SVD for collaborative filtering when number of latent concepts equals the rank

38 Views Asked by At

Consider we perform collaborative filtering on a rating matrix $A$ of rank $r$, where we assume that all the unobserved entries (items that haven't been rated per each user) are marked by zero.

If we use $K = r$ latent concepts (equals the rank of the matrix); will the system predict zero for all the unobserved entries?