Characterising Eye Movement Events with an Unsupervised Hidden Markov Model
Malte Lüken1,2, Šimon Kucharský1,2, Ingmar Visser1
1Department of Psychology, University of Amsterdam, The Netherlands.
Journal of Eye Movement Research
|April 10, 2023
Summary
Researchers developed gazeHMM, an unsupervised generative model for analyzing eye movements. This hidden Markov model algorithm classifies gaze data into distinct events, offering a new method for cognitive process research.
Area of Science:
- Cognitive Science
- Neuroscience
- Computer Science
Background:
- Eye-tracking is crucial for inferring cognitive processes from eye movements.
- Current event parsing relies on algorithms, limiting hypothesis testing.
- An unsupervised, generative model is needed for robust eye-movement data analysis.
Purpose of the Study:
- Develop an unsupervised, generative model for eye-movement data analysis.
- Create an algorithm that classifies gaze data into distinct events without human-coded input.
- Enable hypothesis testing on fitted eye-movement models.
Main Methods:
- Developed gazeHMM, a hidden Markov model (HMM) based generative algorithm.
- Algorithm classifies gaze data into fixations, saccades, and optionally postsaccadic oscillations and smooth pursuits.
- Evaluated performance via simulation studies and benchmark datasets.
Main Results:
- gazeHMM successfully recovered HMM parameters and hidden states in simulations.
- For static stimuli, gazeHMM demonstrated high similarity to human coding and outperformed other algorithms.
- For dynamic stimuli, gazeHMM showed improved similarity compared to most other algorithms, despite rapid state switching.
Conclusions:
- gazeHMM provides a practical, unsupervised method for classifying eye-movement events.
- The model facilitates hypothesis testing and offers an alternative to traditional algorithmic parsing.
- Further development could enhance smooth pursuit classification and incorporate covariates for improved accuracy.
