Message Passing Attention Networks for Document Understanding

Code for the paper Message Passing Attention Networks for Document Understanding.

Requirements

Code is written in Python 3.6 and requires:

• PyTorch 1.1
• gensim 3.8
• scikit-learn 0.21

Word embeddings

Download and unzip the pre-trained word2vec vectors from the following link: https://code.google.com/p/word2vec/

Run the model

For the simple model, run:

python mpad/main.py --path-to-embeddings path

where path points to the word2vec binary file (i.e., GoogleNews-vectors-negative300.bin file).

For the hierarchical models, run:

python hierarchical_mpad/main.py --path-to-embeddings path --graph-of-sentences type

where type can take the values 'clique', 'path' or 'sentence_att', and each value corresponds to one of the three hierarchical models described in the paper.

Cite

Please cite our paper if you use this code:

@inproceedings{nikolentzos2020message,
  title={Message Passing Attention Networks for Document Understanding},
  author={Nikolentzos, Giannis and Tixier, Antoine Jean-Pierre and Vazirgiannis, Michalis},
  booktitle={Proceedings of the 34th AAAI Conference on Artificial Intelligence},
  pages={8544--8551},
  year={2020}
}

---

Provided for academic use only