Model1:Character_embedding-Based BiLSTM-CRF.
Model2:On the basis of model1. Extract n_gram feature from word embedding as auxiliary feature, using Conv1D.
Ps:
Run preprocess.py and utils.py firstly, to get processed train/dev/test data and pre-trained char/word embedding matrix.
The file 'appendix_···.py' is writed by means of 'BosonNLP'.
References:
1.End-to-end Sequence Labeling via Bi-directional LSTM-CNNs-CRF. Xuezhe Ma, Eduard Hovy
2.Bidirectional LSTM-CRF Models for Sequence Tagging. Zhiheng Huang, Wei Xu, Kai Yu.
3.Neural Architectures for Named Entity Recognition. Guillaume Lample et al.