This is a PyTorch implementation of paper: Motif Graph Neural Network

• Python == 3.8
• PyTorch == 1.12.1
• DGL == 0.9.0
• PyTorch Geometric == 1.7.2
• NumPy == 1.18.0
• SciPy == 1.6.2

• MGNN/data/GraphMGNN_DATA/: datasets for graph classification
  • ./AIDS/raw: original dataset of AIDS
  • ./AIDS/processed: processed dataset of AIDS
  • ./ProcessedMotif.7z: processed motif data of AIDS, ENZYMES and MUTAG.

The introduction of ./ENZYMES and ./MUTAG is the same as above.

• MGNN/data/NodeMGNN_DATA/: datasets for node classification
  • ./Cora/raw: original dataset of Cora
  • ./Cora/processed: processed dataset of Cora
  • ./processed/motif_adj4Cora: contain 13 motif-based adjacency matrices of Cora

The introduction of ./CiteSeer, ./PubMed and ./chem2bio2rdf is the same as above.

• MGNN/MGNN_Graph/: graph classification code of MGNN

  • layer.py: implementation of a MGNN layer
  • preprocess.py: build 13 motif adjacency matrices for each graph
  • utils.py: implementation of building 13 motif adjacency matrices for a graph and other utility functions
  • main.py: MGNN implementation, training and evaluation

• MGNN/MGNN_Node/: node classification code of MGNN on Cora, Citeseer and Pubmed

  • layer.py: implementation of a MGNN layer
  • utils.py: implementation of building 13 motif adjacency matrices for a graph and other utility functions
  • main.py: MGNN implementation, training and evaluation

• MGNN/MGNN_CBR/: node classification code of MGNN on Chem2Bio2RDF and the introduction of its subdirectory is the same as MGNN/MGNN_Node/.

1. Please unzip all 7z files to their directory first.

within MGNN/

2. For graph classification task, run the following scripts: ./MGNN_Graph/main.py
3. For node classification task, run the following scripts:

• on Cora, Citeseer and Pubmed: ./MGNN_Node/main.py
• on Chem2Bio2RDF: ./MGNN_CBR/main.py

If you find this code useful, please cite the following:

@ARTICLE{10154572,
  author={Chen, Xuexin and Cai, Ruichu and Fang, Yuan and Wu, Min and Li, Zijian and Hao, Zhifeng},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={Motif Graph Neural Network}, 
  year={2023},
  volume={},
  number={},
  pages={1-15},
  doi={10.1109/TNNLS.2023.3281716}}