why mean normalization is used here, because what we are doing is replacing new user rating with mean of that movie. We can directly replace it with average of that particular row.
Hello @Aman6,
Your approach will lead to a different trained model because you will have imputed some ratings that will be otherwise not used for training. The current approach keeps the training set to contain only actual ratings, while letting new users be getting a mean prediction.
The next question is, should we impute the ratings? My answer is no, because the goal of collabrative filtering is to predict the missing rating from the existing ones. If we had imputed them from the beginning, it would lose the purpose.
Thoughts?
Cheers,
Raymond