Skip to content

Latest commit

 

History

History
113 lines (85 loc) · 3.38 KB

README.english.md

File metadata and controls

113 lines (85 loc) · 3.38 KB
Raw

Deep Learning Chinese Word Segment (https://github.com/koth/kcws)

引用 

本项目模型BiLSTM+CRF参考论文：http://www.aclweb.org/anthology/N16-1030 ,IDCNN+CRF参考论文：https://arxiv.org/abs/1702.02098

Build

  1. Python 2.6+ (but not Python 3)
  2. 安装好bazel代码构建工具，安装好tensorflow（目前本项目需要tf 1.0.0alpha版本以上)
  3. 切换到本项目代码目录，运行./configure
  4. 编译后台服务
    bazel build //kcws/cc:seg_backend_api

Training

  1. Follow Wechat ID, reply kcws to get corpus download link：

    logo

  2. Unzip to DIR

  3. Run:

python kcws/train/process_anno_file.py <DIR> pre_chars_for_w2v.txt
bazel build third_party/word2vec:word2vec

Obtain inital vocabulary table

./bazel-bin/third_party/word2vec/word2vec -train pre_chars_for_w2v.txt -save-vocab pre_vocab.txt -min-count 3

Process low-frequency words

 python kcws/train/replace_unk.py pre_vocab.txt pre_chars_for_w2v.txt chars_for_w2v.txt

Train word2vec

 ./bazel-bin/third_party/word2vec/word2vec -train chars_for_w2v.txt -output vec.txt -size 50 -sample 1e-4 -negative 5 -hs 1 -binary 0 -iter 5

构建训练语料工具

 bazel build kcws/train:generate_training

生成语料

 ./bazel-bin/kcws/train/generate_training vec.txt <语料目录> all.txt

得到train.txt , test.txt文件

 python kcws/train/filter_sentence.py all.txt
  1. 安装好tensorflow,切换到kcws代码目录，运行:

python kcws/train/train_cws.py --word2vec_path vec.txt --train_data_path <绝对路径到train.txt> --test_data_path test.txt --max_sentence_len 80 --learning_rate 0.001  （默认使用IDCNN模型，可设置参数”--use_idcnn False“来切换BiLSTM模型)



    

  1. 生成vocab
    2. 





  ./bazel-bin/kcws/cc/dump_vocab vec.txt kcws/models/basic_vocab.txt


    

  1. 导出训练好的模型
    2. 



 python tools/freeze_graph.py --input_graph logs/graph.pbtxt  --input_checkpoint logs/model.ckpt --output_node_names  "transitions,Reshape_7"   --output_graph kcws/models/seg_model.pbtxt


    

  1. 

    词性标注模型下载  (临时方案，后续文档给出词性标注模型训练，导出等）
    

    

    从 https://pan.baidu.com/s/1bYmABk 下载pos_model.pbtxt到kcws/models/目录下
    

    

    2. 

  2. 

    运行web service
    

    3. 



bazel build kcws/cc:seg_backend_api


  ./bazel-bin/kcws/cc/seg_backend_api --model_path=kcws/models/seg_model.pbtxt(绝对路径到seg_model.pbtxt>)   --vocab_path=kcws/models/basic_vocab.txt   --max_sentence_len=80


词性标注的训练说明：


https://github.com/koth/kcws/blob/master/pos_train.md


自定义词典


目前支持自定义词典是在解码阶段，参考具体使用方式请参考kcws/cc/test_seg.cc
字典为文本格式，每一行格式如下:




<自定义词条>\t<权重>




比如：




蓝瘦香菇	4




权重为一个正整数，一般4以上，越大越重要


demo


http://45.32.100.248:9090/


附： 使用相同模型训练的公司名识别demo:


http://45.32.100.248:18080


Reference 


Guillaume Lample, Miguel Ballesteros, Kazuya Kawakami, Sandeep Subramanian, and Chris Dyer "Neural architectures for named entity recognition," NAACL-HLT, 2016 (http://www.aclweb.org/anthology/N16-1030)