Incorporating anticipation embedding into reinforcement learning framework for multi-hop knowledge graph question answering | Litlas