ApeGNN: Node-Wise Adaptive Aggregation in GNNs for Recommendation

ApeGNN develops a node-wise adaptive diffusion mechanism for information aggregation, in which each node is enabled to adaptively decide its diffusion weights based on the local structure (e.g., degree).

This is our PyTorch implementation for WWW'23 paper:

Environment Requirements

The code has been tested running under Python 3.9.7. The required packages are as follows:

• pytorch == 1.9.1+cu111

Usage Example

Running one trial of Heat Kernel on AMiner:

nohup python main.py --dataset aminer --gnn ApeGNN_HT --pool sum --Ks '[20, 50]' --step 1 --runs 1 --e 1e-7 --gpu_id 1 > ./logs/aminer/ApeGNN_HT.log 2>&1 &

Running one trial of PPR on AMiner:

nohup python main.py --dataset aminer --gnn ApeGNN_APPNP --pool sum --Ks '[20, 50]' --step 1 --runs 1 --e 1e-7 --gpu_id 1 > ./logs/aminer/ApeGNN_APPNP.log 2>&1 &

Dataset

In the main results compared with representative models, we use six processed datasets: Ali, Amazon, AMiner, Gowalla, MovieLens, and Yelp2018.

Dataset	#Users	#Items	#Interactions	Density
Ali	106,042	53,591	907,407	0.016
Amazon	192,403	63,001	1,689,188	0.014
AMiner	5,340	14,967	163,084	0.204
Gowalla	29,858	40,981	1,027,370	0.084
MovieLens	6,040	3,416	999,611	4.362
Yelp2018	31,668	38,048	1,561,406	0.130

Citing

If you find this work is helpful to your research, please consider citing our paper:

@inproceedings{zhang2023apegnn,
  title={ApeGNN: Node-Wise Adaptive Aggregation in GNNs for Recommendation},
  author={Zhang, Dan and Zhu, Yifan and Dong, Yuxiao and Wang, Yuandong and Wenzheng, Feng and Kharlamov, Evgeny and Tang, Jie},
  booktitle={WWW'23},
  year={2023}
}