Junk does what, exactly?

[Mark Smith]

On 10 Mar 2007, at 15:11, mmalc crawford wrote:

> On Mar 9, 2007, at 11:03 PM, LuKreme wrote:
>
>> Apple's Junk mail filtering uses Baysian filters
>>
> Actually it uses latent semantic analysis (<http://www.apple.com/ 
> lae/macosx/jaguar/mail.html>, <http://www.hawkwings.net/2005/12/24/ 
> how-apple-mails-junk-filter-is-not-like-the-others/>).

I remember reading that article when it came out and thinking that it  
seemed robust and that Mail's junk filtering ought to be at least up  
with the best.

However, in repeated lengthy experiments, SpamSieve (used in Mail  
with Mail's own filtering turned off) beats it hands down. It gets  
very good much faster than Mail (after about 500 messages SpamSieve  
is consistently above 98% for identifying spams (Mail is around  
90-95%) and from this point SpamSieve has consistently fewer false  
positives (far less than 1 in 1,000 compared to Mail's approximately  
1 in 500).

SpamSieve is so good, that I have completely abandoned maintaining  
SpamAssassin on the server side.

mark.