Tuning Confidence Bound for Stochastic Bandits with Bandit Distance