Awesome

Source code of Boundary-aware Model for Nested NER

Implementation of Our Paper "A Boundary-aware Model for Nested Named Entity Recognition" in EMNLP-IJCNLP 2019.

Requirements

• python 3
• pytorch
• numpy
• gensim
• scikit-learn
• joblib

Data Format

Our processed GENIA dataset is in ./data/genia.

The data format is the same as in Neural Layered Model, Ju et al. 2018 NAACL

Each line has multiple columns separated by a tab key. Each line contains

word	label1	label2	label3	...	labelN

The number of labels (N) for each word is determined by the maximum nested level in the data set. N=maximum nested level + 1 Each sentence is separated by an empty line. For example, for these two sentences, John killed Mary's husband. He was arrested last night , they contain four entities: John (PER), Mary(PER), Mary's husband(PER),He (PER). The format for these two sentences is listed as following:

John    B-PER   O   O
killed  O   O   O
Mary    B-PER   B-PER   O
's  O   I-PER   O
husband O   I-PER   O
.   O   O   O

He    B-PER   O   O
was  O   O   O
arrested  O   O   O
last  O   O   O
night  O   O   O
.  O   O   O

Pre-trained word embeddings

For GENIA Corpus:

• Pre-trained word embeddings used here is the same as in Neural Layered Model

For GermEval 2014 Corpus:

• Pre-trained word embeddings used here is the same as in paper GermEval-2014: Nested Named Entity Recognition with Neural Networks.
• Since the URL above has been invalid, we provide the pre-processed GermEval 2014 Dataset in the file GermEval2014.zip.

Setup

Download pre-trained embedding above, unzip it, and place PubMed-shuffle-win-30.bin into ./data/embedding/ For GermEval 2014 dataset, the same as GENIA.

Usage

Training

python3 train.py

The trained model will be saved at ./data/model/

Testing

set model_url to the url of saved model in training in main() of eval.py

python3 eval.py

Citation

If you find this repo helpful, please cite the following:

@inproceedings{zheng2019boundary,
  title={A Boundary-aware Neural Model for Nested Named Entity Recognition},
  author={Zheng, Changmeng and Cai, Yi and Xu, Jingyun and Leung, Ho-fung and Xu, Guandong},
  booktitle={Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  pages={357--366},
  year={2019}
}