Temporal knowledge graph representation learning with local and global evolutions

Temporal knowledge graph (TKG) representation learning aims to project entities and relations in TKG to low-dimensional vector space while preserving the evolutionary nature of TKG. Most existing methods separately treat knowledge that happened at different times, which fails to explore how temporal knowledge graph evolves over time. Actually, TKG should evolve both on local and global structures. The local structure evolution describes the formation process of graph structure in a detailed manner, while the global structure evolution refers to the dynamic topology (e.g., community partition) of graph, which is derived from the continuous formation process. Both of them are key factors for understanding the evolutionary nature of TKG. Unfortunately, seldom attention has been paid to this aspect. In this paper, we propose a new TKG representation learning framework with local and global structure evolutions, named EvoExplore. Specifically, we define the local structure evolution as the establishment process of relations between entities, and propose a hierarchical-attention-based temporal point process to capture the formation process of graph structure in a fine-grained manner. For global structure evolution, we propose a novel soft modularity parameterized by entity representations to capture the dynamic community partition of TKG. Finally, we employ a multi-task loss function to jointly optimize the above two parts, which allows EvoExplore to learn the mutual influences of local and global structure evolutions. Experimental results on three realworld datasets demonstrate the superiority of EvoExplore compared with baseline methods.