HINT

Implementation for Hierarchical information matters: Text classification via tree based graph neural network

Usage

Build dependency graph for dataset in data/corpus/ as:

python build_dependency_graph.py <dataset>

Provided datasets include mr,ohsumed,R8andR52.

Build coding tree for dataset as:

python build_coding_tree.py -d <dataset> -k <tree_deepth> -o <onehot> -a <add> -s <stop>
example: python build_coding_tree.py -d mr -k 2 -o True -a False -s False

Start training and inference as:

 python main.py [--dataset DATASET] [--tree_deepth DEEPTH]
                [--epochs EPOCHS] [--batch_size BATCHSIZE]
                [--hidden_dim HIDDEN_DIM] [--learning_rate LEARNING_RATE]
                [--final_dropout DROPOUT] [--input_dim INPUT_DIM]
                [--num_mlp_layers MLP_LAYERS] [--l2rate L2RATE]
                [--tree_pooling_type TREE_POOLING] [--mode MODE]
                [--position_embedding PE] 
example: python main.py -d mr -k 2 -b 4 -md dependency -pe onehot

Citation

@inproceedings{zhang-etal-2022-hierarchical,
title = "Hierarchical Information Matters: Text Classification via Tree Based Graph Neural Network",
author = "Zhang, Chong  and
  Zhu, He  and
  Peng, Xingyu  and
  Wu, Junran  and
  Xu, Ke",
booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
month = oct,
year = "2022",
address = "Gyeongju, Republic of Korea",
publisher = "International Committee on Computational Linguistics",
url = "https://aclanthology.org/2022.coling-1.79",
pages = "950--959"
}