Is There a Difference Between Recommendation and Machine Learning?

There's a number of deep learning type approaches in to the one that's contributed in your paper. I guess we can compare them by accuracy in the traditional machine learning way. But for example, the deepen cross network paper stated that an improvement of point zero zero one in low loss is considered commercially significant. Are there other metrics though, that we might want to look at, considering we're going to run this scale? So not in a research environment, i guess there's kind of two things that you might need to look at - revenue per impression and click through rate.