ChronoR: Rotation Based Temporal Knowledge Graph Embedding

Despite the importance and abundance of temporal knowledge graphs, most current research has been focused on reasoning static graphs. In this paper, we study challenging problem inference over particular, task link prediction. general, is a difficult due to data non-stationarity, heterogeneity, its complex dependencies. We propose Chronological Rotation embedding (ChronoR), novel model for learning representations entities, relations, time. Learning dense frequently used as an efficient versatile method perform The proposed learns k-dimensional rotation transformation parametrized by relation time, such that after each fact's head entity transformed using rotation, it falls near corresponding tail entity. By high dimensional operator, ChronoR captures rich interaction between multi-relational characteristics Temporal Knowledge Graph. Experimentally, show able outperform many state-of-the-art methods benchmark datasets graph