chengsen / PyTorch_TextGCN

The PyTorch 1.6 and Python 3.7 implementation for the paper Graph Convolutional Networks for Text Classification
https://arxiv.org/abs/1809.05679
107 stars 21 forks source link
deep-learning gnn-model graph graph-neural-networks text-classification

Graph Convolutional Networks for Text Classification in PyTorch

PyTorch 1.6 and Python 3.7 implementation of Graph Convolutional Networks for Text Classification [1].

Tested on the 20NG/R8/R52/Ohsumed/MR data set, the code on this repository can achieve the effect of the paper.

Benchmark

dataset 20NG R8 R52 Ohsumed MR
TextGCN(official) 0.8634 0.9707 0.9356 0.6836 0.7674
This repo. 0.8618 0.9704 0.9354 0.6827 0.7643

NOTE: The result of the experiment is to repeat the run 10 times, and then take the average of accuracy.

Requirements

Usage

  1. Process the data first, run data_processor.py (Already done)
  2. Generate graph, run build_graph.py (Already done)
  3. Training model, run trainer.py

References

[1] Yao, L. , Mao, C. , & Luo, Y. . (2018). Graph convolutional networks for text classification.