Enhancing Document-Level Relation Extraction by Entity Knowledge Injection

Document-level relation extraction (RE) aims to identify the relations between entities throughout an entire document. It needs complex reasoning skills synthesize various knowledge such as coreferences and commonsense. Large-scale graphs (KGs) contain a wealth of real-world facts, can provide valuable document-level RE. In this paper, we propose entity injection framework enhance current RE models. Specifically, introduce coreference distillation inject knowledge, endowing model with more general capability reasoning. We also employ representation reconciliation factual aggregate KG representations document into unified space. The experiments on two benchmark datasets validate generalization our consistent improvement several