Does this implementation take time-ordered user-movie sequences as input? It appears to be using each rating independently.