Is a rating or ranking prediction more logical while developing a movie recommendation system? If the amount of ratings per user is high enough, factorizing the user-product matrix is the best thing, in my opinion. However, if the dataset is too sparse, the prediction can be extremely inaccurate.
Knowing this fact, I was exploring the RankSys library and found that one of the contributors is arguing that ranking is more logical. He did not provide any explanation, though. Later on, I talked to some recommender system developers and researchers and got to know that he probably meant ranking is less sensitive to the prediction error due to the gap between a number of ratings and items. The reason ...
