I am doing the lab assignment for collaborative filtering. I noticed that we are making the weights and biases for each user (meaning each user will have their own model to predict). When I look at the cost function, it computes the cost of all the ratings of all users. I don’t understand why we need to compute the cost of all ratings of all users to optimize the weights and bias for each user while we can just compute the cost of each user and use that cost to optimize the weights and bias for that user only?
if I implement collaborative filtering in a movie streaming website, shouldn’t computing the cost of all ratings of all users each time I have a new user to make a predicted model for that user be very costly in terms of performance when the number of users increases?
Yes you are right that performing the same computation all over the dataset will be costly but probably the one here is not very big. There are some techniques to mitigate this issue. But you do need to have knowledge of the behavior of other users in order to make suggestions.
1 Like
Also, each user does not have its own model; the process intends to find similarities among users or items that are being observed. So, really, the other users are crucial to the model.
1 Like