Home

Awesome

DCA

DCA (Dynamic Context Augmentation) provides global entity linking models featuring:

Remarkablely, our DCA models (trained by supervised learning or reinforcement learning) achieved:

Details about DCA can be accessed at: https://arxiv.org/abs/1909.02117.

This implementation refers to the project structure of mulrel-nel.

Written and maintained by Sheng Lin (shenglin@zju.edu.cn) and Xiyuan Yang (yangxiyuan@zju.edu.cn).

Overall Workflow

Alt Text

Data

Download data from here and unzip to the main folder (i.e. your-path/DCA).

The above data archive mainly contains the following resource files:

Installation

Requirements: Python 3.5 or 3.6, Pytorch 0.3, CUDA 7.5 or 8

Important Parameters

mode: train or eval mode.

method: training method, Supervised Learning (SL) or Reinforcement Learning (RL)

order: three decision orders -- offset / size / random. Please refer to our paper for their concrete definition.

n_cands_before_rank: the number of candidates.

tok_top_n4inlink: the number of inlinks for a Wikipedia page (entity) would be considered as candidates for the dynamic context.

tok_top_n4ent: the number of inlinks for a Wikipedia page (entity) would be added into the dynamic context.

isDynamic: 2-hop DCA / 1-hop DCA / without DCA. Corresponding to the experiments of Table 4 in our paper.

dca_method: soft+hard attention / soft attention / average sum. Corresponding to the experiments of Table 5 in our paper.

Running

cd DCA/

export PYTHONPATH=$PYTHONPATH:../

Supervised Learning: python main.py --mode train --order offset --model_path model --method SL

Reinforcement Learning: python main.py --mode train --order offset --model_path model --method RL

Citation

If you find the implementation useful, please cite the following paper: Learning Dynamic Context Augmentation for Global Entity Linking.

@inproceedings{yang2019learning,
  title={Learning Dynamic Context Augmentation for Global Entity Linking},
  author={Yang, Xiyuan and Gu, Xiaotao and Lin, Sheng and Tang, Siliang and Zhuang, Yueting and Wu, Fei and Chen, Zhigang and Hu, Guoping and Ren, Xiang},
  booktitle = {Proceedings of EMNLP-IJCNLP},
  year={2019}
}