Details

Autor(en) / Beteiligte

• Liu, Yang
• Jiang, Xiaoxia
• Zhao, Wenzheng
• Ge, Weiyi
• Hu, Wei

Titel

Dual Graph Convolutional Networks for Document-Level Event Causality Identification

Ist Teil von

• Web and Big Data, p.114-128

Ort / Verlag

Cham: Springer Nature Switzerland

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Event causalities organize events into a graph according to causal logics, which assists humans in decision making by causal reasoning among events. Despite many efforts to identify event causalities, most of them assume that only one causality exists in a sentence or causalities only occur in adjacent sentences, leading to the incapability of detecting multiple causalities or document-level causalities. In this paper, we propose a novel model for document-level event causality identification named DocECI. We define two heterogeneous document graphs, namely text structure graph and mention relation graph, and encode them with relational graph convolutional networks, which gradually aggregate the information of multi-granular nodes in a cascade manner and capture the causality patterns. Experiments on a benchmark dataset show that DocECI outperforms existing models by a significant margin. Moreover, a new experiment is conducted on causality direction identification, which is overlooked by existing models.