Reproduce “Multi-Objective Recommendation via Multivariate Policy Learning” 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

Olivier Jeunen, Jatin Mandav, Ivan Potapov, Nakul Agarwal, Sourabh Vaid, Wenzhe Shi, Aleksei Ustimenko

The authors extend counterfactual risk minimization to learn multivariate recommendation weights from logged randomized experiments, optimizing a platform-chosen long-term reward. They add an effective-sample-size correction to make conservative confidence bounds more reliable in finite samples. In simulation, its best variant reduced required sample size by up to 60.7×; an online test found a positive retention interval on platform B.

Summary by alphaXiv.

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