[mel-base-devel] Spam filter

Jochen Schmidt js at crispylogics.com
Sun May 9 09:34:53 UTC 2010


Am 09.05.2010 um 02:08 schrieb Ron Garret:

> Hello,
> 
> Just found out about Mel-base.  Looks very cool.  I have two questions:

Thank you - I'm actually still quite fond of it when looking at the code years after I wrote it. :-)

> 
> 1.  Why is it called Mel-base?

It originally was called WeirdMail but then Robert Strandh wanted to write a CLIM Mailreader he called "Mel" which is a slang word for Mail in french. I offered my (up to then closed source) mail library and made it public as mel-base (the base library of mel). Perhaps its time to really rename it to something more recognizable. (Have to think hard on it...I'm open for options)

> 
> 2.  The web page says:
> 
> "Mel-base got used to implement a .... Spam-Filter .... The mentioned applications, while not Open Source, are available on request."

The spam filter used a reimplemented method of Paul Grahams naive bayesian method. I'm actually not sure if I have the sources still here. If I would nowadays want to write a spam filter using mel-base I would do it different. The bayesian method alone is not enough. On my mail server (postfix) I use DNS based black lists which filter out quite some stuff. On some highly spammed addresses I use grey listing which means the first mail from unrecognized senders gets denied and the second mail accepted only when tried 10-15 mins later. This removes spam greatly and reduces stress on the mail server but it may happen that incorrectly configured sending hosts are not able to get through (one can see that in the logs then and add them to a white list). I use spamassassin as a generic spam filter - it has a big ruleset which heuristically rates spam and adds a spam header field which my mail reader (Apple Mail) recognizes. A mel-base using spam solution should be able to do DNS queries to black lists, use a rule based rating system (Mycin? Certainty Factors?) and text analysis using different kinds of things e.g classification using "support vector machines".

If you need some help on how to do what you want in mel-base - feel free to ask me. Mel-base is highly extensible and it is quite simple to add support for new protocols.

> 
> To whom should such a request be addressed?

me

I'll see if I can find the sources on one of my backup discs.

ciao,
Jochen

-- 
Jochen Schmidt 
CRISPYLOGICS
Uhlandstr. 9, 90408 Nuremberg

Fon +49 (0)911 517 999 82
Fax +49 (0)911 517 999 83

mailto:(format nil "~(~36r@~36r.~36r~)" 870180 1680085828711918828 16438)
http://www.crispylogics.com





More information about the mel-base-devel mailing list