ner

Named Entity Recognition

We employ the birnn plus CRF architecture as Lample et al. 2016, and experiment on CoNLL-2003 English NER data. Main experimental results are summarized below.

Model	#Params	NER
Lample et al. 2016	-	90.94
LSTM	245K	89.61
GRU	192K	89.35
ATR	87K	88.46
SRU	161K	88.89
LRN	129K	88.56

F1-score.

Requirement

see requirements.txt for full list.

How to Run?

• download and preprocess dataset

  • download the conll2003 dataset from anago (in data folder).
  • download the Glove-6B-100d pre-trained word embedding from: http://nlp.stanford.edu/data/glove.6B.zip

• no hyperparameters are tuned, we keep them all in default.

• training and evaluation

  the running procedure is as follows:

  export CUDA_ROOT=XXX
  export PATH=$CUDA_ROOT/bin:$PATH
  export LD_LIBRARY_PATH=$CUDA_ROOT/lib64:$LD_LIBRARY_PATH
  
  export CUDA_VISIBLE_DEVICES=0
  
  export data_dir=path-of/conll2003/en/ner
  export glove_dir=path-of/glove.6B/glove.6B.100d.txt
  
  RUN_EXP=5
  rnn=lrn
  
  for i in $(seq 1 $RUN_EXP); do 
      exp_dir=exp$i
      mkdir $exp_dir
      cd $exp_dir
  
      export cell_type=$rnn
      python3 ner_glove.py --cell lrn >& log.lrn
  
      cd ../
  done
  
  python scripts/get_test_score.py $rnn exp* >& score.$rnn
  

  Results are reported over 5 runs.

Credits

Source code structure is adapted from annago.