bert-event-extraction

Pytorch Solution of Event Extraction Task using BERT on ACE 2005 corpus

Prerequisites

1. Prepare ACE 2005 dataset.

2. Use nlpcl-lab/ace2005-preprocessing to preprocess ACE 2005 dataset in the same format as the data/sample.json. Then place it in the data directory as follows:

   ├── data
   │     └── test.json
   │     └── dev.json
   │     └── train.json
   │...
   

3. Install the packages.

   pip install pytorch==1.0 pytorch_pretrained_bert==0.6.1 numpy
   

Usage

Train

python train.py

Evaluation

python eval.py --model_path=latest_model.pt

Result

Performance

Method	Trigger Classification (%)			Argument Classification (%)
	Precision	Recall	F1	Precision	Recall	F1
JRNN	66.0	73.0	69.3	54.2	56.7	55.5
JMEE	76.3	71.3	73.7	66.8	54.9	60.3
This model (BERT base)	63.4	71.1	67.7	48.5	34.1	40.0

The performance of this model is low in argument classification even though pretrained BERT model was used. The model is currently being updated to improve the performance.

Reference

• Jointly Multiple Events Extraction via Attention-based Graph Information Aggregation (EMNLP 2018), Liu et al. [paper]
• lx865712528's EMNLP2018-JMEE repository [github]
• Kyubyong's bert_ner repository [github]