Sequence Tagger for Named Entity Recognition (TensorFlow)

This tagger is based on tensorflow and fastText. You need to prepare training, testing data and char, word embedding by yourself. Now, the model is only support for binary classification i.e., TARGET and OTHER but you can easily modify the code to extend the model to multiclass version.

Install requirements

pip install -r requirements.txt

Data format

The data format we used in our model is as follows. The dataset is a list of sentences which contain (segmented word, label) pairs.

TRAINING_DATA = [
   [('吳卓源', 1), ('的', 0), ('星海', 0), ('完整', 0), ('的', 0), ('陳述', 0), ...],
   [('五月天', 1), ('即將', 0), ('在', 0), ('小巨蛋', 0), ('舉辦', 0), ...],
   ...
]

Train

python train.py

Tensorboard monitor

tensorboard --logdir=logs --host=<host> --port=<port>

Reference

• Named Entity Recognition (LSTM + CRF) - Tensorflow

• 基于tensorflow深度学习的中文的命名实体识别