Hierarchical Annotation Event Extraction Method in Multiple Scenarios

In the event extraction task, considering that there may be multiple scenarios in corpus and an argument play different roles under triggers, traditional tagging scheme can only tag each word once, which cannot solve problem of overlap. A hierarchical pipeline model for Chinese based on pretrained Bert was proposed, obtain relevant arguments a way. The structure is selected model, task divided into trigger classification recognition. Firstly, used to generate feature vector transfer it bidirectional gated recurrent unit+conditional random field (BiGRU+CRF) classification; then, marked type features are spliced as known then passed BiGRU+CRF We evaluated our method DUEE, combined with data enhancement mask operation. Experimental results show improved compared other baselines, prove effectiveness corpus.