PGPO: Prototype-Guided Progressive Obfuscation for Privacy-Preserving LLM-Enhanced Recommendation
Paper โข 2611.23236 โข Published โข 2
How to use Shengxiang-Lin/PGPO with PEFT:
Task type is invalid.
This repository contains the LoRA adapter introduced in PGPO: Prototype-Guided Progressive Obfuscation for Privacy-Preserving LLM-Enhanced Recommendation.
PGPO learns topology-aware, privacy-preserving prototype-word obfuscations while preserving inter-item semantic relations for LLM-enhanced recommendation. The adapter is trained with GRPO on top of Qwen3-4B-Base using TRL and Unsloth.