chinese-word-segmentation

Simple chinese word segmentation with experiments on the PKU datatset

Methods

• Pattern based word segmentation
• CRF ++ tagging
• LSTM tagging

Performance

F1

• Pattern Based Segmentation: 0.87
• CRF++ Tagging: 0.93
• LSTM Tagging: 0.86

It seems that the simple LSTM tagger doesn't perform better than CRF++ or even pattern based segmentation.

Tips for improve the performance of the LSTM tagger on the segementation task

• Add Tag Inference as stated in the paper Long Short-Term Memory Neural Networks for Chinese Word Segmentation