From Latent to Observable Position-Based Click Models in Carousel Interfaces
New model ditches hidden variables for real eye-tracking data, achieving the strongest click prediction performance.
Researchers Santiago de Leon-Martinez, Robert Moro, Branislav Kveton, and Maria Bielikova published a paper introducing three new position-based click models for carousel interfaces. Their key innovation is the Observed Examination Position-Based Model (OEPBM), which replaces latent variables with observed eye-tracking signals. In experiments, OEPBM achieved the best click prediction and most realistic user behavior alignment. The work reveals a fundamental limitation: optimizing for clicks alone does not guarantee accurate modeling of user examination patterns in complex UIs.
Why It Matters
This challenges how major platforms like Netflix and YouTube optimize recommendations, pushing for models that understand real user attention, not just clicks.