Specific Time Embedding for Temporal Knowledge Graph Completion

The knowledge graph can be used as a corpus of cognitive computing, in this paper we mainly focus on the temporal knowledge graph. Temporal knowledge graph(TKG), as an extension of static knowledge graph(KG), can be used to deal with dynamic and time-varying knowledge in the real scenario, because many relations are only valid for a certain period, so it can ensure time consistency. Therefore, TKG has received more and more attention. KG embedding (KGE) is an enabling technique for KG completion(KGC), it can complete missing entities in tuples by discovering latent relations between representations. The previous methods mainly focus on static KGC(SKGC), with the emergence of TKG, temporal KGC(TKGC) should be developed. Currently, existing methods for TKGC, either consider changing the representation by temporal information or directly using temporal information to complete. In this paper, we inspired by quantum theory in a sense to propose specific time transE. We note that entities and relations are not time-restricted, only when they are combined to form tuples, the validity of tuples relies on time. We assume that entities and relations can get a determined status after being observed by a specific time, i.e., we use temporal information to get the specific representation of entities and relations. Tuples composed of these specific representations must be related to a specific time, and we use distance model transE to quantify correlation. Finally, through extensive experiments on TKGC datasets, the experimental results verify the validity of our models.