Dynamic knowledge graph completion through time-aware relational message passing

As the structure of knowledge graphs may vary over time, static knowledge graph completion methods do not deal with time-varying knowledge graphs. However, examining the paths between entities and entities' context information can lead to more accurate completion methods. This paper attempts to complete dynamic (time-varying) knowledge graphs by combining time-aware relational paths and relational context. The proposed model can improve dynamic knowledge graph completion methods by leveraging neural networks. Experimental results conducted on two standard datasets, ICEWS14 and ICEWS05-15, indicate our model's superiority in terms of Mean Reciprocal Rank (MRR) and Hit@k over its well-known counterparts, such as DE-TransE and DE-DistMult.