Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Latent to Observable Position-Based Click Models in Carousel Interfaces

Santiago de Leon-Martinez, Robert Moro, Branislav Kveton, Maria Bielikova

TL;DR

This paper proposes three novel position-based models tailored to carousels, including the first position-based model without latent variables that incorporates observed examination signals derived from eye tracking data, called the Observed Examination Position-Based Model (OEPBM).

Abstract

Click models are a central component of learning and evaluation in recommender systems, yet most existing models are designed for single ranked-list interfaces. In contrast, modern recommender platforms increasingly use complex interfaces such as carousels, which consist of multiple swipeable lists that enable complex user browsing behaviors. In this paper, we study position-based click models in carousel interfaces and examine optimization methods, model structure, and alignment with user behavior. We propose three novel position-based models tailored to carousels, including the first position-based model without latent variables that incorporates observed examination signals derived from eye tracking data, called the Observed Examination Position-Based Model (OEPBM). We develop a general implementation of these carousel click models, supporting multiple optimization techniques and conduct experiments comparing gradient-based methods with classical approaches, namely expectation-maximization and maximum likelihood estimation. Our results show that gradient-based optimization consistently achieve better click likelihoods. Among the evaluated models, the OEPBM achieves the strongest performance in click prediction and produces examination patterns that most closely align to user behavior. However, we also demonstrate that strong click fit does not imply realistic modeling of user examination and browsing patterns. This reveals a fundamental limitation of click-only models in complex interfaces and the need for incorporating additional behavioral signals when designing click models for carousel-based recommender systems.

From Latent to Observable Position-Based Click Models in Carousel Interfaces

TL;DR

This paper proposes three novel position-based models tailored to carousels, including the first position-based model without latent variables that incorporates observed examination signals derived from eye tracking data, called the Observed Examination Position-Based Model (OEPBM).

Abstract

Click models are a central component of learning and evaluation in recommender systems, yet most existing models are designed for single ranked-list interfaces. In contrast, modern recommender platforms increasingly use complex interfaces such as carousels, which consist of multiple swipeable lists that enable complex user browsing behaviors. In this paper, we study position-based click models in carousel interfaces and examine optimization methods, model structure, and alignment with user behavior. We propose three novel position-based models tailored to carousels, including the first position-based model without latent variables that incorporates observed examination signals derived from eye tracking data, called the Observed Examination Position-Based Model (OEPBM). We develop a general implementation of these carousel click models, supporting multiple optimization techniques and conduct experiments comparing gradient-based methods with classical approaches, namely expectation-maximization and maximum likelihood estimation. Our results show that gradient-based optimization consistently achieve better click likelihoods. Among the evaluated models, the OEPBM achieves the strongest performance in click prediction and produces examination patterns that most closely align to user behavior. However, we also demonstrate that strong click fit does not imply realistic modeling of user examination and browsing patterns. This reveals a fundamental limitation of click-only models in complex interfaces and the need for incorporating additional behavioral signals when designing click models for carousel-based recommender systems.
Paper Structure (24 sections, 46 equations, 4 figures, 3 tables)

This paper contains 24 sections, 46 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Background: Carousel Cascade Models
  4. Proposal: Extending PBM to Carousels
  5. Implementation of Click Models
  6. TCM and CCM Fit by MLE
  7. CPBM Fit by EM
  8. CPBM Fit by GA on Click Log-Likelihood
  9. RCPBM Fit by GA on Click Log-Likelihood
  10. OEPBM Fit by MLE
  11. OEPBM Fit by GA on Observed Exaination Log-Likelihood
  12. Experimental Methodology
  13. Experimental Results
  14. RQ 1: Does GA outperform EM & MLE?
  15. RQ 2: Do PBM variants better model carousel clicks than cascade models?
  16. ...and 9 more sections

Figures (4)

  • Figure 1: A sample screen of a carousel interface from the user study as presented in RecGaze. It captures an initial presentation with the first 3 carousels shown and mouse-over details of a movie in the top row. Movie posters and mouse-over details were shown in the actual study and are removed here only due to the copyright.
  • Figure 2: Row-column position click distribution for the train set. The majority of clicks are found on the first unswiped set of items (column positions 1-5) and the initial top of the page carousel rows (row positions 1-3).
  • Figure 3: Movie item click frequency for the train set. The majority of items have one click, which can make it difficult to learn an attraction probability.
  • Figure 4: Row-column position click distribution for the test set. The combination of the sparsity of multiple clicks on items and sparsity of multiple clicks on row-column positions leads to a test set that is sparse in covering the item positions of the carousel. Note that there are only 2 clicks (position: 6,8) in the test set for swiped positions.