yzp-tk / text_gcn

Graph Convolutional Networks for Text Classification. AAAI 2019

text_gcn

The implementation of Text GCN in our paper:

Liang Yao, Chengsheng Mao, Yuan Luo. "Graph Convolutional Networks for Text Classification." In 33rd AAAI Conference on Artificial Intelligence (AAAI-19), 7370-7377

Require

Python 2.7 or 3.6

Tensorflow >= 1.4.0

Reproducing Results

1. Run python remove_words.py 20ng

2. Run python build_graph.py 20ng

3. Run python train.py 20ng

4. Change 20ng in above 3 command lines to R8, R52, ohsumed and mr when producing results for other datasets.

Example input data

1. /data/20ng.txt indicates document names, training/test split, document labels. Each line is for a document.

2. /data/corpus/20ng.txt contains raw text of each document, each line is for the corresponding line in /data/20ng.txt

3. prepare_data.py is an example for preparing your own data, note that '\n' is removed in your documents or sentences.

Inductive version

An inductive version of Text GCN is fast_text_gcn, where test documents are not included in training process.