Reproduce “Repeat-bias-aware Optimization of Beyond-accuracy Metrics for Next Basket Recommendation” 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
Yuanna Liu, Ming Li, Mohammad Aliannejadi, Maarten de Rijke
A model-agnostic post-processing approach improves diversity and item fairness in Next Basket Recommendation (NBR) by explicitly mitigating repeat bias, bringing the recommended repeat ratio closer to ground truth while maintaining acceptable prediction accuracy. This is achieved through Mixed-Integer Linear Programming formulations that jointly optimize multiple beyond-accuracy objectives.
Key insights
- Optimizing solely for diversity or item fairness in Next Basket Recommendation (NBR) can, counter-intuitively, increase repeat bias, where systems disproportionately recommend previously purchased items.
- A joint optimization framework that simultaneously considers predicted relevance, diversity, item fairness, and repeat bias mitigation is necessary to achieve more balanced and desirable NBR outcomes.
- Model-agnostic re-ranking, formulated as a Mixed-Integer Linear Program, can effectively address these multiple objectives as a post-processing step for various NBR model architectures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.