hConv-gPool-Net

TensorFlow implementation of Learning Graph Pooling and Hybrid Convolutional Operations for Text Representations (WWW19)

Created by Hongyang Gao at Texas A&M University, Yongjun Chen at Washington State University, and Shuiwang Ji at Texas A&M University.

Introduction

We propose novel graph pooling layer and hybrid convolutional layer for text representation learning. It has been accepted in WWW19.

Detailed information about hConv-gPool-Net is provided in https://arxiv.org/abs/1901.06965.

Citation

@article{gao2019learning,
  title={Learning Graph Pooling and Hybrid Convolutional Operations for Text Representations},
  author={Gao, Hongyang and Chen, Yongjun and Ji, Shuiwang},
  journal={arXiv preprint arXiv:1901.06965},
  year={2019}
}

Data

Since some files are quite large, we use LFS to deal with them. After cloning, please run

"git lfs pull"

to pull large data. You may need to install git LFS.

Results

Results of text classification experiments in terms of classification error rate on the AG’s News, DBPedia, and Yelp Review Polarity datasets. The first two methods are the state-of-the-art models without using any unsupervised data. The last four networks are proposed in this work.

Configure the network

All network hyperparameters are configured in main.py.