A Joint Tagging Event Extraction with Attention Mechanism

Abstract

Event extraction can extract event information from text, which is a very important information extraction task. Currently, most methods assume that there is at most one event in a sentence. However, in practice there may be one or more events in a sentence. Therefore, there may be information overlap between multiple events. To solve this problem, this paper proposes a novel joint learning framework, JointEE. First, the similarity function is used to measure the types of events present in the sentence, and then the sequence annotation model with the attention mechanism is used to jointly extract trigger and arguments. The paper is evaluated on a public event extraction dataset, FewFC. Experiments show that, compared with previous methods, JointEE achieves good results on the overlapping event extraction problem.