This is the source code for the paper, Chinese Word Segmentation Using Temporal Convolutional Networks. It achieves competitive performance to state-of-the-art Bi-LSTM model.

Python: 2.7   
tensorflow-gpu: 1.8.0
numpy: 1.15.4

CoNLL format (prefer BMES tag scheme, compatible with the Chinese word segmentation scheme), with each character its label for one line. Sentences are splited with a null line.

中	B
国	E
财	B
团	E
买	B
下	E
AC	B
米	M
兰	E

代	B
价	E
超	S
10	B
亿	E
欧	B
元	E

1. Training the model on NLPCC2016.

python main.py -o train

2. Training the Baseline model(Bi-LSTM) on NLPCC2016.

python main.py -m bilstm -o train

3. Evaluate the model on NLPCC216

python main.py -o evaluate --model_path saved/tcn/20190303_114309/

4. Evaluate the Baseline model on NLPCC2016.

python main.py -m bilstm -o evaluate --model_path saved/bilstm/20190302_225753/

5. Train the model on the other dataset.

python main.py -o train --train 'train set path'

6. Evaluate the model on the other dataset.

python main.py -o evaluate --test 'test set path' --model_path 'saved model parameters directory path'