Improved Analysis for Bandit Learning in Matching Markets