Reproduce “PGPO: Prototype-Guided Progressive Obfuscation for Privacy-Preserving LLM-Enhanced 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

Shengxiang Lin, Jiajie Su, Pengyang Zhou, Xiang Chen, Xiaolin Zheng, Feng Tian, Chaochao Chen

PGPO obfuscates item attributes for recommendation while preserving relationships among semantic neighborhoods rather than each word’s literal meaning. It selects prototype words, expands variants through a neighborhood graph, and uses three GRPO rewards. On MovieLens-1M, category-label exact-match inversion falls from 0.0641 for raw descriptions to 0.0309, while FREEDOM HR@10 remains similar. These results concern tested protocols, not all attacks or recommenders.

Summary by alphaXiv.

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