Popfile continues to filter my email, and does so excellently. Some statistics (since June):

Bucket	Classification Count	False Positives	False Negatives
legit	2,129 (60%)	9	7
spam	1,444 (40%)	7	9

Paul Graham predicted that the use of these filters would make spam increasingly bland, so that it would be indistinguishable from normal mail. And indeed I get a lot more spam that just says "Hey, check this out:". Invariably, Popfile throws them in spam bucket. How does it know? I checked the analysis of one such instance:

Word	spam	legit
click	0.0065	0.0004
get	0.0060	0.0010
here	0.0055	0.0002
removed	0.0019	0.0000
future	0.0012	0.0002
mailings	0.0012	0.0000

The numbers in the spam and legit column are the relative odds that each word is part of a commercial or legitimate message. There you have it: those "click here to be removed" signoffs actually serve a useful purpose.

posted at: 01:22 | path: /en/computers | permanent link

Foto's van econometristen in de Ardennen (2002 ) staan weer on line, op verzoek.

posted at: 03:41 | path: /nl/fotos | permanent link