Reproduce “PRISM: Personalized Item Representations in User History” 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
Renzhi Wu, Zikun Cui, Junjie Yang, Li Sheng, Yongjun Xie, Jerry Fu, Nipun Mathur, Sri Reddy and 1 more
PRISM builds personalized codes for historical user–item encounters by reusing representations from an upstream ranker. It compresses user-state, user–item, and same-request candidate information into replayable history entries. In an offline Surface 1 evaluation across 13 engagement tasks, PRISM reduces normalized entropy by 0.15%, compared with 0.07% for VISTA; gains rose from 0.06% to 0.17% across tested histories of 26–240 events.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.