Learning Temporal and Spatial Embedding for Temporal Knowledge Graph Reasoning

Temporal knowledge graphs store a large number of temporal facts that simulate the dynamic interactions of entities along the timeline. Since existing temporal knowledge graphs often suffer from incompleteness, it is crucial to build time-aware representation learning models that help to infer the missing temporal facts. However, most of the existing models for temporal knowledge graph reasoning focus on mining temporal associations between entities, and do not fully exploit spatial information contained in entities. To this end, we propose spatial-temporal network(ST-Net), a new representation learning model for temporal knowledge graphs, which has both temporal and spatial awareness capabilities. Specifically, ST-Net enriches the hidden features of entities by simultaneously fusing their temporal and spatial information. At the same time, we introduce the core idea of Copy-Generation Networks, which predicts future facts based on either the historical vocabulary or the whole entity vocabulary. We evaluate our proposed method via link prediction at future times on three benchmark datasets. Through extensive experiments, we demonstrate ST-Net has superior performance on the future link prediction tasks.