On 04/11/2018 11:14 AM, Matus UHLAR - fantomas wrote:
On 04/10/2018 03:49 PM, Motty Cruz wrote:
I apologize here is the email headers and body

https://pastebin.com/bgXrfKaQ

On 10.04.18 16:28, David Jones wrote:
Content analysis details:   (16.0 points, 5.0 required)

pts rule name              description
---- ---------------------- --------------------------------------------------
4.2 RCVD_IN_IVMBL_LASTEXTERNAL RBL: No description available.
                           [178.62.193.238 listed in sip.invaluement.com]
5.2 BAYES_99               BODY: Bayes spam probability is 99 to 100%
                           [score: 0.9996]
3.2 BAYES_999              BODY: Bayes spam probability is 99.9 to 100%
                           [score: 0.9996]
1.2 ENA_RELAY_IN           Relayed through India
0.0 MISSING_MIME_HB_SEP    BODY: Missing blank line between MIME header and
                           body
2.2 ENA_RELAY_NOT_US       Relayed from outside the US and not on whitelists
0.0 ENA_BAD_SPAM           Spam hitting really bad rules.

Since most ofthose rules are 3rd party and other have tuned scores, it's
quite expected that the mail scored 3.5.

(BAYES_999 was apparently not hit, it would score 3.7 then).

we sometimes must accept that a FP appears.
otherwise, there would be no spam and no discussion here :-)


If you read between the lines above there are two points:

1. the IVM RBL is awesome and well worth it's low price

2. if you follow my practice by whitelist_auth as many sender's based on their envelope-from address, then you can aggressively train your Bayes DB based on the content and not the sender to get maximum results.

I don't train every piece of junk or UCE as spam in my Bayes DB. Some senders just need to be reported to Spamcop and blacklist_from because their content looks just like ham. I bet many Bayes DBs out there are "confused" by improper training. I train ham first then spam second so similar content will go toward the spam classification since most of the trusted senders are already covered by a whitelist_auth entry.

Never let end users decide the ham/spam Bayes training either. They don't understand and will mark anything they don't want at that particular moment as spam even if they signed up for it a few hours/days ago.

--
David Jones

Reply via email to