Transition-based Event Extraction

Extracting Entities and Events as a Single Task Using a Transition-Based Neural Model, code for IJCAI 2019 paper.

Dependencies:

• DyNET 2.1
• Pyyaml 5.1
• gensim 3.7.3
• Pytorch 1.1.0
• pytorch-transformer 1.2.0
• flair 0.4.3

Dataset:

ACE2005 https://catalog.ldc.upenn.edu/ldc2006t06

We can not provide full ACE2005 data files due to LDC license, instead, sample JSON files are given in data_files/samples/ for reference.

To facilitate follow up research, we list documents split of train/dev/test in data_files/doc_split/ which is provided by Thien Huu Nguyen et al [2016]

Configurations

• data_config.yaml (for locating file paths)
• joint_config.yaml (for parameters tunning)

Preprocess:

• Put glove.6B.100d.txt in data_files/glove_emb/

• Then make vocabulary and pickle instances by:

python preprocess.py

• (Optional) Generate BERT Embeddings (bert-base-uncased):

python gen_bert_emb.py

Note that if you don`t use BERT, set use_sentence_vec to false in joint_config.yaml.

It performs poorly for argument roles without BERT embeddings (around 45% F-scores on test set). If you don`t want use BERT, consider incorporating dependency features as in Dependency-Bridge (Sha et al [2018]).

Train & Evaluate:

python train.py

Reference:

• Thien Huu Nguyen, Kyunghyun Cho, and Ralph Grishman. Joint Event Extraction via Recurrent Neural Networks, NAACL, 2016
• Lei Sha, Feng Qian, Baobao Chang, and Zhifang Sui. Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction, AAAI, 2018

Citation

@inproceedings{ijcai2019-753,
  title     = {Extracting Entities and Events as a Single Task Using a Transition-Based Neural Model},
  author    = {Zhang, Junchi and Qin, Yanxia and Zhang, Yue and Liu, Mengchi and Ji, Donghong},
  booktitle = {Proceedings of the Twenty-Eighth International Joint Conference on
               Artificial Intelligence, {IJCAI-19}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},             
  pages     = {5422--5428},
  year      = {2019},
  month     = {7},
  doi       = {10.24963/ijcai.2019/753},
  url       = {https://doi.org/10.24963/ijcai.2019/753},
}