Chinese Name Entity Recognition

Use Bi-LSTM and CRF

Requirements

• keras
• keras_contrib
• numpy
• sklearn
• sklearn_crfsuite

Usage

Prepare data

• word2vec: https://github.com/hankcs/HanLP/wiki/word2vec
• train_data and test_data: in data folder, also can download from http://www.pudn.com/Download/item/id/2435241.html
• modify setttings.py for your file path

Process data

python main.py -m preprocess

Train data

python main.py -m train

Evaluate

python main.py -m eval

Show NER demo

In this mode, you can input one sentence from command line, then the program ouput all entities.

python main.py -m demo

Result

	precision	recall	f1-score
B-LOC	0.935	0.896	0.915
I-LOC	0.929	0.878	0.903
B-ORG	0.870	0.834	0.852
I-ORG	0.905	0.887	0.896
B-PER	0.917	0.791	0.849
I-PER	0.898	0.842	0.869
avg	0.912	0.865	0.887