运行环境：python3 TensorFlow 1.2
修改CreateData.py中的input和output（input为文件train.txt的路径）
python CreateData.py
将输出的output文件按90%:10%的比例分成训练集和验证集，改名为train_data和test_data放到data_path文件夹中
python main.py --mode=train
可以对main.py中的超参数修改，如
python main.py --mode=train --epoch=30
main.py中，有两行
elif args.mode == 'demo\':
运行
python main.py --mode=demo --demo_model=1543146557
这里暂时使用注释的方式切换两种使用方式。如想使用后者，请将前者注释，后者取消注释。前者可以直接输入句子进行测试并产生输出，后者可以预测文件并产生输出。在使用后者时记得修改里面的input和output路径
最后将输出的output文件中所有0替换成O
最后结果：F1 Score =0.885836
原项目的地址https://github.com/Determined22/zh-NER-TF
以下是原项目的readme.md
This repository includes the code for buliding a very simple character-based BiLSTM-CRF sequence labelling model for Chinese Named Entity Recognition task. Its goal is to recognize three types of Named Entity: PERSON, LOCATION and ORGANIZATION.
This code works on Python 3 & TensorFlow 1.2 and the following repository https://github.com/guillaumegenthial/sequence_tagging gives me much help.
This model is similar to the models provied by paper [1] and [2]. Its structure looks just like the following illustration:
For one Chinese sentence, each character in this sentence has / will have a tag which belongs to the set {O, B-PER, I-PER, B-LOC, I-LOC, B-ORG, I-ORG}.
The first layer, look-up layer, aims at transforming character representation from one-hot vector into character embedding. In this code I initialize the embedding matrix randomly and I know it looks too simple. We could add some language knowledge later. For example, do tokenization and use pre-trained word-level embedding, then every character in one token could be initialized with this token's word embedding. In addition, we can get the character embedding by combining low-level features (please see paper[2]'s section 4.1 and paper[3]'s section 3.3 for more details).
The second layer, BiLSTM layer, can efficiently use both past and future input information and extract features automatically.
The third layer, CRF layer, labels the tag for each character in one sentence. If we use Softmax layer for labelling we might get ungrammatic tag sequences beacuse Softmax could only label each position independently. We know that 'I-LOC' cannot follow 'B-PER' but Softmax don't know. Compared to Softmax layer, CRF layer could use sentence-level tag information and model the transition behavior of each two different tags.
| #sentence | #PER | #LOC | #ORG | |
|---|---|---|---|---|
| train | 46364 | 17615 | 36517 | 20571 |
| test | 4365 | 1973 | 2877 | 1331 |
It looks like a portion of MSRA corpus.
python main.py --mode=train
python main.py --mode=test --demo_model=1521112368
Please set the parameter --demo_model to the model which you want to test. 1521112368 is the model trained by me.
An official evaluation tool: here (click 'Instructions')
My test performance:
| P | R | F | F (PER) | F (LOC) | F (ORG) |
|---|---|---|---|---|---|
| 0.8945 | 0.8752 | 0.8847 | 0.8688 | 0.9118 | 0.8515 |
python main.py --mode=demo --demo_model=1521112368
You can input one Chinese sentence and the model will return the recognition result:
[1] Bidirectional LSTM-CRF Models for Sequence Tagging
[2] Neural Architectures for Named Entity Recognition
[3] Character-Based LSTM-CRF with Radical-Level Features for Chinese Named Entity Recognition