A Symbolic Approach for Task-related Gaze Classification.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Accurate classification of visual tasks from eye-tracking data is crucial for understanding user behavior.
• Existing methods often rely on complex deep learning models, limiting interpretability and robustness.

Purpose of the Study:

• To introduce a novel symbolic framework for task-related gaze classification.
• To develop an interpretable and robust distance metric for eye-tracking recordings.
• To enable accurate clustering of visual tasks without black-box models.

Main Methods:

• Encoding multivariate eye-tracking feature signals into symbolic string sequences.
• Developing a distance metric based on symbolic representations.
• Applying the framework to the ETRA 2019 dataset for task clustering.

Main Results:

• The symbolic representation allows for accurate clustering of visual tasks.
• The approach demonstrates performance comparable to state-of-the-art deep learning methods.
• The method provides an interpretable alternative to black-box models.

Conclusions:

• Symbolic encoding offers a robust and interpretable framework for gaze classification.
• This approach facilitates effective task clustering in eye-tracking studies.
• The method presents a viable alternative to deep learning for analyzing visual task data.