eyeNotate: Interactive Annotation of Mobile Eye Tracking Data Based on Few-Shot Image Classification

Area of Science:

• Human-Computer Interaction
• Cognitive Psychology
• Usability Engineering

Background:

• Mobile eye tracking is crucial for understanding visual attention in psychology and interaction design.
• Analyzing mobile eye tracking data is currently a manual and time-intensive process.
• Existing methods lack efficiency and scalability for large datasets.

Purpose of the Study:

• To develop and evaluate eyeNotate, a novel web-based tool for semi-automatic annotation of mobile eye tracking data.
• To compare the efficiency, validity, and reliability of a baseline annotation tool versus one enhanced with machine learning (IML-support).
• To assess the usability and user experience of the eyeNotate tool through expert evaluation.

Main Methods:

• Development of eyeNotate, a web-based annotation tool with baseline and IML-support versions.
• An expert study (n=3) comparing the two versions on usability, annotation validity, reliability, and efficiency.
• Re-annotation of existing mobile eye tracking data (n=48) by trained annotators.
• Semi-structured interviews to gather qualitative feedback on IML feature integration.
• A post hoc experiment evaluating image classification models for automated annotation.

Main Results:

• The IML-support version of eyeNotate demonstrated improved efficiency and comparable annotation validity and reliability compared to the baseline.
• Expert annotators perceived the IML-support features positively, aiding in the fixation-to-area mapping process.
• Qualitative feedback highlighted the utility of machine learning suggestions in streamlining the annotation workflow.
• Post hoc experiments confirmed the potential of image classification models for scalable data annotation.

Conclusions:

• eyeNotate offers a significant advancement in mobile eye tracking data analysis, reducing manual effort.
• The integration of few-shot learning models enhances annotation efficiency without compromising data quality.
• The tool is a valuable asset for researchers in psychology and human-centered design, facilitating more efficient visual attention studies.