Reproduce “Incentive-Aware Recommender Systems in Two-Sided Markets” 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
Xiaowu Dai, Wenlu Xu, Yuan Qi, Michael I. Jordan
This research introduces "incentive-aware" recommender systems for two-sided markets, developing Adaptive Recommendation Policy (ARP) and Modified Adaptive Recommendation Policy (MARP) algorithms. These systems explicitly account for user opportunity costs and guarantee ex-post fairness, achieving asymptotically optimal regret rates of O("sqrt(mT log(mT))") and O("sqrt(T log m)") respectively.
Key insights
- The capacity for platform exploration is inherently limited by user incentive constraints, meaning learning under these constraints is slower than in unconstrained environments.
- Having a known "good" arm or established credibility is crucial for platforms to mitigate the "cold-start problem" for new items and effectively incentivize exploration.
- By leveraging cumulative loss as feedback, platforms can implicitly infer and adapt to heterogeneous and private user opportunity costs, enabling incentive-aware design even without explicit preference data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.