Reproduce “SUBER: An RL Environment with Simulated Human Behavior for Recommender Systems” 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
Nathan Corecco, Giorgio Piatti, Luca A. Lanzendörfer, Flint Xiaofeng Fan, Roger Wattenhofer
Researchers at ETH Zurich and the National University of Singapore developed SUBER, a reinforcement learning environment that utilizes Large Language Models to simulate realistic user behavior for training and evaluating recommender systems. The framework successfully trains diverse RL agents, with the A2C algorithm achieving a MAP@10 of 0.88 and MRR@10 of 0.96 while aligning recommendations with simulated user preferences.
Key insights
- Large Language Models can effectively simulate nuanced human preferences and decision-making for recommender systems, demonstrating high fidelity with human judgment.
- Careful prompt engineering, including specific rating scales (e.g., 0-9 digits) and few-shot examples, significantly improves LLM accuracy in simulating user responses.
- An effective item retrieval mechanism, particularly similarity-based, is essential for LLMs to leverage historical user interactions accurately.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.