Status of Spam/Junk mail filter?

[johann_p]

I see some UI for that in the current nightly (299211972), but I wonder if this is
already supposed to work? Is the backend working? I also wonder how this thing
handles training and retraining - how does the user initiate this? Will it be possible
to clear the classifier to retrain it from zero? From what I see no the UI is rather confusing
and probably not usable for arbitrary spam filter backend solutions (I think having the option
to plug in backends would be extremely useful both for researchers and users).
I think there are several things that should be considered now so that the solution
will not be fixed to something inflexible later (like with the labels solution).

[cdn]

I see some UI for that in the current nightly (299211972)

are you from the future ? 990 years ahead of the rest of us, and they still haven't finished spam filtering

spam filtering as of 20021108 is AFAIK still under development backend-wise

[johann_p]

I should definitely give up blind typing with all the different keyboard layouts around :/ - that was meant to read 2002110721.

Found the bug in the meantime: seems to be bug 163188

It also seems there is work for a filter plugin architecture under way - great!

OBbuild: so far I haven't seen any major flaws for that build in browser and mailnews on linux, which is quite remarkable after that many checkins.

[drosoph]

Looks like the front end UI is in place ... the MOVING of Junk Mail has been turned off, as well as the classifying of Junk Mail (Is/Is Not) ... currently all mail shows up as unknown and there is no way to mark it as Junk/Not Junk ... YET ! ...

Cant wait for the Bayesian filtering to get checked in !

[DIGITALgimpus]

I personally have been using SpamAssassin (http://www.spamassassin.com) through ActivePerl on Windows XP (for UNIX of Mac OS X even better) with Pop3Proxy...

Then have the mail labled as spam moved to a spam folder.


Works wonderfuly.

I almost wonder if Bayesian filtering will be a step back for me if I use it.


Though I would love to still give it a try.

[Wildcard]

Hope the backend gets checked in soon. This is going to be a killer feature

[drosoph]

Looks like yet ANOTHER instance of SEA builds not enable all the features ... using a ZIP build and removing the previous installation (via DELETE, not add/remove programs) makes the MAKR MESSAGE AS JUNK/NOT JUNK enabled ...
QUICKLY it is already marking incoming junk mail properly ..

What's more .. you can create a MANUAL filter to move any mail labelled "JUNK" into the trash bin or some special "SPAM" folder .. .. gotta love it !!!!

What logic is in the backend at this point that is working ... if BAYES is not turned on yet ?????

Anyone know ?

[mythdraug]

Looks like yet ANOTHER instance of SEA builds not enable all the features ... using a ZIP build and removing the previous installation (via DELETE, not add/remove programs) makes the MAKR MESSAGE AS JUNK/NOT JUNK enabled ...
QUICKLY it is already marking incoming junk mail properly ..

Well.. I guess that would resolve bug 179150

[ckjnigel]

http://www.mozilla.org/mailnews/spam.html
In the mailnews ng Pratik & others talk about "training" by deliberately putting junk in the inbox. Anybody know how I can get all that spam from Hotmail into a Mozmail-handled inbox?

[slippytoad]

Maybe it's just me but 2002111308 Win2k doesn't do anythign when you select a message to mark as junk. I've made the column visible but it has no column name nor does an icon show up in there. Tools->Junk Mail Controls lets you set it up but it has no effect on anything. Anything I'm missing, or maybe the code just hasn't landed yet?

[whiprush]

I almost wonder if Bayesian filtering will be a step back for me if I use it.

I'm also happily using spamassassin on my linux mail server too, and I've been wondering about this as well. I'm not too familiar with Bayesian filtering techniques, but I do know that spamassassin works nearly 95% of the time for me. Any experts care to chime in? Do you have to train a Bayesian filter?

Of course, if they are indeed working on a plugin architecture, I'm sure a spamassassin plugin would become very popular.

[velociraptor]

I almost wonder if Bayesian filtering will be a step back for me if I use it.

I'm also happily using spamassassin on my linux mail server too, and I've been wondering about this as well. I'm not too familiar with Bayesian filtering techniques, but I do know that spamassassin works nearly 95% of the time for me. Any experts care to chime in? Do you have to train a Bayesian filter?

Of course, if they are indeed working on a plugin architecture, I'm sure a spamassassin plugin would become very popular.

Yes you need to train a Bayesian filter (or more correctly "naive bayesian classifier") and it will work best if you have already a larger collection of junk and non-junk examples.
A well trained classifier can well exceed 95% though. You can never be sure of course, depending on what strategies spammers will use in the futures.

However, the important point is not so much to really catch all spam you get, but NOT to falsely mark non-junk messages as junk (false positives). Naive bayesian classification is quite good with that. So even if it detects only say 95% of the junk correctly it should detect close to 100% of the noin-junk correctly. As far as I can see until now, the implementation in Mozilla is quite good in that (couldnt do a more exact formal test yet, since the controls either dont allow it or I dont know how to do it... )

[slippytoad]

How does one go about using it in Mozilla? I am using 1.3a nightly zip builds into a fresh dir.

[Levin]

I have build 2002111208 on Mac OS 10.2.2.

I was surprised to see the Junk icons and button.

I can now mark spams as Junk with the Junk button. Onced marked, the icon for each messages changes to a Junk status icon.

However, I except to see some AI that will mark the junk messages automatically and let me delete these messages quickly.

I see that the Spam filtering is making good progress.

Good work.

[slippytoad]

I'm using the Orbit theme and the junkmail icons do not appear. Switching back to Modern makes everything much better!