Chinese Event Temporal Relation Identification via Document-Level Graph
Event temporal relation identification is a challenging subtask of information extraction.Most previous works are focused on identifying sentence-level temporal relation,failing to address document-level relation(i.e.,in-tra-sentence,adjacent-sentence and nonadjacent-sentence relation).To address this issue,we propose a model of e-vent temporal relation identification on document-level graph.It constructs two Graph Convolutional Networks to encode syntactic information and event interaction information,respectively.The experimental results on the Chinese ACE2005-extended dataset show that the proposed model achieves 71.81%in micro-F1 measure,with 1.76%improvement compared with the best baseline.
NETL
NSTL
万方数据
