Iterative rule-guided reasoning over sparse knowledge graphs with deep reinforcement learning

Yi Xia, Mingjing Lan, Junyong Luo, Xiaohui Chen, Gang Zhou

Abstract: Highlights•To the best of our knowledge, our proposed SparKGR constitutes the first iterative inference framework for sparse knowledge graph reasoning with reinforcement learning.•SparKGR tightly integrates the embedding-based, rule-based, and path-based methods to perform effective reasoning over knowledge graphs.•With our proposed dynamic path completion and iterative rule guidance strategies, the reinforcement learning agent in SparKGR explores efficiently in the process of interacting with knowledge graphs, and compensate for their incompleteness and sparsity.