From the spambayes mail list:
Patent application on adaptive spam filtering
Patent application on whitelists, blacklists, challenge-response, and
digital signatures used in spam-fighting
I have seen some fear expressed that the first patent above covers all Bayesian spam filters.
I looked at this last night.
I am not a lawyer, so don't go to the bank on what I say. And I didn't spend a huge amount of time on it.
But I do have some experience with patents, and I do understand the spambayes approach and the gist of the patent. It is my impression that the patent does not have a scope that encompasses Graham-derived ("Bayesian") filters, because they do not calculate "first" and "second" "symantic anchors" as the term is used in Claim 1.
They seem to be trying to make a straightforward adaptation of technology that works well for classifying documents according to subject area, latent semantic analysis, into the spam realm.
It would be very, very interesting to code and test their algorithm's performance against that of spambayes or bogofilter, for example.
One aspect of using latent semantic analysis is that it treats synonyms of known spammy words much as it does the spammy words themselves. It's sophisticated technology. But I'm not sure that its advantages matter much for spam detection with the kind of data we have available. It would be very interesting to know.
10:40:07 PM
|