Junk does what, exactly?

[LuKreme]

On 9-Mar-2007, at 19:41, Macs R We wrote:
> I can't find any good reference as to exactly what effect the  
> "Junk" button has in Apple Mail.
>
> Sure, it moves a message to the junk mailbox, that much is clear --  
> but what gets "learned" in terms of future mail?

Apple's Junk mail filtering uses Baysian filters, so a message is  
submitted to bayes and this improves the checking of future messages.

>   Does it become more wary about messages with similar contents?

Simplistically, yes.

>   Does it give the sender a demerit?  Neither, both, something  
> else?  If so, how many demerits put all of that sender's messages  
> in the Junk bin?

It doesn't work that ways.  Bayes looks at everything and comes up  
with a probability a message is junk, based on the other messages  
you've marked and, in this case, Apple's supplied starting database.

> And how can you see what Mail's plans are for future messages?   
> "Learning" based rules don't appear in the Rules panel.  For  
> example, don't expect to find a rule saying, "It the contents  
> contain Viagra, and the contents contain shipping, it's junk mail,"  
> unless you put it there yourself.

That's not how bayes works.

<http://en.wikipedia.org/wiki/Bayesian_spam_filtering> is a good  
starting point, also I think the SpamAssassin wiki has some writing  
on Bayes filtering.

> In fact, a number of online references seem to imply that once you  
> take Mail out of "learning" mode, the Junk button no longer causes  
> anything to be learned.  I never thought that was the case, but  
> I've just realized that I can't point to any authoritative source  
> that says otherwise.

There's little specific information on the exact mechanisms.  I've  
not looked into it much because I use SpamAssassin on my server to  
deal with Spam.  The only Junk mail.app deals with is the account  
that is wide-open (by design).

>
> Here's my motivation: today, a new client complained to me that  
> once upon a time she "junked" a single chain-letter message from a  
> friend, and now she regrets it, because all that friend's messages  
> are going into the Junk mailbox.  (That struck me as odd, because  
> my experience has been that you have to hammer a sender multiple  
> times before he is effectively blacklisted, if ever -- usually I  
> get tired and just write a rule.)  She couldn't exhibit any proof,  
> since her Junk mailbox was empty.  She wanted me to do something  
> specific to undo what she had done.  As I remarked previously, it's  
> not like there's some file I can go to and remove the line that  
> says "Junk all mail from Hermione."

Marking other messages as not spam would balance out

> So I told her next time a message of Hermione's got junked, she  
> should go into the Junk mailbox and un-Junk it -- on the theory  
> that whatever weak censure a single Junk button may have caused, a  
> single un-Junk button should correct.  Then she said, "I can't tell  
> when there are things in there, and they don't stay in there  
> long."  Now first off, I know for a fact that the unread message  
> count shows up right next to the Junk mailbox.  Secondly, I can  
> find no preference setting that would automatically empty the Junk  
> mailbox, so I don't think that even happens.

Sure, Preferences -> Accounts -> Mailbox Behaviors -> Junk -> Delete  
junk messages when: (Never|One Day Old| One Week Old| One Month Old| 
Quitting Mail)

> Anyway, it struck me after this encounter that Mail's claims of  
> "learning" are vague and unsatisfying, in terms of exactly WHAT it  
> thinks it is learning, how often it has to "learn" something before  
> it "sticks," and what you can do to modify what it has learned.  I  
> could find nothing in a Google search other than the usual vague  
> promises that "Mail learns stuff when you hit the button."

Mail Help -> Discover Mail -> Dealing with junk email-> Too much of  
my legitimate email is getting marked as junk

> Train Mail that the messages are legitimate. Whenever you receive  
> messages that have been incorrectly marked as junk, select them and  
> click Not Junk in the toolbar or choose Message > Mark > As Not  
> Junk Mail. Over time, Mail will learn the messages are not junk.
> Add the senders to your Address Book. To do so, click the arrow in  
> a Smart Address and choose "Add to Address Book" from the pop-up  
> menu. Or select the message and choose Message > "Add Sender to  
> Address Book".
> Review your Mail Preferences for junk mail, as well as any rules  
> you might have created to handle junk mail, and make changes as  
> necessary.

For the record, unless she messed up her settings, no message from  
anyone in your address book OR your previous recipients should be  
marked as junk.

Preferences -> Junk Mail
The following types of messages are exempt from junk mail filtering:
   [x] Sender of the message is in my Address Book
   [x] Sender of the message is in my Previous Recipients
   [ ] Message is addressed to my full name

(those are my settings)


-- 
Rid yourself of doubt --
                                 or should you?
-George Carlin