Temporal Knowledge Graph Completion Based on Temporal Dependency and Commonsense Guidance

The Temporal Knowledge Graph (TKG) extends traditional knowledge graphs by introducing a temporal dimension, but the challenges of temporal dependency and sequential characteristics make completion and prediction more difficult. Existing methods for Temporal Knowledge Graph Completion (TKGC) often struggle to simultaneously consider the temporal dependency of events and the consistency of common sense, which limits their reasoning effectiveness. To address this issue, this paper proposes a new model, TDCGC. This model combines a Temporal Transformer with an innovative temporal dependency rule learning algorithm, utilizing predicate embedding regularization guided by temporal dependency rules along with Temporal Transformer for temporal modeling. The Temporal Transformer effectively captures the temporal dependencies between events, while the commonsense-based time-invariant representations are used to score events. Moreover, the model can effectively capture complex temporal interactions between events by delving deeply into the long-term dependencies of time series, thereby enhancing the completion capabilities of the knowledge graph. Experimental results show that this model improves the Mean Reciprocal Rank (MRR) by 1.1% and 1.2% on the ICEWS14 and ICEWS05-15 datasets, respectively, demonstrating a significant enhancement in its adaptability and reasoning ability when handling temporal-dependent data.