Matrix factorization is a way to generate low rank matrix when multiplying two differentmatrices, and thanks to that we can assimilate missed reviews, derived by examining theassociations between the users and the items. Hence, we can predict if a user is likely to givea movie high rating or low rating, and given that information the system recommend themovie to that user.
and
which estimates as close as possible the matrix
, in other words, we want to minimize a function
Our Pipeline can be summarized as follows :
In this project we addressed the problem of Matrix factorization, where we implemented two approaches to solve it which are : stochastic gradient descent and Alternating least squares.Through this study, we can conclude that stochastic gradient descent with biases is a good solution for this problem, since it gives better results and takes less running time.