Reproduce “Fatigue-aware Bandits for Dependent Click Models” 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
Junyu Cao, Wei Sun, Zuo-Jun, Shen, Markus Ettl
Researchers at UC Berkeley and IBM Research developed a fatigue-aware Dependent Click Model (DCM) for recommender systems, integrating marketing and content fatigue to enhance user engagement. The approach yields UCB-based online learning algorithms with sublinear regret bounds and empirically outperforms a standard explore-then-exploit benchmark in maximizing clicks on both simulated and real-world datasets.
Key insights
- The optimal recommendation sequence is determined solely by item intrinsic relevance and the content fatigue discount factor, independent of user click and non-click resumption probabilities.
- Explicitly modeling both marketing fatigue (user exit at any point) and content fatigue (decreasing attractiveness of similar items) creates a more realistic and robust user behavior prediction framework.
- UCB-based algorithms can effectively learn complex, interdependent parameters in an online setting, achieving sublinear regret bounds despite the non-stationary nature of item attractiveness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.