Chinese word segmentation model based on BI_GRU_AT_HN_CRF_6

Abstract

Chinese word segmentation is an indispensable step in natural language processing, and it is also the most important step. At present, the use of recurrent neural network to Chinese word segmentation model has become a new trend. The researchers proposed various models based on the LSTM network model of long and short memory and the word segmentation method based on the GRU network model. Both LSTM and GRU are a type of circulatory neural network that inherits the ability to automatically learn and long-short term memory characteristics. However, in the process of Chinese word segmentation, as the length of the sentence becomes longer, the inter-dependent feature distance in the context becomes farther, resulting in the loss of the historical feature information and future feature information that the given sentence depends on, thereby reducing the accuracy of word segmentation. In order to solve this problem, this paper introduces the attention mechanism and proposes the BI_GRU_AT_HW_CRF_6 neural network segmentation model. Experiments show that with the introduction of attentional mechanism, with the change of sentence, there is a better performance in accuracy, training, and forecasting data speed.