Reproduce “DEAR: Deep Reinforcement Learning for Online Advertising Impression in 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
Xiangyu Zhao, Changsheng Gu, Haoshenglun Zhang, Xiwang Yang, Xiaobing Liu, Jiliang Tang, Hui Liu
Researchers from Michigan State University and Bytedance developed DEAR, a Deep Reinforcement Learning framework that addresses the trade-off between advertising revenue and user experience in recommender systems. It employs a novel Deep Q-network to simultaneously determine whether, which, and where to place an ad, achieving a 6.712% improvement in accumulated reward over hierarchical DQN baselines on a real-world Douyin dataset.
Key insights
- Reinforcement learning is critical for long-term optimization in balancing ad revenue and user experience, outperforming supervised learning approaches.
- The proposed novel DQN architecture effectively and efficiently manages the simultaneous and interdependent decisions of ad interpolation, offering improved stability over conventional and hierarchical RL methods for this task.
- Accurate modeling of dynamic user preferences through sequential processing of both recommended and advertised item histories is crucial for optimal ad placement decisions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.