High-ascii characters found in subject header SARE_HEAD_8BIT_SPAM

[vuser1]

I noticed that some normal emails are marked as SPAM. It happens not so often, but happens. Rule level is normal, not aggressive. Looks like there is an option to improve. See:

Subject: SPAM: Регистрация в системе A1 Агрегатор
raw subject: SPAM: =?utf-8?B?0KDQtdCz0LjRgdGC0YDQsNGG0LjRjyDQsiDRgdC40YHRgtC10LzQtSBBMSDQkNCz0YDQtdCz0LDRgtC+0YA=?=

In headers:
SARE_HEAD_8BIT_SPAM 0.888 High-ascii characters found in subject header
SUBJ_ALL_CAPS 1.806 Subject is all capitals

1) Notice that subject is not all capitals. Looks like bug.
2) Why it complains about non-ascii characters, what's the problem? Should this check be OFF in Proxmox - it is international software, I hope This would decrease the chance of "false positive" for emails with UTF-8 subjects.

 

[dietmar]

Please try to disable:

Configuration/SpamDetector/Options/UseAdvancedTests

 

[vuser1]

OK, I've did it. I hope it doesn't switch off useful tests like "too many spaces" or "message is HTML only". Anyway, do you think that non-ascii characters should not increase spam level?

 

[dietmar]

Anyway, do you think that non-ascii characters should not increase spam level?

No. But Spam detection rules are heuristics - mass checks results determine if a rule makes sense or not.

 

[vuser1]

Interesting! Do you mean the system learns and will decrease weight of a rule? How does it work, how it understand that email was not spam?

 

[dietmar]

Interesting! Do you mean the system learns and will decrease weight of a rule?

Rule developers run "mass checks" to test rules.

Besides, the system uses various statistical systems (BAYES, AWL) to 'learn'.

 

[vuser1]

Do I understand right that Proxmox Mail Gateway does not learn on-the-fly, according to my emails. Instead, it relies on statistical databases of built-in engines such as SpamAssassin?

 

[dietmar]

Do I understand right that Proxmox Mail Gateway does not learn on-the-fly, ...

I just wrote the oposite - what exactly is your question?

 

[vuser1]

1) Why non-ascii characters in subject increase spam level? This does not make sense outside english-speaking countries.
2) You wrote that proxmox 'learn' and decrease weight of the rule SARE_HEAD_8BIT_SPAM. How it can learn if users do not tell it that email is not spam? Users receive messages with "SPAM:" in subject and they can't send feedback to proxmox that message is not spam.

Perhaps I do not understand some very basic concept of proxmox MG...

 

[dietmar]

1) Why non-ascii characters in subject increase spam level? This does not make sense outside english-speaking countries.

It make sense, because most spam is written in english.

2) You wrote that proxmox 'learn' and decrease weight of the rule SARE_HEAD_8BIT_SPAM.

It uses Bayes filtering and other adaptive/statistical methods to learn.

 

[dietmar]

Perhaps I do not understand some very basic concept of proxmox MG...

Yes. One basic concept is that more than one test needs to trigger.

- Dietmar