Chinese Word Segmentation Model Based On Improved Bidirectional LSTM-CRF

Natural language,as the most important contact tool,is the primary way for communication by humans.Natural language processing(NLP)is the is the method and technology of processing language information by computer.Chinese word segmentation(CWS)is the basic of Chinese natural language processing.The quality of word segmentation will directly affect the rest of NLP tasks.At present,the most common solution is the classical machine learning model based on sequence labeling.Recently,with the artificial intelligence tide rising again,Bidirectional LSTM-CRF neural network model used bidirectional LSTM layer for capturing the forward and backward context and chain conditional random field layer for output label decoding prediction.It has been widely utilized in various kinds of NLP tasks,and functions well.As for CWS task,we propose a contribution rate to balance the matrix's value in forward LSTM layer and backward LSTM layer based on classical Bi-LSTM Model.It not only improves the long-term dependence,but also modeling based on the difference between forward and backward context.Bidirectional LSTM-CRF neural network model used bidirectional LSTM layer for capturing the forward and backward context and chain conditional random field layer for output label decoding prediction.It has been widely utilized in various kinds of NLP tasks,and functions well.we proposed a Bidirectional LSTMN-CRF neural network model in this paper,replaced LSTM unit with attention-based LSTMN unit of Bi-LSTM-CRF model.New model solves the memory compression problem during the decoding period.