Very high level overview of a company that provides an email marketing solution based on customer's past buying habits. From the end:
“I still get e-mails from Amazon recommending books based on the Jared Diamond titles I bought three years ago,” he said. “But I get nothing about my interest in gardening.
Same Author recommendtions are much easier, an stronger, than same subject. And it's hard to notice shifts in customer interests- you might think that you're really interested in gardening, but you only buy 2 books on it, compared with dozens of another subject. I find myself frustrated with the same thing, even though I know I can look at the list of my purchases and searches and see a different pattern appearing there than I think there should be.
On the similar interests filtering topic, I first heard about this stuff via Firefly at the Media Lab. Wired has an article from long ago about its demise, Firefly's Dim Light Snuffed Out.
After all, Firefly is more than just another failed Microsoft Web venture. As far back as 1996, the technology, and the community that piggybacked on top of it, stood out as one of the most potent properties anywhere.
In essence, Firefly was a collaborative filter -- a technology that asked users what they liked, learned their tastes in music, then got them in touch with people having similar tastes.
Five years and several new paradigms later -- and following the company's 1998 buyout by Microsoft -- the light is going out for good on the forums. The underlying technology will live on, however, powering Redmond's e-commerce efforts.
Some of the service's users clearly long for the good old days.
"What the hell happened to the fly?" wrote one displaced Firefly user in an MSN forum. "It went down for a few days and then BLAM!!!!!! ... They decided to shut it down ... Does anybody remember when there was over 400 people on at one time in the fly?"
MIT professor Patti Maes does. She headed up the software agents group at MIT's Media Lab and led the development of the technology that would eventually spin off to become Firefly.
So that would make it 10 years ago when I was a junior or senior in college. This kind of recommendations filtering has changed a ton in that time, but it also has remained pretty static. Sure, now you can use MapReduce and we have several orders of magnitude more data, but at the root it's still the same basic algorithm.