Reproduce “Reinforced Prompt Personalization for Recommendation with Large Language Models” 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
Wenyu Mao, Jiancan Wu, Weijian Chen, Chongming Gao, Xiang Wang, Xiangnan He
Researchers from the University of Science and Technology of China introduce Reinforced Prompt Personalization (RPP/RPP+), a multi-agent reinforcement learning framework that automatically generates personalized discrete prompts for individual users in recommender systems. This approach significantly improves recommendation accuracy across various LLMs compared to fixed, task-wise prompting methods and traditional recommender models.
Key insights
- Fixed, task-wise prompts are inherently inadequate for personalized recommendations with LLMs due to user diversity and LLM prompt sensitivity.
- Decomposing prompts into patterns like 'role-playing' and 'reasoning guidance' allows for effective, targeted personalization, with 'reasoning guidance' showing the most impact.
- A multi-agent reinforcement learning approach is effective for optimizing discrete prompt components, demonstrating strong generalization across various LLMs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.