Reproduce “Not All Branches Are Equal: Adaptive Semantic ID Construction for Generative 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
Guy Hadad, Haggai Roitman, Bracha Shapira
AST builds semantic item IDs as recursively clustered paths, selecting each node’s branch count by silhouette score. In offline next-item tests across five Amazon domains, AST-V2 reports a 13.08% overall relative gain over the strongest non-AST baseline across four metrics. It does not beat AST-V1 in every cell: Sports NDCG@10 is 2.54% for AST-V1 and 2.51% for AST-V2.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.