Reproduce “Personalized and Multi-View Representation for Federated Cold-Start Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jaehyung Lim, Wonbin Kweon, Woojoo Kim, Junyoung Kim, Dongha Kim, Hwanjo Yu
Pohang University of Science and Technology researchers developed PMFRec, a federated learning framework that generates personalized and multi-view item representations for cold-start recommendations, improving performance by up to 7.56% over state-of-the-art baselines. This approach enhances personalization for new items and captures diverse semantic facets from item attributes while maintaining communication efficiency and superior robustness under local differential privacy.
Key insights
- Combining personalized user-specific interpretations with a globally learned multi-view attribute representation is critical for effective federated cold-start recommendation.
- Multi-view encoding with item-adaptive gating and diversity-promoting objectives (orthogonality, load-balancing) effectively addresses semantic entanglement and improves the quality of item representations.
- A regularizer-free client-side training approach significantly enhances robustness to Local Differential Privacy noise and improves communication efficiency in federated settings by avoiding signal interference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.