MEMBER

This is the official implementation of MEMBER (Mixture-of-Experts for Multi-BEhavior Recommendation)

(Accepted for ACM CIKM 2025 Full Research Paper)

---

📑 Paper & Appendix

• Paper: arXiv:2508.19507
• Accepted at: ACM CIKM 2025 (Full Research Paper)
• Online Appendix: See MEMBER-Online Appendix.pdf in the root folder.

---

📊 Datasets

We use three widely adopted multi-behavior recommendation datasets:

• Tmall
• Taobao
• Jdata

Preprocessing

cd data/{data_name}
python preprocess.py

---

📈 Evaluation

We report three evaluation results:

1. Overall performance under the standard evaluation
2. Performance on visited items
3. Performance on unvisited items

How to Run MEMBER

cd METHOD

• Tmall

python main.py --data_name tmall --con_s 0.1 --temp_s 0.6  --con_us 0.1 --temp_us 0.7 --gen 0.1 --lambda_s 0.6 --alpha 2

• Taobao

python main.py --data_name taobao --con_s 0.1 --temp_s 0.8 --con_us 0.1 --temp_us 0.7 --gen 0.1 --lambda_us 0.6

• Jdata

python main.py --data_name jdata --con_s 0.1 --temp_s 0.6 --con_us 0.01 --temp_us 1.0 --gen 0.01 --lambda_s 0.4 --lambda_us 0.4 --alpha 2

📚 Citation

If you find MEMBER useful, please cite our paper:

@inproceedings{kim2025self,
  title     = {A Self-Supervised Mixture of Experts Framework for Multi-behavior Recommendation},
  author    = {Kim, Kyungho and Kim, Sunwoo and Lee, Geon and Shin, Kijung},
  booktitle = {CIKM},
  year      = {2025}
}