Skip to content

Latest commit

 

History

History
76 lines (64 loc) · 2.54 KB

README.md

File metadata and controls

76 lines (64 loc) · 2.54 KB

On the Effectiveness of Hybrid Pooling in Mixup-Based Graph Learning for Language Processing

Implementation of Journal of Systems and Software (JSS 2024) paper: On the Effectiveness of Hybrid Pooling in Mixup-Based Graph Learning for Language Processing [arxiv].

We build this project on the top of GNN-FakeNews and Project_CodeNet. Please refer to these projects for more details.

Introduction

The performance of linear interpolation methods, especially those used for graph learning, can be significantly influenced by graph pooling operators. To investigate this, we conduct a comprehensive empirical study by applying Manifold-Mixup to a formal characterization of graph pooling. This study encompasses 11 graph pooling operations, including 9 hybrid and 2 non-hybrid pooling operators, to explore how these operators affect the performance of Mixup-based graph learning.

Requirements

On Ubuntu:

Installation

  • install python packages
pip install tqdm
pip install pandas
pip install ogb
pip install keras
pip install scikit-learn
pip install scipy
  • Task: Problem Classification
Python>=3.6
cuDNN>=7.6
PyTorch>=version 1.8.0) 
Pytorch Geometric>=version 1.6.3
CUDA 11.0
  • Task: Fake News Detection
Python>=3.6
cuDNN>=7.6
Pytorch>=version 1.8.0
Pytorch Geometric>=version 1.6.3
CUDA 11.0

Experiments

The script to run the experiments is ./run.sh

  • Task: Problem Classification
cd text_detection

./run.sh
  • Task: Fake News Detection
cd Problem_classification/gnn-based-experiments

./run.sh

Dataset

Citation

If you use the code in your research, please cite:

    @article{DONG2024112139,
    title = {On the effectiveness of hybrid pooling in mixup-based graph learning for language processing},
    journal = {Journal of Systems and Software},
    volume = {216},
    pages = {112139},
    year = {2024},
    issn = {0164-1212},
    doi = {https://doi.org/10.1016/j.jss.2024.112139},
    author = {Zeming Dong and Qiang Hu and Zhenya Zhang and Yuejun Guo and Maxime Cordy and Mike Papadakis and Yves Le Traon and Jianjun Zhao},
    }