Impressive work and great explanations! I know first hand that explaining these kind of things can be very though. It looks like you invested a lot of time in this article and I am very grateful for it. I learned a great deal!

Now I am very tempted to try this out in my own domain (Recommendations). Especially, since I work with implicit preference datasets which are inherently noisy (clicks are considered positive engagements).