Reproduce “Learning List-wise Representation in Reinforcement Learning for Ads Allocation with Multiple Auxiliary Tasks” 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
Ze Wang, Guogang Liao, Xiaowen Shi, Xiaoxu Wu, Chuheng Zhang, Yongkang Wang, Xingxing Wang, Dong Wang
A reinforcement learning framework enhances list-wise representation learning for ads allocation by integrating three domain-specific auxiliary tasks. This approach improved ads revenue by 2.92% and user experience scores by 1.13% in online A/B tests on a major food delivery platform.
Key insights
- Domain-specific auxiliary tasks, tailored to e-commerce ads allocation, effectively address the challenges of high-dimensional state-action spaces by guiding the learning of more robust and generalizable list-wise representations.
- The combination of reconstruction (for preserving key item info), prediction (for dense user behavior signals), and contrastive learning (for representation differentiation) significantly improves the RL agent's performance.
- Careful hyperparameter tuning for auxiliary task weights and the amount of information used for reconstruction is crucial for optimizing the balance between different learning objectives and maximizing overall performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.