Reproduce “RecSim: A Configurable Simulation Platform 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

Eugene Ie, Chih-wei Hsu, Martin Mladenov, Vihan Jain, Sanmit Narvekar, Jing Wang, Rui Wu, Craig Boutilier

RECSIM is an open-source, configurable simulation platform from Google Research designed to facilitate the development and evaluation of reinforcement learning algorithms for sequential interactive recommender systems. It allows researchers to author stylized environments, enabling the study of complex user dynamics, partial observability, and long-term optimization challenges while fostering academic-industrial collaboration.

Key insights

  • Stylized simulation environments are effective for studying complex aspects of interactive recommender systems, such as the value of exploration for discovering unobservable user latent states for long-term engagement.
  • Decomposition techniques for handling large combinatorial action spaces in slate recommendation can yield substantial performance gains and demonstrate robustness to shifts in user choice models.
  • Hierarchical policies utilizing temporal aggregation or regularization can amplify the advantage function, making learned policies more robust to low signal-to-noise ratio (SNR) and slow-evolving user states in long-horizon environments.

Summary by alphaXiv.

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