Knowledge Graphs (KGs), both manually and automatically constructed, are far from complete. Recently, Knowledge Graph Completion (KGC) has been receiving a great deal of attention. Link prediction is a key task in KGC. In this work, we research multi-hop link prediction based on reinforcement learning (RL) in KGs. More specifically, we propose a novel RL framework for learning more accurate link prediction models: we frame link prediction problem in KGs as an inference problem in probabilistic graphical model (PGM) and use maximum entropy RL to maximize the expected return. Our experimental evaluations show that our proposed method performs well on FB15K-237 and NELL-995 datasets.
Paper
Full text
A Multi-Hop Link Prediction Approach Based on Reinforcement Learning in Knowledge Graphs
Semantic Scholar · Computer Science · 2018
Abstract
Knowledge Graphs (KGs), both manually and automatically constructed, are far from complete. Recently, Knowledge Graph Completion (KGC) has been receiving a great deal of attention. Link prediction is a key task in KGC. In this work, we research multi-hop link prediction based on reinforcement learning (RL) in KGs. More specifically, we propose a novel RL framework for learning more accurate link prediction models: we frame link prediction problem in KGs as an inference problem in probabilistic graphical model (PGM) and use maximum entropy RL to maximize the expected return. Our experimental evaluations show that our proposed method performs well on FB15K-237 and NELL-995 datasets.