RTA: A reinforcement learning-based temporal knowledge graph question answering model

Yu Zhu, Tinghuai Ma, Shengjie Sun, Huan Rong, Yexin Bian, Kai Huang

Abstract: Highlights•We design a Reinforcement learning Temporal knowledge graph question Answer (RTA) framework, which is specifically designed for the TKGQA task.•We note that complex temporal questions’ multiple entities may lead to model’s redundant information or interference in reasoning; our model mitigates it by extracting context to select topic entities in question understanding.•We introduce reinforcement learning for reasoning in TKG and implement a dynamic path-matching module in the policy network to aggregate path features and help the agent explore the path effectively.