On Sun, 13 Jan 2008, Mike Cisar wrote:
>However, these last bunch seem to have a trick, the only other text in the
>message aside from the URL seems to be a date string. Somehow that must
>totally be screwing with Bayes since those messages are also triggering
>BAYES_00 or BAYES_02 and pretty much obliterating the "btnI" scores with a
>high negative. I'm hesitant to train them as spam since I'm not sure
>whether that would do more harm.
>
>Any thoughts?
Ah! I saw those too and was wondering about the purpose of the date -
thanks for the explanation. :)
I agree with your hesitation.
I'd avoid feeding those to Bayes, and just focus on the offending trait.
I've already cranked up my "btnI" score to 7.65 (i.e. 150% of kill).
Anyone using the "btnI" rule listed earlier in this thread, should add the
"ls" TLD (the tiny yet breathtakingly beautiful Lesotho) to your lists of
google variants. Far better would be to just match on ALL google domains,
as John previously pointed out.
I've also seen this fascinating new variant:
"open google. enter nowsuperpill. press i am feeling lucky button!"
I had to laugh at that one. :)
- "Chip"
