Skip to content

jiangpinglei/BERT_ChineseWordSegment

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

BERT_ChineseWordSegment

Try to implement a Chinese word segment work based on Google BERT!

The corpus is extracted from The People's Daily (Chinese: 人民日报, Renmin Ribao).


First git clone https://github.com/google-research/bert.git

Second put the three scripts: modeling.py、optimization.py、tokenization.py into this project, structure is as follows:

BERT_ChinesewordSegment

    |____ PEOPLEdata
    |____ output
    |____ modeling.py
    |____ optimization.py
    |____ tokenization.py
    |____ run_cut.py
    |____ evaluation.py

Third download the Chinese pre-trained bert model BERT-Base, Chinese

And then set pre-trained model path and data path environment: $BERT_CHINESE_DIR、$PEOPLEcut

run

python3 run_cut.py   --task_name="people"   --do_train=True   --do_predict=True  --data_dir=$PEOPLEcut    --vocab_file=$BERT_CHINESE_DIR/vocab.txt   --bert_config_file=$BERT_CHINESE_DIR/bert_config.json   --init_checkpoint=$BERT_CHINESE_DIR/bert_model.ckpt    --max_seq_length=128    --train_batch_size=32    --learning_rate=2e-5   --num_train_epochs=3.0    --output_dir=./output/result_cut/

It will take about 28 minutes with 3 epochs on a GPU.

This will produce an evaluate output like this:

INFO:tensorflow:***** Eval results *****
INFO:tensorflow:  count = 9925
INFO:tensorflow:  precision_avg = 0.9794
INFO:tensorflow:  recall_avg = 0.9780
INFO:tensorflow:  f1_avg = 0.9783
INFO:tensorflow:  error_avg = 0.0213

And the word segmentation results will be seen in ./output/result_cut/seg_result.txt

If you want learn more details, see the code analysis(in Chinese)简书:BERT系列(五)——中文分词实践...

About

A Chinese word segment model based on BERT, F1-Score 97%

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages