Active Gaze Labeling: Visualization for Trust Building

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Data Visualization

Background:

• Manual annotation of Areas of Interest (AOIs) for gaze data analysis is labor-intensive and prone to labeling ambiguities.
• Efficient and accurate AOI annotation is crucial for reliable gaze data interpretation.

Purpose of the Study:

• To develop an interactive, user-centered system for efficient and explainable AOI annotation.
• To enhance trust in machine learning-based AOI classification through uncertainty-aware visualization.

Main Methods:

• An integrated workflow combining visualization (EyeFlower glyphs, dimensionality reduction) with machine learning and user feedback.
• Uncertainty-aware visualization to build user trust in the classification process.
• Hardware-agnostic approach supporting both stationary and mobile eye-tracking data.

Main Results:

• Demonstrated an interactive labeling approach that streamlines AOI annotation.
• Showcased the effectiveness of uncertainty-aware visualization in fostering user trust.
• Validated the versatility and hardware-agnostic nature of the proposed system.

Conclusions:

• The interactive, explainable annotation system significantly improves the efficiency and accuracy of AOI labeling for gaze data.
• The approach enhances user trust in automated classification, facilitating more reliable gaze analysis.
• This method offers a versatile solution for diverse eye-tracking applications.