Uyghur Language Recognition Method based on BIGRU_IDCNN_ATT_CRF

Abstract

Named entity recognition plays a very important role in the field of natural language processing. Aiming at the special semantic morphology and scarcity of data in Uyghur named entity recognition, a neural network model based on BIGRU_IDCNN_ATT_CRF is proposed. First, extract the long-dependent semantic information of the Uyghur language context through the bidirectional gated recurrent neural network (BIGRU), and then uses the word vector through iterated dilated convolutional neural network (IDCNN) to increase the perception field to reduce the number of neurons and training parameters. Then use the self-attention mechanism to weight the features extracted from BIGRU_IDCNN to strengthen key features and weaken useless features. Finally, Conditional Random Field (CRF) is used for label prediction. It is concluded through experiments that the accuracy, recall and F1 value of this model on the Uyghur language data set are 85.0%, 84.3% and 84.58%, respectively, which can significantly improve the Uyghur language recognition task compared with the existing models.