Exploiting Coreference and Schema Structure for Document-level Event Extraction

Abstract

Document-level event extraction (DEE) extracts structured information of events from a document. Previous studies focus on improving the model architecture. We argue that exploiting data characteristics is also important. We propose to utilize coreference information to obtain better document-level entity representations, and propose the concept of core roles to adjust the schema structure to alleviate error propagation. Experiments demonstrate that our data exploitation methods significantly improve the performance of existing models on both the role-level and record-level metrics.

Tasks

Reproductions