Skip to content

Latest commit

 

History

History
43 lines (40 loc) · 2.1 KB

README.md

File metadata and controls

43 lines (40 loc) · 2.1 KB
Raw

Every Document Owns Its Structure: Inductive Text Classification via Graph Neural Networks in Pytorch

The code and dataset for ACL2020 paper Every Document Owns Its Structure: Inductive Text Classification via Graph Neural Networks,implemented in Pytorch. Some functions are based on TextGCN.Thank for their work. The original code implemented in Tensorflow is TextING.Thank for their work too.

Comprehensions

Every Document Owns Its Structure: Inductive Text Classification via Graph Neural Networks论文理解

Requirements

  • Python 3.6+
  • Pytorch 1.7.1(other versions may also work~)
  • Scipy 1.5.1

Usage

Download pre-trained word embeddings glove.6B.300d.txt from here and unzip to the repository. Build graphs from the datasets in data/corpus/ as:

python build_graph.py [DATASET] [WINSIZE]

Examples:

python build_graph.py R8

Provided datasets include mr,ohsumed,R8andR52. The default sliding window size is 3. To use your own dataset, put the text file under data/corpus/ and the label file under data/ as other datasets do. Preprocess the text by running remove_words.py before building the graphs. Start training and inference as:

python train.py [--dataset DATASET] [--learning_rate LR]
                [--epochs EPOCHS] [--batch_size BATCHSIZE]
                [--hidden HIDDEN] [--steps STEPS]
                [--dropout DROPOUT] [--weight_decay WD]

Examples:

python train.py --dataset R8

To reproduce the result, large hidden size and batch size are suggested as long as your memory allows. We report our result based on 96 hidden size with 1 batch. For the sake of memory efficiency, you may change according to your hardware. Program uses cpu by default.

Reproduced Results

MR R8 R52 Ohsumed
- - 96.89 (98.04 in report) - -

Thank TextING again~