undefined | Better HN

0 pointsgaryrob3y ago0 comments

A statistical technique I developed was incorporated into a number of award-winning spam filters, including SpamAssassin.[1]

I'm also apparently the original inventor of the tracking cookie, which had the implication that no one was able to patent it. It was presented in a patent of mine[2] that was about a collaborative filtering technique for recommending ads; I'd come up with the tracking cookie mechanism to support that technique. So, it didn't attempt to patent the tracking cookie separately; but because it was the first publication describing the method, no one else could patent it either. In 2021 a joint legal brief filed by Google and Twitter together, defending themselves against a patent troll, called it "Robinson's Cookie". My patent is owned by Google now. It contained a lot of details for giving users control of the data derived from tracking; that part was pretty much ignored by people implementing it.

[1] https://www.linuxjournal.com/article/6467 [2] https://patents.google.com/patent/US5918014A

0 comments

garyrobOP3y ago

The spam filter stuff used Paul Graham's word probabilities described in his seminal article A Plan For Spam. It changed those probabilities a bit, to better account for the number of emails a word appeared in. But in the main, my article was about a statistical method (NOT Bayesian) for combining those probabilities. The word probabilities were Bayesian, but the way of combining them used frequentist statistics. Even so, spam filters that used the technique were always referred to as Bayesian as if nothing frequentist was involved.

HaZeust3y ago

Legendary story, we've used your spam filter contributions - thank you!

j / k navigate · click thread line to collapse