473752 - Training Junk Mail Controls doesn't affect other recipient addresses.

Status: RESOLVED INVALID

(MailNews Core :: Filters, defect)

Description

[Michael D. Crawford]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.4; en-US; rv:1.9.0.5) Gecko/2008120121 Firefox/3.0.5
Build Identifier: version 2.0.0.19 (20081209)

In general, I have never found Thunderbird's Junk Mail Controls to work very well, despite several years of diligent training.  I found three bugs tonight, and will cross-reference them all from each report after I file them them.

I have several different email addresses each with its own Thunderbird account.  A while back I asked if training for one recipient would cause junk to be recognized for other recipients, and I was told that it would be, but it's clear to me that that's not the case.

Tonight I tried with three recipients, each of which had about a hundred thousand messages in its inbox - mostly Junk.

Select the Inbox for one recipient account.

In the Subject or Sender search box, enter a keyword that is almost certain to match only to Junk.  "Replica" and "debt" are good candidates.

Go through all the search matches and mark all the Junk.  I actually scrolled through all the hits, just to make sure I didn't inadvertently mark something that wasn't junk.

Once all the Junk has been marked, select the Inbox of a different account, then from the Tools menu select Run Junk Mail Controls on Folder, and then, after noting how many messages are in the Inbox, Delete Mail Marked as Junk in Folder.

Very few - if any - Junk mails will be deleted.  The most I ever see is a hundred; much more commonly less than ten, sometimes none at all.  But if I now enter that same keyword in the search box, I can see one to three thousand matching messages, all of which are very obviously junk.

I did all this just now for both "Replica" and "debt".  I did it successively for three different accounts; even after doing this for the second account, running the Junk Mail Controls on the third account marked only a few messages.



Reproducible: Always

Steps to Reproduce:
1.  See the Details section.
2.
3.
Actual Results:  
Very few obviously-Junk messages are marked as such.

Expected Results:  
Nearly all the Junk messages containing the given keyword in their Subject will be marked as Junk.

First-generation MacBook Pro with a 1.83 GHz Core Duo (not Core 2 Duo) and Mac OS X 10.4.11 (Tiger).

Comment 1

[Michael D. Crawford]

Attachment: GZipped Macintosh System Profiler configuration report

After decompressing this file, you can open this configuration report in the System Profiler, which is in /Application/Utilities.

I GZipped it because the original was 800k, much larger than Bugzilla would allow me to post.

Comment 2

[Wayne Mery (:wsmwk)]

xref bug 473756, bug 473758

Comment 3

[Michael D. Crawford]

thanks Wayne.

While I reported all these bugs on version 2.0.0.19 on Mac OS X, I've been seeing all three of these for years, and on Windows 2000 as well.

Comment 4

[Kent James (:rkent)]

(In reply to comment #0)
> 
> Go through all the search matches and mark all the Junk.
> 
> Once all the Junk has been marked, select the Inbox of a different account,
> then from the Tools menu select Run Junk Mail Controls on Folder

As I commented in a different bug, you are describing a process where you only mark junk and not good. This is guaranteed to lead to poor results.

I can also assure you that there is only a single database of bayesian tokens which is shared over all accounts. Yet each account will have very distinct markers beyond simply the email address. If most of your good training was done on one account, then it will be very difficult to convince the spam filter that emails are spam on that account when training from a different account.

The root cause of all of this is problems in the UI that do not encourage users to properly train the bayesian filter.

Comment 5

[Wayne Mery (:wsmwk)]

invalid based on comment 4. There are however deficiencies as noted above, and covered in other bugs