I am currently studying the recommendation system part and learnt that we have cost function for weights and bias for each user and we have cost function for features of each movie and to find one we need other. but shouldn’t the features be in the data to make a recommendation. for example movies are given and feature is it is romantic or action. so according to video we learnt with the weights and bias given of each user we can find these features of a movie which doesn’t make sense. plz help me clear this doubt
In the type of recommender system discussed here, we have no features of the movies -we have only the ratings.
The ratings are the labels for each movie. Those are the outputs.
So we have to derive both the features and their weights.
Andrew’s discussion of “action” and “romance” is just to give an intuitive explanation for how the algorithm might work.
But in practice, we don’t have those features in the dataset. We only have the ratings from users for the movies they have watched.
But later he merged to cost functions and made one with both weights bias and features. moreover if there are no features(x) then how is the prediction(ratings) made and if no predictions are made there wont be weights and bias of each user.
Have you watched all of the lectures on this topic?
no i am midway
Please post back after you’ve watched all of the lectures.
I have finished the lectures and I still have doubt in collabaritive filtering recommendation system
After having watched the lectures, what is your doubt specifically?