Text this: Thompson Sampling for Non-Stationary Bandit Problems