Att-ChemdNER

---

This repo contains the source code and dataset for the following paper:

• Ling Luo, Zhihao Yang, Pei Yang, Yin Zhang, Lei Wang, Hongfei Lin, Jian Wang. An Attention-based BiLSTM-CRF Approach to Document-level Chemical Named Entity Recognition. Bioinformatics, 2017.

Dependency package

Att-ChemdNER uses the following dependencies:

• Python 2.7
• Theano 0.9.0
• numpy 1.12.1

Content

• data
  • CHENDNER corpus
  • CDR corpus
• models
  • The basic BiLSTM-CRF model
  • The Att-BiLSTM-CRF model
  • The 50-dimensional word embedding
• src
  • backend
  • evaluation: evaluate result of NER task
  • activations.py: activation functions
  • initializations.py
  • loader.py: load the data set
  • model.py: build the model
  • nn.py: the layers of the network architecture
  • optimizaiton.py: optimization method
  • utils.py
  • train.py: train a basic BiLSTM-CRF model
  • AttenTrain.py: train a Att-BiLSTM-CRF model
  • tagger.py: tag the document using the BiLSTM-CRF model
  • AttenTrain.py: tag the document using the Att-BiLSTM-CRF model

Train a basic BiLSTM-CRF model

To train a basic BiLSTM-CRF model, you need to provide the file of the training set, development set,testing set and word embedding model, and run the train.py script:

python train.py --train trainfile --dev devfile --test testfile --pre_emb word_embedding.model 

Train a Att-BiLSTM-CRF model

To train our Att-BiLSTM-CRF model, you need to provide the file of the training set, development set,testing set and word embedding model, and run the AttenTrain.py script:

python AttenTrain.py --train trainfile --dev devfile --test testfile --pre_emb word_embedding.model 

Tag the documents using the BiLSTM-CRF model

Recognize the chemical entities from the documents using the pretrained BiLSTM-CRF model, and you need to provide the pretrained model, inputfile and outputfile:

python tagger.py --model BiLSTM-CRF.model --input inputfile --output outputfile

The inputfile should contain one document by line, and they have to be tokenized.

Tag the documents using the Att-BiLSTM-CRF model

Recognize the chemical entities from the documents using the pretrained Att-BiLSTM-CRF model, and you need to provide the pretrained model, inputfile and outputfile:

python Atten_tagger.py --model Att-BiLSTM-CRF.model --input inputfile --output outputfile

The inputfile should contain one document by line, and they have to be tokenized.

---