Richard Hart said: >I'm talking about the >relatively recent random-text/random-name scourge, also known as >Bayesian poisoning. Nonsense sentences and randomly generated names are >driving me crazy.
Recent? Been getting those for a year at least. Is it a new strain you're talking about? >Please tell me how you filtered those. I'm inundated. I simply mark them as spam. That's it. Some get trough, like 2 similar messages a week out of hundreds. I did spend time years ago training SpamSieve with hundreds of messages, but lately I just let SpamSieve train itself. I do use 2 spam filters after the evaluation one. One is set to mark messages with a spam value of 20 with a spam label and put them in the spam folder. "Don't apply subsequent filters" are unchecked. The other toss messagees with a spam value of 50 in the mail trash and check "Don't apply subsequent filters". This method has ridden me of having to wade trough the PowerMail trash for false positives. Almost none get to there. I have an accuracy rate of 99.7% since 2004 (more than 150 000 messages filtered with an average of 51 spam messages a day). 482 false negatives out 50 000+ spam ones which is less than 1%. There are 5 000+ "Good" messages and about 6 200 Spam ones in the corpus. Mikael Tech facts: PM 5.5.2 Swedish | OS X 10.4.5 | Powerbook G4/550Mhz | 1GB RAM | 80GB HD