Learning Optimal Dialogue Management Rules by Using Reinforcement Learning and Inductive Logic Programming