Unsupervised Clustering of Eye Tracking Data

Abstract

The reading behavior on maps can strongly vary with factors such as background knowledge, mental model, task or the visual design of a map. Therefore, in cartography, eye tracking experiments have a long tradition to foster the visual attention. In this work-in-progress, we use an unsupervised machine learning pipeline for clustering eye tracking data. In particular, we focus on methods that help to validate and evaluate the clustering results since this is a common issue of unsupervised machine learning. First results indicate that validation using the silhouette score alone is a poor choice and should, for example, be accompanied by a visual validation using t-distributed stochastic neighbor embedding (t-SNE).