Tr-Sth: Temporal Knowledge Reasoning Based on Short-Term Historical Events

The paper introduces TR-STH, a novel Temporal Knowledge Graph (TKG) reasoning model. Unlike models that only focus on historical events that have already occurred and neglect short-term historical events, TR-STH reveals complex event connections and proposes three reasoning rules to deeply explore the implicit indirect historical relations between entities, thereby effectively expanding the set of historical events. Considering that short-term historical events may have a greater impact on the future, the model also introduces an exponential decay function, assigning corresponding weights based on the distance between historical events and the prediction time point, effectively amplifying the influence of neighboring events. We conduct experiments on two public datasets, YAGO and WIKI, and the results show that all indicators are improved by an average of 5 to 10 percentage points compared to the sub-optimal model. In addition, ablation studies also demonstrate the positive impact of each model component, providing a new solution for future research on the reasoning of temporal knowledge.