WGCN

The implementation of Word Grounded Graph Convolutional Network.

Authors: Zhibin Lu ([email protected]), Qianqian Xie ([email protected]), Benyou Wang ([email protected]), Jian-Yun Nie ([email protected])

Overview

This is the implementation of Word Grounded Graph Convolutional Network. If you make use of this code or the WGCN or WGraph approach in your work, please cite the following paper:

 @inproceedings{ZhibinluWGCN,
     author    = {Zhibin Lu and Qianqian Xie and Benyou Wang and Jian-Yun Nie},
     title     = {Word Grounded Graph Convolutional Network},
     publisher = {arXiv},
     year      = {2023},
  }

Requirements

• Python 3.7.2
• PyTorch 1.0
• scikit-learn 0.20.1
• scipy 1.1.0
• numpy 1.15.4
• glove.6B.300d.txt (copy to data/ dir)

Datasets

1. Demo dataset is mr, in data/ dir.

Pre-processing

1. Run python build_graph.py mr

Trainning

1. For Original GCN, Text GCN, run python train_tgcn.py
2. For MLP, run python train_mlp.py
3. For WGCN, run python train_wgcn.py
4. For WGCN using a Vocabuary embedding, run python train_wgcn_vocab_embedding.py (download glove.6B.300d.txt and copy to data/ dir)
5. For WGCN using a word embedding X, run python train_wgcn_word_embedding.py (download glove.6B.300d.txt and copy to data/ dir)