SegmentationCRF

Chinese word segmentation with CRF++.

Running Environment

• Windows 10
• Python 3.6
• CRF++ 0.58

If running in a different environment, this program may not work properly.

For instance, considering the difference of newline between Windows and Linux, you need to modify Line 32 of segment.py.

Usage

1. Prepare the correct environment mentioned above.
2. Run python prepare.py.
3. Prepare your own template for CRF++ or use template.utf8 instead.
4. Run CRF++-0.58/crf_learn.exe (For Windows only) to train your own model or use the models in CRF_Model/.
5. Run python segment.py to segment.

Test Result

Using the two models in CRF_Model/, the F1 Scores of testing files are listed as follows:

	crf_model_pku	crf_model_both
pku_test	0.931	0.880
msr_test	0.857	0.936

Conclusion

Maybe adjusting the template and training parameters can make the result better, but as it takes too much time to do it, I just stop here.

Anyway, this is just the beginning of many to explore.