xPath

This is a PyTorch implementation for our AAAI'23 paper: Towards Fine-grained Explainability for Heterogeneous Graph Neural Network.

Here also include the supplementary materials for proof of theorems in the paper and additional experimental results.

An alternative implementation based on Mindspore is coming soon.

Requirements

• Python 3.7
• torch~=1.10.0
• dgl~=0.8.2
• numpy~=1.21.6
• tqdm~=4.64.0

Datasets

The heterogeneous graph datasets we use are DBLP, ACM, and IMDB. Datasets are provided in the data folder. Taking DBLP dataset as an example:

• dblp_graph.bin: Heterogeneous graph。
• dblp_index_60.bin: Training, validation, and test set for SimpleHGN.
• dblp_index_2000.bin: Training, validation, and test set for HGT.

Train HGNs

Trained HGNs are provided in ckpt/{dataset_name}/bk for reproducing the results in our paper. To retrain the HGNs, run

# edit configurations in config.py
python train.py

Generate explanations

# edit configurations in config.py
python main.py