Chinese Emergency Event Recognition Based on BiGRU-AM Model

Abstract

A Chinese emergency event recognition based on bidirectional gated recurrent unit BiGRU-AM model with attention mechanism is proposed to resolve the limitation of traditional event recognition methods and the poor interpretability of general recurrent neural networks in respect of information features with different degrees of importance. Firstly, text corpus was trained to generate word vectors, and contextual information features were extracted through BiGRU, and then attention mechanism was introduced into BiGRU network to make feature extraction more selective. Finally, the learned features were activated by softmax function to output recognition results. Simulation results show that this method improves the accuracy and recall rate of emergency recognition, and the F value is superior to other methods.