GENE

Code for TextGraphs 2021 paper "GENE: Global Event Network Embedding"

@inproceedings{zeng-etal-2021-gene,
    title = "GENE: Global Event Network Embedding",
    author = "Zeng, Qi  and
      Li, Manling  and
      Lai, Tuan  and
      Ji, Heng  and
      Bansal, Mohit  and
      Tong, Hanghang",
    booktitle = "Proceedings of the Graph-based Methods for Natural Language Processing (TextGraphs)",
    month = Jun,
    year = "2021",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
}

Requirements

numpy
torch
dgl
sklearn
allennlp

Data

The current data directory only includes the sample data. ACE05 Datset requires LDC License (Access from LDC and preprocessing following OneIE.

You may contact qizeng2@illinois.edu for the preprocessed (enhanced) data.

Train

CUDA_VISIBLE_DEVICES=0 python main.py --mode 'train' --version 'test' --model_base 'SEM_ARC' 

• mode: train, infer, eval
• model_base: SEM_ARC, SEM, ARC, SKG, DGI
• version: name for this model

Check args.py for more tunable hyperparameters.

Eval

The evaluation code of Event Coreference can be found in event-coref folder with a separate README.

The evaluation for Node Typing and Argument Role Classification can be run with:

CUDA_VISIBLE_DEVICES=0 python main_hetero.py --mode 'eval' --version 'test' --load_emb 'SEM_ARC.test'

• load_emb: [MODELBASE].[VERSION]