TCL: Temporal Constraint Learning from Temporal Knowledge Graphs

Djawad Bekkoucha, Nadjib Lazaar, Pierre-Henri Paris, FATIHA SAIS

Keywords: Temporal Knowledge Graph, Knowledge Graph Consistency, Qualitative Constraint Network, Constraint Acquisition

TL;DR: This paper proposes a method for learning consistent temporal constraint networks from temporal knowledge graphs.

Abstract: Temporal Knowledge Graphs (TKGs) encode large collections of time-stamped facts. However, this information is often temporally inconsistent, which limits the effectiveness of abstraction and temporal reasoning. In this work, we model temporal relations between facts using Temporal Constraint Networks (TCNs). We introduce TCL, an approach that learns a compact set of temporal constraints directly from TKGs. To ensure the coherence of the learned network, we develop a propagation algorithm inspired by path consistency. The resulting TCNs provide a compact and informative representation of the temporal structure implicitly encoded in TKGs. They can serve as structured inputs for downstream tasks such as temporal reasoning, inconsistency detection, knowledge graph validation, and predictive inference. Experimentally, we demonstrate the quality of the TCNs produced by TCL through the obtained domains reduction rate, the quality of the generated inferences, and the support of the learned constraints, thereby highlighting the benefit of combining observed information with inferred knowledge.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 19