HHGT: Hierarchical Heterogeneous Graph Transformer for Heterogeneous Graph Representation Learning

This repository hosts the code, datasets of HHGT, as well as the instructions to the selected baselines.

🎉 News

• [x] 🚀 [2024.10] 🎯🎯📢📢 Our 🌟HHGT🌟 has been accepted by WSDM'2025 Research Track (17.3% acceptance rate)! Congrats! 🎉🎉🎉

Baseline links

• M2V/PTE/HIN2Vec/ComplEx/AspEm/HAN/R-GCN: https://github.com/yangji9181/HNE
• SHGP: https://github.com/kepsail/SHGP
• AGAT: https://github.com/Qidong-Liu/Aspect-Aware-Graph-Attention-Network
• HGT: https://github.com/dmlc/dgl/tree/master/examples/pytorch/hgt
• GTN/FastGTN: https://github.com/seongjunyun/Graph_Transformer_Networks
• HINormer: https://github.com/Ffffffffire/HINormer
• NAGphormer: https://github.com/JHL-HUST/NAGphormer

Requirements

The code has been tested under Python 3.9, with the following packages installed (along with their dependencies):

• torch >= 1.9.0
• pytorch-lightning >= 1.4.4
• torchmetrics >= 0.5.0
• torch-scatter >= 2.0.9
• torch-sparse >= 0.6.12
• numpy
• pandas
• tqdm
• yaml

Files in the folder

• /data: Put both acm and mag datasets under it to store the prepared data.
• /dataloader: Codes of the dataloader.
• /models: Codes of the HHGT model, semi-supervised node classification task.
• /utils: Codes for some utils.
• /lightning_logs: Store the trained model parameters, setting files, checkpoints, logs and results.
• main.py: The main entrance of running.
• /records: Store the output node embeddings for evaluation.

Node Classification Task

train HHGT by

# train HHGT.
python main.py --setting_path *.yaml

#  for example
#  MAG dataset
python main.py --setting_path lightning_logs/mag_best/mag_settings_cla.yaml

The *.yaml is the configuration file.

And if you want to adjust the hyperparameters of the model, you can modify it in *.yaml, or create a similar configuration file, and specify --setting_path like this:

python main.py --setting_path yourpath.yaml

Checkpoints, logs, and results during training will be stored in the directory: ./lightning_logs/version_0

Load the node embeddings and evaluate the results by:

# test 
python cla_pre.py 

PS: Keep the configuration file unchanged during training.