Reproduce “A Model-based Multi-Agent Personalized Short-Video Recommender System” 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
Peilun Zhou, Xiaoxiao Xu, Lantao Hu, Han Li, Peng Jiang
The Model-based Multi-agent Ranking Framework (MMRF) combines multi-agent reinforcement learning with model-based techniques to optimize multiple user engagement metrics and address sample selection bias in short-video recommendation. Deployed in a large-scale production environment, it demonstrated sustained improvements in watch-time, follow actions, and session depth during online A/B testing.
Key insights
- Reformulating recommendation as a multi-agent Markov Decision Process allows for effective concurrent optimization of multiple user engagement metrics.
- Model-based simulation of user feedback for non-impression samples, combined with uncertainty estimation, provides a robust method to address sample selection bias.
- An attention-based collaboration mechanism enables specialized agents to share relevant information and improve collective decision-making in a multi-objective setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.