Chinese Named Entity Recognition Based on CNN-BiLSTM-CRF

Named Entity Recognition (NER) is an important basic task in natural language processing (NLP). In recent years, the method of word representations enhancement by character embedding has significantly enhanced the effect of entity recognition. However, this kind of character embedding method only works on alphabetic spelling words such as English, and the same method is not suitable for Chinese. Aiming at the inherent characteristics of Chinese as morpheme writing, we propose a novel neural network model based on CNN-BiLSTM-CRF in this paper. Convolution neural network (CNN) extracts the glyph embeddings with morphological features from each Chinese character, which are concatenated with the character embeddings with semantic feature information and fed to the BiLSTM-CRF network. We evaluate our model on the third SIGHAN Bakeoff MSRA dataset for simplified Chinese NER task. The experimental results show that our model reaches 91.09% in F-scores which does not rely on the hand-designed features and domain knowledge.