Answers are grounded in this paper and cite their sources.
Temporal knowledge graph (TKG) embedding has received increasing attention in the academia. However, most existing methods are extensions of traditional translation models. Due to their intrinsic limitations, it is often difficult for such methods to effectively model essential characteristics of TKG, namely three basic relation patterns including symmetry/antisymmetry, inversion, and composition. In this paper, a new 3-Dimensional Rotation Temporal Embedding (3DRTE) method is proposed. Firstly, we selectively fuse temporal and relational features of fact triples by taking advantages of self-attention mechanism in processing sequential information. Then, entities are modelled as points in three-dimensional space, and the relations are interpreted as two isoclinic rotations between entities with Quaternion. Experimental results on several public datasets show that our method obtains state-of-the-art results.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2020) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: