Fig. 2
From: Reliable knowledge graph fact prediction via reinforcement learning
Illustrations of a framework for a KG fact prediction model based on RL. (a) The KG environment is modeled as an MDP environment. The black and blue lines represent a target relation trained by RL and a reasoning path obtained by the agent through a random walk, respectively. (b) The agent interacts with the MDP environment and takes action based on the policy network to extend the reasoning path