MAE and related metrics provide a static view of the predictive performance
of a recommender system outside of time. Often, though, the
performance of the recommender over time is interesting: does the recommender
generate better results as more users use it? For individual
users, how does accuracy change as they add more ratings?