Re: [Evolution] Spam Blocking :: Bayesian Filtering



On Fri, 2002-08-16 at 13:59, Chris Ness wrote:

I've seen Spamassassin in action.  This email account uses it.  It does
an ok job but I get a lot of false positives (mostly in my dumbass
friends HTML emails, and yes I do call them dumbasses to their face for
sending HTML email). I know I can experiment with the "sensitivity" of
the score, but I don't want to have to do that and increase my spam
intake.

I've got my share of dumbasses, too. ;)

On this account, SA gets about a 1% false positive rate with default
scoring.

Yes, you will get spam even with this method but a "delete as spam"
keystroke would tell Evo to update it's DB and recalculate probabilities
to use on the next sync.

You could do the same thing by configuring a single filter to "add to
spam DB" and another to add to the non-spam DB. Give us a way to run one
filter on a message list selection. Then configure the filters to run
some external program that populates those databases. (That part's in
the development release.)

The actual filtering part can still be in the delivery path, just like
SpamAssassin. It would consult per-user DB's, much as SA can be
customized on a per-user basis.

In fact you need a good supply of spam to create a initial DB of
probabilities and non-spam too.  There is a start-up time, but well
worth it overall.

Easy for me. I just dump it all in a big spam folder, so I'd just select
everything in that and feed it to the is-spam filter, and select
everything else and feed it to the not-spam filter.

Why not uninstall spamassassin?  

Then it *is* only at one point, being the end user.  This leaves clock
ticks for the servers.  Of course we could never do this unless Bayesian
filtering became common place in email clients as we would leave
non-Bayesian filtering clients 'high and dry.'

Why not use both? Configure SA with a high threshold to catch the
egregious stuff and use the personal Bayesian filter to catch what
sneaks through.





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]