FEE: An Event Extraction Model for Flood and Drought Disaster Based on Sequence Labeling

Due to unique climatic characteristics, China suffers from frequent disasters such as typhoons, droughts, and floods almost every year. It is necessary to study the disasters from history. The event extraction model is effective to mine disaster events from historical documents to help people organize historical event information in a more structured way. Common approaches transform the event extraction task into a sequence labeling task by modeling words with a static vector regardless of polysemy, which affects the performance of event extraction. We propose an event extraction model named FEE. It introduces dynamic word vectors to the BiLSTM-CRF mechanism, which solves the problem that existing models do not consider the ambiguity of words, and makes the result of event extraction more accurate. Experimental results on flood and drought disaster dataset and general dataset show that our model outperforms traditional models on precision, recall, and F
1
.