Context-Aware Heterogeneous Graph Interactive Learning for Document-Level Event Extraction

doi:10.1007/978-981-95-3456-2_6

Context-Aware Heterogeneous Graph Interactive Learning for Document-Level Event Extraction

2026-01-01

0

PRE

AI

DOI：10.1007/978-981-95-3456-2_6

Original

Original request for help

Save

Abstract

En 中文

Document-level event extraction focuses on extracting structured event records from unstructured documents. Previous approaches exhibit limitations in representing or utilizing the context and entity information due to the complex semantic relationships in documents, leading to accuracy decline in event extraction. In this paper, we propose a lightweight end-to-end framework to address this deficiency via heterogeneous graph interaction and common context-aware fusion. Firstly, aiming to capture the complex semantic relationships within documents, we employ a heterogeneous graph convolution network to simulate information interactions between varying granularity. Secondly, during the argument combination extraction stage, we design a strategy that fuses entity information and common context to improve the prediction accuracy of entity correlation and reduce error propagation. Experiments on the public dataset show that our lightweight model outperforms other methods in both single-event and overall evaluations.

Keywords:

event extraction

heterogeneous graph interaction

common context-aware information

entity relevance

Context-Aware Heterogeneous Graph Interactive Learning for Document-Level Event Extraction

Abstract

Journal

Organization