Logic and Commonsense-Guided Temporal Knowledge Graph Completion

A temporal knowledge graph (TKG) stores the events derived from data involving time. Predicting is extremely challenging due to time-sensitive property of events. Besides, previous TKG completion (TKGC) approaches cannot represent both timeliness and causality properties events, simultaneously. To address these challenges, we propose a Logic Commonsense-Guided Embedding model (LCGE) jointly learn representation together with time-independent perspective commonsense. Specifically, design rule learning algorithm construct rule-guided predicate embedding regularization strategy for among Furthermore, could accurately evaluate plausibility via auxiliary commonsense knowledge. The experimental results TKGC task illustrate significant performance improvements our compared existing approaches. More interestingly, able provide explainability predicted in view causal inference. appendix, source code datasets this paper are available at https://github.com/ngl567/LCGE.