Event Extraction with Deep Contextualized Word Representation and Multi-attention Layer

Abstract

One common application of text mining is event extraction. The purpose of an event extraction task is to identify event triggers of a certain event type in the text and to find related arguments. In recent years, the technology to automatically extract events from text has drawn researchers’ attention. However, the existing works including feature based systems and neural network base models don’t capture the contextual information well. Besides, it is still difficult to extract deep semantic relations when finding related arguments for events. To address these issues, we propose a novel model for event extraction using multi-attention layers and deep contextualized word representation. Furthermore, we put forward an attention function suitable for event extraction tasks. Experimental results show that our model outperforms the state-of-the-art models on ACE2005.