Reproduce “A Feedback Shift Correction in Predicting Conversion Rates under Delayed Feedback” 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
Shota Yasui, Gota Morishita, Komei Fujita, Masashi Shibata
Researchers at CyberAgent introduce the "feedback shift" concept to address the challenge of delayed feedback in online advertising's conversion rate prediction. Their Importance Weighting approach improved conversion volume by 31% and reduced training time by over 70 times compared to existing methods, showing enhanced predictive accuracy and efficiency.
Key insights
- The delayed feedback problem can be precisely characterized as a "feedback shift," allowing for a targeted and theoretically sound correction method.
- Importance Weighting, traditionally used for covariate shift, can be effectively extended and proven consistent for correcting this specific type of conditional label distribution shift.
- Real-world conversion delays exhibit complex, non-parametric patterns, which are implicitly captured by the flexible LightGBM models used for FSIW estimation, outperforming methods relying on simple delay distribution assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.