Reproduce “Towards Validating Long-Term User Feedbacks in Interactive Recommendation 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
Hojoon Lee, Dongyoon Hwang, Kyushik Min, Jaegul Choo
The paper investigates the validity of using public review datasets to benchmark Reinforcement Learning (RL) based Interactive Recommender Systems (IRS), finding that these datasets often lack significant long-term user feedback. A simple greedy recommendation model frequently matches or exceeds the performance of complex RL models in optimizing cumulative rewards, suggesting a re-evaluation of current evaluation protocols.
Key insights
- Sophisticated RL-based IRS models often do not significantly outperform a simple greedy approach when evaluated on standard public review datasets.
- Increasing the emphasis on future rewards by raising the discount factor typically degrades the performance of RL models on these datasets.
- Public review datasets commonly used for benchmarking IRS exhibit only marginal long-term effects in user feedback.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.