TopKGAT: A Top-K Objective-Driven Architecture for Recommendation

This is the official PyTorch implementation for our WWW 2026 paper.

Sirui Chen, Jiawei Chen, Canghong Jin, Sheng Zhou, Jingbang Chen, Wujie Sun and Can Wang. TopKGAT: A Top-K Objective-Driven Architecture for Recommendation arXiv link

Environment

• python==3.9.19
• numpy==1.26.4
• pandas==2.2.1
• torch==2.2.2

Datasets

Dataset	#Users	#Items	#Interactions
Ali-Display	17,730	10,036	173,111
Epinions	17,893	17,659	301,378
Food	14,382	31,288	456,925
Gowalla	55,833	118,744	1,753,362

Training & Evaluation

• Ali-Display

python -u code/main.py --data=Ali-Display --TopKformer_layers=4 --emb_learning_rate=1e-1 --emb_reg_lambda=1e-4

• Epinions

python -u code/main.py --data=Epinions --TopKformer_layers=4 --emb_learning_rate=1e-1 --emb_reg_lambda=1e-4

• Food

python -u code/main.py --data=Food --TopKformer_layers=4 --emb_learning_rate=1e-2 --emb_reg_lambda=0

• Gowalla

python -u code/main.py --data=Gowalla --TopKformer_layers=3 --emb_learning_rate=1e-1 --emb_reg_lambda=0

Citation

If you find the paper useful in your research, please consider citing:

@article{chen2026topkgat,
  title={TopKGAT: A Top-K Objective-Driven Architecture for Recommendation},
  author={Chen, Sirui and Chen, Jiawei and Jin, Canghong and Zhou, Sheng and Chen, Jingbang and Sun, Wujie and Wang, Can},
  journal={arXiv preprint arXiv:2601.18432},
  year={2026}
}