Sunday, May 19, 2013

Amazon needs Machine Learning about Assholes

You know what's missing from Amazon etc. model for making recommendations?

A human model.

I'm always getting recommendations or seeing ads for stuff I just bought.  If I bought 48 rolls of toilet paper for $13.98, is there a reason to try to sell me more right away?  Nope.  There should be a human model.  At first it could be generic, but eventually using my "big data" profile (e.g. single guy) to estimate my toilet paper consumption.   It should bombard me with toilet paper deals a couple weeks before my supply runs out.

Ditto for wiper blades, shampoo, appliances.  I buy a dishwasher and they push adds on my for another dishwasher?   Detergent, or dishes even (people with dishwashers have more dishes).   But another dishwasher?

Clearly there is no human model of consumption behind these recommendations.  Just recent purchase info. 

If they had a few simple profiles - with age, marital status, sex, etc. and had some profile for each item (or category since there are too many items) of a reasonable expected consumption, they'd do much better at predicting what people are likely to want to buy.

For the toilet paper example, a human model would tell them that most people have just one asshole.


Post a Comment

Links to this post:

Create a Link

<< Home