Position-Aware Subgraph Neural Networks with Data-Efficient Learning

This paper: Position-Aware Subgraph Neural Networks with Data-Efficient Learning, is submitted to WSDM 2023.

Dataset

PADEL and baseline methods are implemented on the hpo_metab, hpo_neuro, and em_user datasets, firstly relsed by SubGNN. We provide these datasets in different data-efficient situations:

datasets-PADEL.7z
├─em_user
│      edge_list.txt
│      subgraphs.pth
│      subgraphs_10.pth
│      subgraphs_20.pth
│      subgraphs_30.pth
│      subgraphs_40.pth
│      subgraphs_50.pth
│
├─hpo_metab
│      edge_list.txt
│      subgraphs.pth
│      subgraphs_10.pth
│      subgraphs_20.pth
│      subgraphs_30.pth
│      subgraphs_40.pth
│      subgraphs_50.pth
│
└─hpo_neuro
        edge_list.txt
        subgraphs.pth
        subgraphs_10.pth
        subgraphs_20.pth
        subgraphs_30.pth
        subgraphs_40.pth
        subgraphs_50.pth
------ 

edge_list.txt is the orignal edge list file for the base graph. subgraphs.pth is the original subgraph file with subgraph and labels. subgraphs_X0 means the new subgraph file consisting of X0% of the training set and the original validation/ test set.

Codes

We provide pseudo-code for random 1-hop subgraph diffusion and PADEL's training pipline in PADEL_pseudo_code.pdf, and the source code has been released!

Todo List

• Dataset
• Pseudo-code
• Source code
• Setup guidelines

Please let me know if you have any question :-)