Discontinuous NER with Pointer Network
Codes for the AAAI 2021 paper: Rethinking Boundaries: End-To-End Recognition of Discontinuous Mentions with Pointer Networks.
Requirement
python>=1.6
numpy>=1.13.3
torch>=0.4.0
Datasets
Two benchmark datasets for discontinuous NER.
Download them and put at ./data
folds.
- CADEC
- ShARe13
Data format preprocessing. Please process the annotation as following format:
Upset stomach and the feeling that I may need to throw up .
0,1 ADR|10,11 ADR
See the example data in ./data/examples.
Experiments
To train the parser, run the following script:
python ./framework/main.py
Change the parameters for training, testing.
Citation
@inproceedings{FeiDisNERAAAI21,
author = {Hao Fei and Donghong Ji and Bobo Li and
Yijiang Liu and Yafeng Ren and Fei Li},
title = {Rethinking Boundaries: End-To-End Recognition of Discontinuous Mentions with Pointer Networks},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
pages = {12785--12793},
year = {2021},
}
License
The code is released under Apache License 2.0 for Non-commercial use only.