CRNet: Modeling Concurrent Events Over Temporal Knowledge Graph

Temporal knowledge graph (TKG) reasoning, which aims to extrapolate missing facts in TKGs, is vital for many significant applications, such as event prediction. Previous studies have attempted to equip entities and relations with temporal information in historical timestamps and have achieved promising performance. While ignoring the likelihood that future occurrences would occur simultaneously, they independently forecast the missing data. However, there are complicated connections between future concurrent events that might correlate with and influence one another. Therefore, we propose our Concurrent Reasoning Network (CRNet) to leverage event concurrency in both historical and future timestamps for TKG reasoning. Specifically, we select the top-k candidate events for each missing event and construct a candidate graph based on the candidate events of all missing events at the future timestamp. The candidate graph connects missing facts by sharing the same entities. Furthermore, we employ a novel relational graph attention network to represent the interactions of candidate events. We evaluate our proposal by the entity prediction task on three well-known public event-based TKG datasets. Extensive experimental results show that our CRNet complete future missing facts with a 15–20% improvement over MRR.

Read The Full Article.