Settings for learning

[krazykanuck]

Been using 2010 for a week now and and seeing the same style SPAM consistently learning as 149.

examples

ScreenShot001.jpg (2.8 KiB) Viewed 3764 times

ScreenShot002.jpg (3.6 KiB) Viewed 3764 times

ScreenShot005.jpg (3.19 KiB) Viewed 3764 times

I haven't change the learning settings from the default.
I am just wondering if the short sentence is not giving the learning filter enough to work with.
What else is the learning filter looking for?
Are there settings that I can change that might help catch these?
I am assuming that V6 was catching these since I don't remember seeing them until I switched to 2010.
I've been a user since 3.3 so by the time I got to V6 allot of learning had occurred so maybe the answer is I just have to wait.

Any suggestions will be appreciated

Thanks

[Sidewinder]

Been using 2010 for a week now and and seeing the same style SPAM consistently learning as 149.

examples

ScreenShot001.jpg

ScreenShot002.jpg

ScreenShot005.jpg

I haven't change the learning settings from the default.
I am just wondering if the short sentence is not giving the learning filter enough to work with.
What else is the learning filter looking for?
Are there settings that I can change that might help catch these?
I am assuming that V6 was catching these since I don't remember seeing them until I switched to 2010.
I've been a user since 3.3 so by the time I got to V6 allot of learning had occurred so maybe the answer is I just have to wait.

Any suggestions will be appreciated

Thanks

It will take a number of these type of messages to train the Bayesian evaluation. Unfortunately the spammers also know that the short sentences make their output harder to detect. I prefer to set the Bayesian range to low. Settings, Spam Tools, Learning Settings, Bayesian sensitivity to Low. I also set my Good token weight to 1.00 but you may want to wait awhile until you complete more training time to make this change.

I also have my Settings, Spam Tools, Spam Ratings set as follows: Friends +200, Blacklist -200 and Origin of Spam and First Alert at -150.

[stan_qaz]

2010 has been tweaked to be less sure of the spamminess of short messages, the Bayesian score is lowered by a scaling factor as the number of words drop. The good side of this is that non-spam messages are less likely to be tagged by the learning tool.

Really your best bet is to use the combination of tools that MW offers, copmare to the v6 one-at-a-time method the 2010 combination method is a lot more powerful and accurate. USe the blacklist for wildcard domains you don't get good mail from, use the source of spam tool with a decent selection of servers, likely 3 or 4 is the best number, FirstAlert is working decently in 2010 without the false positive problems it had in v6 and last are the filters that you can do some really nice things with.

There is a sticky post with some filter examples and a copy of my "I hate everyone" blacklist here somewhere if you'd like to give them a shot.

http://forum.firetrust.com/viewtopic.ph ... 689#p19689 - my blacklist, you don't need the numbers in 2010.

http://forum.firetrust.com/viewtopic.php?f=50&t=5575 - basic filters

http://forum.firetrust.com/viewtopic.php?f=50&t=5642 - regex (very powerful) filters