Gaze-Aware Scatterplot Analysis

Abstract

We present a task-driven approach to designing implicit gaze-aware interactions for scatterplot analysis. Building on Sarikaya and Gleicher’s taxonomy of scatterplot tasks, we focus on two foundational object-centric tasks—Identify Object and Verify Object—to guide the development of real-time, gaze-responsive interventions. Our design uses a gaze-based interest model to adapt the visualization interface without requiring explicit input from the user. We contribute four interaction techniques: rendering order, reference axes, hover speed, and click accuracy. These techniques respond passively to user attention, aiming to reduce occlusion, improve feedback responsiveness, and support more precise interactions. A user study with 24 participants demonstrates that gaze-aware adaptations can improve task efficiency and reduce cognitive load. Hover Speed emerged as the most effective and preferred technique, while other interventions showed promise with opportunities for refinement. Our findings underscore the potential of real-time attention modelling in visualization systems and motivate future research on adaptive, user-aware interaction design.

images/teaser.jpg

Overview of the gaze-aware interaction workflow. Gaze input informs an interest model, driving the system to adapt visualization behavior.

Scatterplots are foundational, yet visually demanding. We mitigate visual overload using human visual attention.

Building on scatterplot task taxonomies, we focus on identifying and verifying objects to develop real-time, gaze-responsive interventions. Our gaze-based interest model infers intent directly from eye movements, predicting interest rather than just tracking positions.

We introduce four core interactions: rendering order, reference axes, hover speed, and click accuracy. These techniques passively respond to user attention, reducing occlusion and offering greater precision—demonstrating how implicit interaction can drastically reduce cognitive load.

Techniques

Four Implicit Interactions

01 / Rendering Order

Reordering points to reduce occlusion.

Dynamically adjusts the rendering order based on the user's focus. During saccadic movements, points associated with the inferred interest color are re-rendered on top to enable faster identification without distraction.

images/rendering-order.jpg

02 / Reference Axis

Local axes for precise verification.

When fixating on a dense region, a contextual axis overlay appears locally. This supports reading values accurately in visually crowded domains.

images/reference-axis.jpg

03 / Hover Speed

Accelerated feedback for targets.

Tooltips for points of inferred interest appear significantly faster than non-interest points. Evaluated as the most preferred technique, it reduced tooltip delay by 61% overall.

images/hover-speed.jpg

04 / Click Accuracy

Expanded target geomentry.

Adjusts the clickable areas using a weighted Voronoi diagram tailored by interest. Points holding visual attention subtly expand their interaction regions, mitigating misclicks in dense plots.

images/click-accuracy.jpg

Evaluation

Results & Insights

Evaluated with 24 participants, gaze-driven interventions showed measurable improvements for visual analysis tasks and reduced perceived cognitive latency.

Key Findings

Hover Speed is Most Effective

The Hover Speed technique saved over a minute of cumulative delay per user (61% reduction), marking it the preferred interaction. 83.3% of participants reported it as non-distracting and naturally additive.

Gaze Mitigates Complex Cognitive Load

In combined technique tests, participants completed complex Verify Object tasks significantly faster (avg 4.9 min vs 6.4 min) and reported the lowest workload (NASA-TLX down to 9.1 vs 12.8).

Varied Impact on Interaction Types

While Hover Speed excelled, Rendering Order and Reference Axis updates had mixed results, with some users perceiving them as unexpected visual shifts. However, Click Accuracy helped participants click more confidently once recognized.

Promising Target Intent Modelling

The passive interest model correctly predicted target color focus in 50% of trials, significantly above the 33% random baseline. Over 90% of participants expressed intent to use gaze interactions in future analytical environments.

Performance Data

Tooltip Wait Time Reduction -61%
Verify Task Time (All vs Hover) 4.9m vs 6.4m
Target Prediction Accuracy 50%
NASA-TLX Workload (Combined) 9.1 (vs 12.8)
Future Usage Intent >90%

images/completion-time-chart.png

Mean (±SD) completion time indicates significant pairwise differences (* p < .05)

The Team

Authors

TS

Tania Sanai Shimabukuro

Researcher, MSc

Ontario Tech University

FW

Feiyang Wang

Researcher, MSc

Ontario Tech University

MS

Mariana Shimabukuro

Researcher, Associate Professor

Ontario Tech University

CC

Christopher Collins

Professor

Ontario Tech University

Citation

T. S. Shimabukuro, F. Wang, M. Shimabukuro, and C. Collins, "Designing Implicit Gaze-Aware Interactions for Scatterplots," 2026.

@inproceedings{shimabukuro2026gaze, title={Designing Implicit Gaze-Aware Interactions for Scatterplots}, author={Shimabukuro, Tania Sanai and Wang, Feiyang and Shimabukuro, Mariana and Collins, Christopher}, year={2026} }