An Improvement of Diachronic Embedding for Temporal Knowledge Graph Completion

Knowledge graph completion (KGC) aims to predict missing information in a knowledge graph (KG). Knowledge embedding approaches learn the representations of entities and relations in a static knowledge graph, playing an essential role in knowledge graph completion. However, the previous methods have been designed for static knowledge graphs. A temporal knowledge graph (TKG), on th other hand, contains timestamp-based facts indicating relationships between entities at different times. R. Goel et al. (2020) introduced novel models for temporal knowledge graph completion through existing static models with a diachronic entity embedding function, which are able to provide characteristics of entities at any point in time. Moreover, their method is flexible since it can be potentially combined with any existing static model. In this paper, we propose two models, - (DE-RotatE and DE-RotatE-sinc) -, to combine the diachronic entity embedding with RotatE - the model is introduced by Z. Sun et al. (2019). Through experiments, we show that the results of our proposed models are better than those in the work by R. Goel at al. Our source code is available on Github, https://github.com/anhntt1202/DE-RotatE.