You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Roman Sozinov <le...@newmail.ru> on 2007/06/21 14:20:13 UTC
Bayes became to work very bad
I'm using spamassassin about 1 year and for that period I already have good
BAYES tokens base.
But about 2 weaks ago began something wrong - my system became to catch spam
very bad.
About 80% of spam have BAYES_50 score :(
What's wrong?
I'm using Spamassassin 3.1.8 with mysql backend (awl & bayes)
In my bayes base there are:
spam_count - 17539
ham_count - 4895
token_count - 453505
Roman
--
View this message in context: http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11231927
Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayes became to work very bad
Posted by jean-philippe luiggi <je...@didconcept.com>.
Hello Roman,
Perhaps we could get other advices but i think
learning is still a good thing.
About the old BAYSES-base, i've no opinion, i think it may still be
valuable but i've no idea of the accuracy.
Best regards,
Jean-philippe.
On Thu, 21 Jun 2007 06:25:22 -0700 (PDT)
Roman Sozinov <le...@newmail.ru> wrote:
>
>
>
> jean-philippe luiggi-2 wrote:
> >
> >
> > In the last one i got, there were 3 words relating to
> > spam for more than 200 regular.
> >
> > Best regards.
> >
> Thanks for quick answer, but what about advice. What to do? Learn?
> What to do with old BAYES-base?
>
> Roman
> --
> View this message in context:
> http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11232591
> Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
>
Re: Bayes became to work very bad
Posted by Roman Sozinov <le...@newmail.ru>.
jean-philippe luiggi-2 wrote:
>
> Hello Roman,
>
> It's not a problem with SA but with the bayes's concept instead.
> Keeping an "big" but "old" bayes's database doesn't mean
> you'll catch the new spam.
> Each day, spammers change them in order to evade detection so in
> order to be as accurate as possible you need to learn often.
>
> In the last one i got, there were 3 words relating to
> spam for more than 200 regular.
>
> Best regards.
>
Thanks for quick answer, but what about advice. What to do? Learn? What to
do with old BAYES-base?
Roman
--
View this message in context: http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11232591
Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayes became to work very bad
Posted by jean-philippe luiggi <je...@didconcept.com>.
Hello Roman,
It's not a problem with SA but with the bayes's concept instead.
Keeping an "big" but "old" bayes's database doesn't mean
you'll catch the new spam.
Each day, spammers change them in order to evade detection so in
order to be as accurate as possible you need to learn often.
In the last one i got, there were 3 words relating to
spam for more than 200 regular.
Best regards.
n Thu, 21 Jun 2007 05:20:13 -0700 (PDT)
Roman Sozinov <le...@newmail.ru> wrote:
>
> I'm using spamassassin about 1 year and for that period I already
> have good BAYES tokens base.
> But about 2 weaks ago began something wrong - my system became to
> catch spam very bad.
> About 80% of spam have BAYES_50 score :(
> What's wrong?
>
> I'm using Spamassassin 3.1.8 with mysql backend (awl & bayes)
> In my bayes base there are:
> spam_count - 17539
> ham_count - 4895
> token_count - 453505
>
> Roman
> --
> View this message in context:
> http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11231927
> Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
>
>
>
>
>
>
> !DSPAM:1,467a6f05135801198567559!
Re: AW: Bayes became to work very bad
Posted by Matthias Haegele <mh...@linuxrocks.dyndns.org>.
Joerg Reisslein schrieb:
>
> Mit freundlichen Gru?en
> Do you have a link for the botnet plugin?
$searchmachine "download botnet plugin spamassassin"
http://people.ucsc.edu/~jrudd/spamassassin/
Docs in tarball provide details for install.
--
hth
MH
Dont send mail to: ubecatcher@linuxrocks.dyndns.org
--
AW: Bayes became to work very bad
Posted by Joerg Reisslein <j....@schmitt-aufzuege.de>.
Mit freundlichen Gru?en
Do you have a link for the botnet plugin?
cheers
> -----Ursprungliche Nachricht-----
> Von: Roman Sozinov [mailto:levsha@newmail.ru]
> Gesendet: Donnerstag, 21. Juni 2007 16:01
> An: users@spamassassin.apache.org
> Betreff: Re: Bayes became to work very bad
>
>
>
>
> Matthias Haegele-2 wrote:
> >
> > Use blacklists, the botnet plugin, SARE rules, sa-update ...?
> > upgrade to a newer SA release?
> >
> Thanks :)
> What about my BAYES base? delete it? Do it from new?
> --
> View this message in context: http://www.nabble.com/Bayes-became-to-
> work-very-bad-tf3958378.html#a11233759
> Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayes became to work very bad
Posted by Roman Sozinov <le...@newmail.ru>.
Matthias Haegele-2 wrote:
>
> Use blacklists, the botnet plugin, SARE rules, sa-update ...?
> upgrade to a newer SA release?
>
Thanks :)
What about my BAYES base? delete it? Do it from new?
--
View this message in context: http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11233759
Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayes became to work very bad
Posted by Matthias Haegele <mh...@linuxrocks.dyndns.org>.
Roman Sozinov schrieb:
>
> Matthias Haegele-2 wrote:
>> You use sa-learn till now?. Some people suggest not to learn old
>> spam/ham ...
>> Think its pretty normal that bayes hits are not very good on new spam
>> (spammer tweak their messages every day to slip the filters ...).
>> Some new spam messages here only get BAYES_00 ...
>>
> sa-learn sometimes use (~3 times in a week)
> So what about some advice? :)
Use blacklists, the botnet plugin, SARE rules, sa-update ...?
upgrade to a newer SA release?
--
Grüsse/Greetings
MH
Dont send mail to: ubecatcher@linuxrocks.dyndns.org
--
Re: Bayes became to work very bad
Posted by Roman Sozinov <le...@newmail.ru>.
Matthias Haegele-2 wrote:
>
> You use sa-learn till now?. Some people suggest not to learn old
> spam/ham ...
> Think its pretty normal that bayes hits are not very good on new spam
> (spammer tweak their messages every day to slip the filters ...).
> Some new spam messages here only get BAYES_00 ...
>
sa-learn sometimes use (~3 times in a week)
So what about some advice? :)
--
View this message in context: http://www.nabble.com/Bayes-became-to-work-very-bad-tf3958378.html#a11232378
Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayes became to work very bad
Posted by Matthias Haegele <mh...@linuxrocks.dyndns.org>.
Roman Sozinov schrieb:
> I'm using spamassassin about 1 year and for that period I already have good
> BAYES tokens base.
> But about 2 weaks ago began something wrong - my system became to catch spam
> very bad.
> About 80% of spam have BAYES_50 score :(
> What's wrong?
You use sa-learn till now?. Some people suggest not to learn old
spam/ham ...
Think its pretty normal that bayes hits are not very good on new spam
(spammer tweak their messages every day to slip the filters ...).
Some new spam messages here only get BAYES_00 ...
> I'm using Spamassassin 3.1.8 with mysql backend (awl & bayes)
> In my bayes base there are:
> spam_count - 17539
> ham_count - 4895
> token_count - 453505
>
> Roman
--
Grüsse/Greetings
MH
Dont send mail to: ubecatcher@linuxrocks.dyndns.org
--