Efficient Frameworks for Generalized Low-Rank Matrix Bandit Problems