Probabilistic Residual User Clustering
Published in IJCAI2025 Workshop / Submitted to TMLR, 2024
This work proposes PRUC (Probabilistic Residual User Clustering), a causal Bayesian framework that clusters users and models residuals between predicted and true ratings to enhance recommendation accuracy.
Key Contributions
- Introduced a plug-and-play architecture compatible with diverse deep learning recommenders
- Improved performance in cold-start and domain-shift settings
- Demonstrated significant improvement across benchmark datasets while uncovering meaningful user clusters via latent variable inference
Status: Accepted at IJCAI2025 Workshop on Causal Learning for Recommendation Systems, Under Review at TMLR
Recommended citation: Wenyuan Wang, Yusong Zhao, Zihao Xu, Hengyi Wang, Shreya Venugopal, Desmond Lobo, Chengzhi Mao, Qi Xu, Zhigang Hua, Yan Xie, Bo Long, Shuang Yang, Hao Wang. "Probabilistic Residual User Clustering." IJCAI2025 Workshop / Submitted to TMLR.
Download Paper | Download Slides