18 ms·
I remember reading a blog post somewhere describing the problem with modern recommendation engines being their reactive nature. Say you purchase a cell phone on
by tonysdg 8y ago
I remember reading a blog post somewhere describing the problem with modern recommendation engines being their reactive nature. Say you purchase a cell phone on Amazon -- they'll start showing you other cell phones, cell phone accessories, more electronics, etc. The problem is, you just bought a phone -- you don't need 5 more, and you max need 1 or 2 cases. So the recommendations are garbage, since the engine has no way of knowing what your current mindset is and what you actually might be interested in.
In hindsight, that's obvious, but it struck me at the time. I don't ever want to see my thoughts get read by a computer (personal choice), but until such technology exists, recommendation engines will continue to suck.
- ghaff 8y agoThere are a bunch of problems. Paul Lamere has done a number of talks on the topic at SXSW and the like. I know him from Sun Labs but I think he's at Spotify these days. https://www.coursera.org/lecture/matrix-factorization/recommending-music-interview-with-paul-lamere-665tz https://www.coursera.org/lecture/matrix-factorization/recomm...
- derefr 8y ago> the engine has no way of knowing what your current mindset is and what you actually might be interested in You'd think, with enough data, you could build causal models of purchases of product types. Amazon certainly has associative models (people who buy this also buy that)—it seems like they would just need to add a temporal dimension to the input data.