Rankformer: A Graph Transformer for Recommendation based on Ranking Objective

This is the PyTorch implementation for our WWW 2025 paper.

Sirui Chen, Shen Han, Jiawei Chen, Binbin Hu, Sheng Zhou, Gang Wang, Yan Feng, Chun Chen, Can Wang. Rankformer: A Graph Transformer for Recommendation based on Ranking Objective 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
Amazon-CDs	51,266	46,463	731,734
Yelp2018	167,037	79,471	1,970,721

Training & Evaluation

• Ali-Display

# Rankformer
python -u code/main.py --data=Ali-Display \
    --use_gcn \
    --use_rankformer --rankformer_layers=4 --rankformer_tau=0.5 \

# Rankformer-CL
python -u code/main.py --data=Ali-Display \
    --use_cl \
    --use_gcn --gcn_layers=2 --gcn_left=0.5 --gcn_right=0.5 \
    --use_rankformer --rankformer_layers=5 --rankformer_tau=0.1 \
    --learning_rate=1e-3 --loss_batch_size=2048 --valid_interval=1

• Epinions

# Rankformer
python -u code/main.py --data=Epinions \
    --use_gcn \
    --use_rankformer --rankformer_layers=4 --rankformer_tau=0.4
    
# Rankformer-CL
python -u code/main.py --data=Epinions \
    --use_cl \
    --use_gcn --gcn_layers=2 --gcn_left=0.5 --gcn_right=0.5 \
    --use_rankformer --rankformer_layers=3 --rankformer_tau=0.2 \
    --learning_rate=1e-3 --loss_batch_size=2048 --valid_interval=1

• Amazon-CDs

# Rankformer
python -u code/main.py --data=Amazon-CDs \
    --use_gcn \
    --use_rankformer --rankformer_layers=2 --rankformer_tau=0.5
    
# Rankformer-CL
python -u code/main.py --data=Amazon-CDs \
    --use_cl \
    --use_gcn --gcn_layers=2 --gcn_left=0.5 --gcn_right=0.5 \
    --use_rankformer --rankformer_layers=4 --rankformer_tau=0.2 \
    --learning_rate=1e-3 --loss_batch_size=2048 --valid_interval=1

• Yelp2018

# Rankformer
python -u code/main.py --data=Yelp2018 \
    --use_gcn \
    --use_rankformer --rankformer_layers=3 --rankformer_tau=0.4
    
# Rankformer-CL
python -u code/main.py --data=Yelp2018 \
    --use_cl \
    --use_gcn --gcn_layers=2 --gcn_left=0.5 --gcn_right=0.5 \
    --use_rankformer --rankformer_layers=4 --rankformer_tau=0.3 \
    --learning_rate=1e-3 --loss_batch_size=2048 --valid_interval=1

Citation

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

@inproceedings{chen2025rankformer,
  title={Rankformer: A Graph Transformer for Recommendation based on Ranking Objective},
  author={Chen, Sirui and Han, Shen and Chen, Jiawei and Hu, Binbin and Zhou, Sheng and Wang, Gang and Feng, Yan and Chen, Chun and Wang, Can},
  booktitle={Proceedings of the ACM on Web Conference 2025},
  pages={3037--3048},
  year={2025}
}