Details about the Bayesian module

Hi to all,

I was wondering about the Bayesian filter in Thunderbird. I know other
spam filters with Bayesian modules that consider all incoming mail
good as long as the filter itself or the user does not specifically
mark mails as spam. This is necessary for the Bayesian module to learn
what good mails look like.

How is it with Thunderbird? Does Thunderbird learns good mails
automatically from all the mails that arrive in the inbox?

Thanks for your answers!

Niko Klausner <oos@ofdka.bwl.de> wrote:

> Hi to all,
> 
> I was wondering about the Bayesian filter in Thunderbird. I know
> other spam filters with Bayesian modules that consider all
> incoming mail good as long as the filter itself or the user does
> not specifically mark mails as spam. This is necessary for the
> Bayesian module to learn what good mails look like.
> 
> How is it with Thunderbird? Does Thunderbird learns good mails
> automatically from all the mails that arrive in the inbox?

No, you must train it just like the Bayesian filters but you'll 
probably not be very happy with it, you might consider running POPFile 
or K9.

Niko Klausner schreef:
> Hi to all,
> 
> I was wondering about the Bayesian filter in Thunderbird. I know other
> spam filters with Bayesian modules that consider all incoming mail
> good as long as the filter itself or the user does not specifically
> mark mails as spam. This is necessary for the Bayesian module to learn
> what good mails look like.
> 
> How is it with Thunderbird? Does Thunderbird learns good mails
> automatically from all the mails that arrive in the inbox?
> 
> Thanks for your answers!
> 
Hi, recognizing good from the bad requires training, as you probably 
know. The message menu has an option to mark the message in focus as 
""not-junk"". Alternatively, you can mark the Junk column:
once for Junk 
and once more for Not-Junk.(Note: initially the Junk state is Undecided, 
which is not visible in the column.)

My adaptive filter, trained along time ago, catches almost any spam, 
that passes the filter of the Mail Account services. I am quite happy 
with its performance.

On 14/04/08 09:55, Mel wrote:
> Niko Klausner schreef:
>> Hi to all,
>>
>> I was wondering about the Bayesian filter in Thunderbird. I know other
>> spam filters with Bayesian modules that consider all incoming mail
>> good as long as the filter itself or the user does not specifically
>> mark mails as spam. This is necessary for the Bayesian module to learn
>> what good mails look like.
>>
>> How is it with Thunderbird? Does Thunderbird learns good mails
>> automatically from all the mails that arrive in the inbox?
>>
>> Thanks for your answers!
>>
> Hi, recognizing good from the bad requires training, as you probably
> know. The message menu has an option to mark the message in focus as
> ""not-junk"". Alternatively, you can mark the Junk
column: once for Junk
> and once more for Not-Junk.(Note: initially the Junk state is Undecided,
> which is not visible in the column.)
>
> My adaptive filter, trained along time ago, catches almost any spam,
> that passes the filter of the Mail Account services. I am quite happy
> with its performance.
>
> Regards, M.

You can also include "Set junk Status as" "[not] Junk" as an
action in 
your mail filters, if you know that some kind of mail should never be 
spam -- but be cautious, it could teach your filters wrongly if applied 
indiscriminately.

My junk filters are quite good but not perfect: at times I have to mark 
as "junk" something that was left in my Inbox instead of falling into

the junk folder. The opposite (fishing "good" mail out of Junk after 
marking it as non-junk) I practically never need to do anymore. Both 
should of course be done when appropriate, to "teach" the junk
filtering 
system.


Best regards,
Tony.
--