Reproduce “Personalized Graph Signal Processing for Collaborative Filtering” 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

Jiahao Liu, Dongsheng Li, Hansu Gu, Tun Lu, Peng Zhang, Li Shang, Ning Gu

Personalized Graph Signal Processing (PGSP) reframes collaborative filtering as signal propagation over a graph containing users, items, and their cross-relations. By enriching sparse interaction histories with user similarity and mixing global low-pass filtering with local higher-frequency information, PGSP outperforms prior matrix-factorization, GNN, and graph-signal baselines on Gowalla, Yelp2018, and Amazon-Book, improving average Recall and NDCG over GF-CF by 1.83% and 2.61%, respectively.

Key insights

  • Preference signals are not uniformly noise-separated by frequency: globally smooth components capture broad interests, while locally smooth but globally rough components can encode personalization.
  • User similarity is especially useful for sparse histories because interactions from neighboring users provide indirect evidence about items a target user may prefer.
  • The contribution of higher-frequency information depends on data density; it is strongest on sparse datasets such as Amazon-Book and comparatively small on denser MovieLens-1M.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.