Spam evaluation - detailed information

Forum for MailWasher Pro 7 and/or older 2011/2012 versions.
roy_ved
Mystified Moa
Posts: 7
Joined: Wed Mar 10, 2021 1:56 am

Spam evaluation - detailed information

Tue Oct 12, 2021 6:25 pm

Sometimes I don't understand why an email have been given a certain score.
The sender is not on the blacklist, no words in the email appears in any filter etc. Still the email might be evaluated as spam.
This requires me to constantly keep an eye on my recycle bin to pick up "good emails" that have been autodeleted.

Would it be possible to "expand" the evaluation into a list, and let us see how a score is calculated? Then it would be easier to find what to tune to get a more precise evaluation.

Origin: a points
Blacklist: b points
Custom filter 1: c points (bad phrase 1)
Custom filter 2: d points (bad phrase 2)
etc
etc
Overall score: -210 points
User avatar
rusticdog
Firetrust Monkey
Posts: 15864
Joined: Mon Jun 13, 2005 6:27 pm

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 6:38 pm

If you right click a message >> select Show Evaluation Details >> you'll get the panel on the right that does this breakdown between the various tools.

However it only gives the number for the Learning, which I suspect is what is causing the false detection.

Something that might be a quick fix, is go into Settings >> Spam Tools >> Learning Settings >> set the Good Token Weight to 3 >> Save.
It will currently be 2.0, which means a good word is worth twice as much as a bad. Setting that to 3.0 gives more emphasis on the good words that are present in an email.

Another possible quick fix, is under Settings >> General >> Checking Mail >> set the Spam Throttle to 500 lines per email >> Save.
This gives MailWasher more email data to make a decision on, making it more accurate.



To dig down and view the learning breakdown, from the email preview screen, top left click Show Email Info >> then select the Spam Tools tab >> it's raw data so not pretty, but the list of words used is at the bottom.

The closer the number is to 1.0 the spammier it is.

For example from my learning, these words are considered good
WORD: helpdesk prob=0.011000 occurrences=1
WORD: hideaway prob=0.011000 occurrences=2

These words are mid-tier
WORD: fantastic prob=0.414984 occurrences=1
WORD: discount prob=0.572218 occurrences=1

These words are very spammy
WORD: controvert prob=0.999800 occurrences=1
WORD: fecundity prob=0.999800 occurrences=1


This can probably offer a sense maybe of what's going wrong, perhaps very few good words and lots of bad.
Feel free to copy/paste that info into a text file and email to me at forum@firetrust.com to have a look at.

Cheers
roy_ved
Mystified Moa
Posts: 7
Joined: Wed Mar 10, 2021 1:56 am

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 9:52 pm

Thank you.
My Good Token Weight is already set to 3
I have adjusted the Spam Throttle to 500 lines per email (hope that helps)
The "Show Email Info >> then select the Spam Tools tab" doesn't show anything except what filter that was matched (if matched against a filter). No other data.

Thank you for your suggestions. I will test and see if the filtering improves.
However I still would like to see what phrase in a filter was matched against what phrase/word/source code in an email. A detailed spam score report.

The Topic Reply Notification from this forum went straight to the recycle bin :o

(I have MW Pro 7.12.54 w/lifetime license)
User avatar
rusticdog
Firetrust Monkey
Posts: 15864
Joined: Mon Jun 13, 2005 6:27 pm

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 10:36 pm

It sounds like you have auto-delete set based off the Learning, you can check that under Settings >> Spam Tools >> Spam Ratings >> down the bottom auto-delete if rating reaches xx. I would raise xx, bring it up to -140 as that should only remove the certain spam emails.

You don't see text like this in the Show Email Info >> Spam Tools tab ?

Code: Select all

DECODED_BASE64_SUBJECT: 5
SUBJECT_WORDS: 14
FROM_WORDS: 4
ENCODING: 8bit
CONTENT_TYPE: text/plain; charset=UTF-8
NUM_PARTS: 1
ENCODING: 8bit
TYPE: text/plain; charset=utf-8
BYTES: 885
NUM_RAW_WORDS: 45
URL: https://forum.firetrust.com
MULT: 1.943077E-027
COMB: 6.722343E-002
RAWSPAMYNESSE: -1.000000E+000
RAWSPAMYNESS: -1.00000
SPAMYNESSE: -1.000000E+000
SPAMYNESS: -1.00000
GOODCOUNT: 22
BADCOUNT: 0
GOODWORDCOUNT: 7993
GGAIN: 1.000000
BADWORDCOUNT: 13146
BGAIN: 1.000000
INTRESTINGWORDCOUNT: 20
WORDCOUNT: 33
TOTAL_WORDCOUNT_FACTOR: 1.0000
INTERESTING_WORDCOUNT_FACTOR: 1.0000
WORD: asubject_reply      	prob=0.011000       	occurrences=1
WORD: asubject_topic      	prob=0.011000       	occurrences=1
WORD: byclicking          	prob=0.011000       	occurrences=1
WORD: from_<forum@firetrust.com>	prob=0.011000       	occurrences=1
WORD: from_forum          	prob=0.011000       	occurrences=1
WORD: subject_phxpjjphbhb 	prob=0.011000       	occurrences=1
WORD: subject_pjphbpr     	prob=0.011000       	occurrences=1
WORD: the"unsubscribe     	prob=0.011000       	occurrences=1
WORD: topic               	prob=0.011000       	occurrences=1
WORD: from_firetrust      	prob=0.025912       	occurrences=1
WORD: bottom              	prob=0.063282       	occurrences=1
WORD: following           	prob=0.070634       	occurrences=3
WORD: support             	prob=0.079918       	occurrences=1
WORD: asubject_notification	prob=0.125158       	occurrences=1
WORD: above               	prob=0.233134       	occurrences=1
WORD: asubject_information	prob=0.233134       	occurrences=1
WORD: detailed            	prob=0.288431       	occurrences=1
WORD: longer              	prob=0.301031       	occurrences=1
WORD: reply               	prob=0.359486       	occurrences=1
WORD: either              	prob=0.419287       	occurrences=1
roy_ved
Mystified Moa
Posts: 7
Joined: Wed Mar 10, 2021 1:56 am

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 10:45 pm

Nope!
Or, just now an email popped in that shows this, but all other emails marked as spam are blank.
Attachments
screenshot.JPG
screenshot.JPG (58.97 KiB) Viewed 3156 times
User avatar
rusticdog
Firetrust Monkey
Posts: 15864
Joined: Mon Jun 13, 2005 6:27 pm

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 10:49 pm

Right sorry, I'm an idiot. We don't store that information for deleted messages.

That brings us back to the auto-delete setting for the Learning, what was that set to ?
roy_ved
Mystified Moa
Posts: 7
Joined: Wed Mar 10, 2021 1:56 am

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 11:09 pm

Thank you for taking the time!
Please find attached the screenshots

(PS! I am leaving for a meeting now)
Attachments
screenshot3.PNG
screenshot3.PNG (38.57 KiB) Viewed 3155 times
screenshot2.PNG
screenshot2.PNG (58.67 KiB) Viewed 3155 times
User avatar
rusticdog
Firetrust Monkey
Posts: 15864
Joined: Mon Jun 13, 2005 6:27 pm

Re: Spam evaluation - detailed information

Tue Oct 12, 2021 11:30 pm

OK great, so that Good Token Weight looks like it's set to 0.00. Can you set that to at least 2.0 >> Save >> restart MailWasher >> check that the change saved.

Then you can also see the auto-delete rating is currently set to -83, so pull that slider back to -140 to start with. You'll then see less emails in the Bin and more in the main screen, where you can correct the training. After a couple of days assuming MailWasher is getting it right identifying the spam and good, try moving it down to -120. Just see how that goes for a while.
roy_ved
Mystified Moa
Posts: 7
Joined: Wed Mar 10, 2021 1:56 am

Re: Spam evaluation - detailed information

Wed Oct 13, 2021 1:54 am

Thank you.
I will adjust my settings, and try for a while.

Return to “MailWasher Pro 7”