Reproduce “ReSID: Rethinking Long-Tail Retrieval in Generative Recommendation through Item Reachability” 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

Muyu Zou, Shunye Tang, Yi Xu, Haibo Xing, Yandong Wang, Lingyu Mu, Jinxin Hu

ReSID improves long-tail item retrieval in generative recommendation by guiding beam search with popularity calibration and user-semantic alignment, then matching catalog items whose Semantic IDs differ from generated sequences in up to three positions. On In-house, long-tail Recall@200 is 36.62%, versus Ghost’s 18.94%, under a narrower beam; recovery remains limited by paths surviving decoding.

Summary by alphaXiv.

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