Perceiving the World: Question-guided Reinforcement Learning for Text-based Games