View previous topic :: View next topic |
Author |
Message |
ehm
Joined: 25 Jan 2007 Posts: 8
|
Posted: Wed Jun 20, 2007 3:06 pm Post subject: Learning process is not 'learning' :( |
|
|
I'm quite concern on the software.
I used the full 30 days before try it and the antispamsniper seams to learn very slow, then I bought it, and now, after almost 2 months using, I still receiving "Famous Replica Watches ROLEX Cartier Bvlgari" every single day on my 'normal' e-mail box. Every day it is from a different e-mail, but it's there,
Seams that AS is learning nothing at all.
That is only on example, but I'm receiving same content spam e-mails every day and every day I mark it as JUNK using the button but next day I will receive another one.
What could a be doing wrong?
Note that junk sent by a previous blocked email are being held the way it's supose to be. The problem is if the sender change the e-mail account, the AS is not blocking it.
I'm sure I didn't understand something on the software and there is a way to make it learn properly.
Please help ! |
|
Back to top |
|
|
vetaltm Author
Joined: 05 Feb 2006 Posts: 751
|
Posted: Wed Jun 20, 2007 7:32 pm Post subject: |
|
|
Most probably there was a mistake during training the plug-in, when spam was marked as non-spam. This explains slow learning and the filtering mistakes. You can check the filtering log and look at the spam ratio for "Replica Watches" messages. If the ratio is around 40-60%, then the classification database is poisoned. The algorithm in latest versions is sensitive to training mistakes, because it tries to make his best to keep high classification quality for all messages it ever seen.
You can clear the classification database (there is a special button on Options page) and train the plug-in again. Make sure that you never send spam messages, even for testing, because by default the plug-in learns all outgoing messages as non-spam. The training mistakes are often accur on marking as spam or non-spam a lot of messages simultaneously, because it is easy to miss a message of other class in this case. The best way to avoid the training mistakes - training the plug-in only when it makes a filtering mistake. |
|
Back to top |
|
|
ehm
Joined: 25 Jan 2007 Posts: 8
|
Posted: Fri Jun 22, 2007 2:38 pm Post subject: |
|
|
OK. I will delete the file and start the training again. I don't send spam nor similar messages, so I would guess that somehow I classify a SPAM as a 'not spam' in the past and this is why I'm having it.
I will post here in few days/week with the results.
PS: the watch replicas topic was only an example. I'm receiving same viagra, mortgage, and other same kind of e-mails as a 'non spam' every day. |
|
Back to top |
|
|
vetaltm Author
Joined: 05 Feb 2006 Posts: 751
|
Posted: Fri Jun 22, 2007 3:43 pm Post subject: |
|
|
ehm wrote: | PS: the watch replicas topic was only an example. I'm receiving same viagra, mortgage, and other same kind of e-mails as a 'non spam' every day. |
The situation you described, when a lot of obvious spam is recognized as good, never occurs on marking 1-2 spam messages as non-spam. Probably in your case the spam messages are regularly learned as good by the plug-in. It is possible if you receive many spam messages with spoofed sender, i.e. with good address in From field (for example with your own address). Try to turn off auto-learning for the messages from friendly addresses ("Learn the messages from friends as non-spam" on Filtering page). As an alternative, turn on the flag "Ignore email addresses without display names" on Filtering page. The plug-in checks the display names for email addresses in this case. When this flag is enabled the spam messages with good email address in From field, but with a wrong or empty display name in address, are not recognized as non-spam and not learned as good by the plug-in. The spoofed addresses rarely contain good display names, so this additional checking should help. |
|
Back to top |
|
|
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2005 phpBB Group
|