Graph-Convolutional-Networks-With-Argument-Aware-Pooling-for-Event-Detection

implement model in paper Graph Convolutional Networks With Argument-Aware Pooling for Event Detection

Preprocessing data:

• Transform format data from ACE format to json format from here: https://github.com/nlpcl-lab/ace2005-preprocessing.git
• build vocabularies : rebuild event vocab(BIO format in this project, include vocab[PAD] =-100), entity vocab( BIO format include pad label, id=0), word vocab( include pad token with id=0, unknow token with id =1) or just use vocabs from data folder
• use load_data_json and window_encode2 functions in utils.py to build data that will be fed into model

Train model:

• hyperparameters of model are stored in Config class
• example for training model have in file model.py( use EDModel2)

References:

Graph Convolutional Networks With Argument-Aware Pooling for Event Detection, Thien Huu Nguyen, Ralph Grishman
Jointly Multiple Events Extraction via Attention-based Graph Information Aggregation, Xiao Liu† and Zhunchen Luo‡ and Heyan Huang†∗