Improved Exploration in Factored Average-Reward MDPs