From: Hill, B. <hi...@nl...> - 2009-07-15 18:36:14
|
LABANVOYE Franck wrote: > Sent: Wednesday, July 15, 2009 11:58 AM > To: 'ass...@li...' > Subject: [Assp-test] Bayesian Spam > > Hi, > > I have problem with bayesian detection. > I run lastest ASSP RC0.3.15, and i rebuild the bayesian DB > with clean folder of spam and notspam file (cf rebuildrun.txt). > > But, some spam pass the bayesian check (no header added in > the mail), but when i test in the mail analyser GUI, the > result is spam probability of 1. > The database is the same in base case. > And the clean database block less spam than the previous DB. > > Any idea ? > > Thanks ..... > ./storage/errors/spam > File Count: 0 > Processing... storage/errors/spam with 0 files > Imported Files: 0 > Finished in 1 second(s) > > ./storage/errors/notspam > File Count: 0 > Processing... storage/errors/notspam with 0 files > Imported Files: 0 > Finished in 1 second(s) There's only so much just the plain old spam and notspam folders can do. You need to start fine tuning it by forwarding emails to the email interface. For example, you get an email in your inbox that should be spam, but it's in your inbox. Now forward that email to asspspam@yourspamdomain(whatever it is).com. That email will now be put in the errors/spam folder now. ASSP will take that into consideration by adding extra weight to the word pairs contained in the files there. After the next database rebuild, ASSP will know not to allow it through again (a simplified explanation). The same applies to emails that are recognized as spam, but are not. They will go into the errors/notspam folder. The same thing happens to these files except the reverse (assp knows the word pairs are good, thus allowing them through next time). Hope that explains it a little bit! Kind Regards, Brett |