replied to and sent to /dev/null.
Now back to the subject since we talked about it, I'd like to start a project "cl-bayes" that will be a Bayesian Patter filtering library.
If we can get someone to write a mail parsing library, putting the two together to make a CL spam filter will be trivial.
Thanks,
Anthony