Temporal KGs
KG: Knowledge Graph
Overview
Temporal knowledge graphs model entities and their relationships as they change over time, supporting link and event forecasting, temporal question answering, and applications such as clinical timelines and conversational memory. Common approaches include time-aware embeddings and rotations, sequence and state-space models, Hawkes processes, graph-path search, and temporal logic; newer systems also combine graphs with language models for retrieval and reasoning. Evaluation spans benchmarks such as ICEWS, GDELT, and MultiTQ, with tasks ranging from future-event prediction to questions about when facts held. A central challenge is reliable extrapolation under distribution shift or sparse histories, while simple memorization can remain competitive and aggregate metrics may hide failures on rare events or complex temporal constraints.
Latest papers
69 papers7 papers in the last four weeks, down 22% on the four weeks before. 0.0% of all new papers.