Incorporating multi-perspective information into reinforcement learning to address multi-hop knowledge graph question answering.