Amazon's recommendation engine, I believe, relies heavily on collaborative filtering. The theory behind collaborative filtering reads something like: I will tend to like things that people like me like, where "people like me" is defined to be: people that perform similar actions, like buy the books I buy, or rate movies they way I do. Now, take a moment to go the Amazon page for the Uranium Ore product located at:
and check out the "Customers that bought this item also bought ..." section. So after laughing for about 5 minutes, I started to think of why this might happen.
Here's a theory I had: there exists a group of users that like to spoof Amazon's engine for laughs. So they go out and provide a bunch of recommendations on goofy products, like those appearing as a recommendation for Uranium Ore. Then, they comb Amazon for products that don't have a lot of purchase transactions. Since Amazon's recommendation engine relies more heavily on purchase data to determine similar users to then pivot into recommendations, in the absence of this data, I reasoned, it probably relies more heavily on user provided ratings. So, without having to spend any money, the spoofers can simply rate the product well, and voila, out pop the G-String recommendations.
Well I don't think this explanation cuts it. I went through and followed recommendations made by a bunch of users w/ comments like: This is the best strawberry jam I've ever tasted. It works well with toast AND English muffins, and these users certainly did not make recommendations of the G-String kind. So, what gives?