A Study of Event Multi-triple Extraction Methods Based on Edge-Enhanced Graph Convolution Networks

Yu Zhang, Yimeng Li, Chaomurilige, Yu Weng

Abstract: The event multi-triple extraction task consists of identifying event trigger words and extracting relevant event arguments, and is the basis for some natural language processing downstream tasks. Joint approaches are often considered in existing work, but the coupling and constraints of trigger word recognition and argument recognition pose difficulties for concrete implementation. In this paper, we propose a pipelined approach to event multi-triple extraction, use the event trigger words and event categories generated by EE-GCN as an indication of event extraction, NER as a technical tool to extract event indicator. Based on syntactic dependencies we propose syntactic rules for the selection of arguments, According to this rule, event category, event trigger words and their corresponding arguments are formed into event Multi-triple. The experimental results show that the method is effective in extracting event multi-triple from the text in the ACE2005 dataset.