On Friday 25 June 2004 22:14, Graham wrote:
Maybe I'm not training the filter properly but I can't achieve a high enough accuracy to avoid the need to inspect the ones that creep through, which amounts to 20-30 a day.
I have found that you need a really generous quantity of mail in both the spam and ham folders and then training sa-learn by pointing it to both does the trick.
For some funny reason I have heard that it helps to have more ham than spam when doing this. Also it is good advice to use real spam you have received and not training samples, or someone else's trained filters.
Another thing I do in the war against spam is use different names within my mx domain to subscribe to each service. So when for example I registered to ebay I used ebay@mydomain.com the ALUG is aluglist@mydomain.com and so on.
This has two benefits, firstly when I do get spam I can tell where they harvested my address from, and secondly when the spam level gets too high for a particular address, I can then configure the servers at plusnet to dump that particular address.
Also if I have to put my email address up on a website as a mailto: contact, I tend use some javascript to mask it from harvesters.
I still pick up my older addresses and they are full of spam, my current address only picks up about 10 junk mails a week and sa usually catches all of them.