Re: Problems with Bayesian filtering
Laird Breyer <[email protected]> Thu, 26 Feb 2004 12:07:49 +1000
| Newsgroups | gmane.ietf.asrg.filtering |
|---|---|
| Message-ID | <20040226020748.GA31549@ender> |
On Feb 25 2004, John Graham-Cumming wrote: > > No filter can become a combined HTML, Javascript, Flash etc. interpreter. > > They don't need to. If they recognize the different elements being used > that's often enough to recognize the message as spam. As for > Javascript, that's already happening: Nice to see that the spammers are on the ball ;-) What's problematic with this however is how it interacts with the things a filter normally throws out. Unless you're claiming that the mere existence of a tag such as > <script language="JavaScript"> is generally a sufficient clue, you'll be sucked into dealing with, analysing and managing all the miriad tokens within the script body. At some point, you have to decide that some parts should be skipped when seen, and that's where the next attack lies (which is what I mean by unwinnable). There has to be a better way of dealing with such meta data, as a black box rather than interpreting it. -- Laird Breyer.