[neil]

SpamAssassin. I use it at the procmail level for most of our domains and many of our customer's domains.

 

[myself, aka me]

I set it to 'exclusive' in hotmail. Too bad if something important ends up in my junk folder, I ain't readin' it

 

[averydonovan]

I'm trying out that POPFile program. It looks like a good concent and I'll have to see how well it works. Right now I use a lot of filters in PocoMail to try to put the spam in the right places but it is a rather blunt instrument to use. PocoMail does have a scripting ability that I haven't used yet and I might try it out eventually. I think I recieve anywhere between 30 and 60 spams a day. Half of it is filtered by Yahoo, but the rest I have to sort through.

 

[Ob3ron]

Quote:

. I think I recieve anywhere between 30 and 60 spams a day

Given that amount it shouldn't take very long for you get POPFile trained

. I only get roughly 5 spam per day, so I won't see any reduction in that for a while longer than yourself!

 

[andrzejpw]

Popfile made a mistake this morning.

Emails classified: 143
Classification errors: 12

--------------------------------------------------------------------------------

Accuracy: 91.6%

 

[averydonovan]

Quote:

Originally posted by andrzejpw Popfile made a mistake this morning. Emails classified: 143 Classification errors: 12 -------------------------------------------------------------------------------- Accuracy: 91.6%

That's still not bad. My manual filter system combined with Yahoo has only been hitting 70% lately, it used to be more accurate but spammers kept finding ways to get around my filters and Yahoo's filters. I've turned off Yahoo's SPAM Guard for now just to get all the messages through so I can train POPFile. And at least POPFile doesn't delete messages it just classifies them so you can correct any errors it makes without losing messages.

Last night I fed POPFile 1400 messages from my Earthlink mail account, only one of which was not spam, and it identified all but 8 of the spams as such.

 

[bubbaj]

spamgourmet.com

A free service that allows you to create temporary e-mail addresses. Give them out to whoever and when mail comes in it gets forwarded to your real account. You can specify a count of messages you'll allow through. You can turn off an address completely, or you can allow an address all the time.

You can even set it up so that if you reply to a message sent to you, the recepient will only see your temmporary e-mail address.

 

[andrzejpw]

No, my popfile isn't bad. If you discount the first day or two of usage(where it will make mistakes), I'm at around 98 percent!

 

[redshifter]

i switched from hotmail to softhome.net pop account. with hotmail i got 80% spam, with softhome i have yet to recieve one single spam, and this is without filters or any other filtering programs.

i also test spam filtering for msn client software. i had to write spam mails to test the different junk mail level ratings, but nothing i wrote got caught. then i pasted some actual spam i recieved into the test mails and the msn filter caught each one. i guess i just have no talent as a spammer!

bloody vikings!

 

[Mr.PD]

I just picked up some BBQ flavour Spam today at the store.

 

[jim]

Thanks for the link to popfile. I'm interested in this kind of stuff (how it works, that is).

Quote:

Last night I fed POPFile 1400 messages from my Earthlink mail account, only one of which was not spam, and it identified all but 8 of the spams as such.

I'm not sure how popfile is implemented, but this is generally not a good way for training machine learning algorithms. One of the basic ideas in machine learning is that the probabalistic distribution of your training data should be as close as possible to the distribution of your testing data, otherwise you may not learn very well. (you will have skewed results) You should train it on the complete set of emails that you normally get, good ones and bad.

Popfile most likely has preemptive knowledge of specific features to look at for identifying spam, which would account for it's high success rate right off the bat, but if you were learning from scratch this would be a very bad thing to do. The program would misclassify "good" emails with high error rates.

</academic hat off>

 

[bootman]

I use www.spamcop.net to report spam.