Hi Bill!

On Mon, 30 Aug 2004 19:58:06 -0400 (EDT), William Stearns wrote:
> Good evening, Mat,
> On Tue, 31 Aug 2004, Mat Bowen wrote:
>> If this has been covered before please point me in the right
>> direction
>> (i've had a search but couldn't find anything).
>> What are the implications of running the DNSBL tests but disabling
>> their scoring, instead letting the Bayes filter pick up on the
>> success
>> of the tests? I realise the DNSBL tests work 'out of the box' and
>> that
>> would be lost, but on the other hand would it not be better to let
>> Bayes work out the scoring based on these results?

> I must admit I'm a bit confused.  If you trust the DNSBL results
> enough to affect bayes' learning, why would you also not trust them
> enough
> to affect the overall score?

Personally i'm happy with the current setup but i was asked this
question and it got me thinking, so i thought the knowledgable people
on this list would be able to share their opinions.

> That said, if you really want this, how about setting the scores
> for each to, say, 0.1?  That'll be so small it'll be unlikely to
> affect
> the overall score significantly, but still have the tokens
> available for
> bayes learning.

Yes that would do what i was suggesting, but why not do that from the
start, why assign scores to these tests at all? The reasoning behind
the question is that Bayes will be better suited to judge how to score
these tests based on individual users spam/ham. For example, some
tests blacklist yahoo.com, obviously for a user with lots of contacts
who use yahoo.com will want their scoring to adapt to give a lower
score to this test, as classification by Bayes would do.

Does that make any sense? What do you think?

Thank you,
Mat

Reply via email to