Tensor Decomposition-Based Temporal Knowledge Graph Embedding
Lifan Lin, Kun She · 2020
In order to meet the problems caused by sparse data and computational efficiency, knowledge graph (KG) is adopted to represent the semantic information of entities and relations as dense and low-dimensional vectors. While conventional KG representation methods mainly focuse on static data. These methods fail to deal with data that evolves with time which may only be valid for a certain period of time. To accommodate this problem, a temporal KG embedding model based on tensor decomposition is proposed in this paper, which regards the fact set in the KG as a fourth-order tensor including head entities, relations, tail entities and time dimensions. This method can be further generalized to other static KG embedding based on tensor decomposition. With experiments on temporal datasets extracted from real-world KG, extensive experiment results show that our approach outperforms state-of-the-art methods of KG embedding.