Reproduce “Beyond Semantic IDs: Atomic Creator Tokens for Large-Scale 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
Zhuosheng Liu, Yihan Wu, Ziyun Xu, Xiaoxuan Liu, Longfei Yun, Kevin Han, Mingze Gao, Pengfei Wang and 3 more
ACT gives each creator a dedicated trainable token rather than representing creators with shared quantized codes. The authors compare ACT with a matched semantic ID system, where several creators can share a code. On production data at the 10Y vocabulary tier, ACT reaches 43.41% Recall@15, versus 23.58% for SID. The finding applies to in-vocabulary targets, not necessarily sparse long-tail creators.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.