For all the sophistication claimed in the article, it feels like my recommendations could just as accurately be driven by simple keywords. I wonder how often, if ever, they compare their current solution to much simpler algorithms?