The word2vec BiLSTM-CRF model for CCKS2019 Chinese clinical named entity recognition.

• python 3.6
• gensim 3.4.0
• jieba 0.39
• keras 2.2.4
• keras_contrib 2.0.8
• numpy 1.16.4
• pandas 0.24.2

The dataset is provided by the CCKS2019.

• ./data/
  • original_data/
    • tagged_data/
    • untagged_data/
  • processed_data/
    • test_data.txt
    • train_data.txt
    • untagged_test_data.txt

• In the "original_data" directory:
  • Each data file in the "tagged_data" should be in the following format:
    • Each line is a JSON object, with "originalText" and "entities" as JSON keys;
    • The JSON value of "entities" is a list of JSON object, and each JSON object represents an entity with "entity_name", "start_pos", "end_pos", "label_type", "overlap" as its JSON keys;
  • Each data file in the "untagged_data" should be in the following format:
    • Each line is a JSON object, with "originalText" as the JSON key;
• In the "processed_data" directory: "train_data.txt" and "test_data.txt" should be in the following format:

患 O
者 O
罹 O
患 O
胃 B-疾病和诊断
癌 I-疾病和诊断

每 O
个 O
例 O
子 O
空 O
行 O
分 O
隔 O

• "untagged_test_data.txt" should be in the following format:

患
者
罹
患
胃
癌

每
个
例
子
空
行
分
隔

• Please download the dataset from CCKS2019 by yourself.
• Put the tagged data under the directory "/data/original_data/tagged_data/".
• Put the untagged data under the directory "/data/original_data/untagged_data/".

python preprocess --tagged True
python preprocess --tagged False

python main.py --mode train

python main.py --mode test

The prediction, standard results and evaluation would be saved as "test_results.json", "true_results.json" and "eval_results.csv", respectively.

python main.py --mode predict

The prediction would be saved as "pred_results.json".