Publikation
CLEG: Classifying Listener Expertise Through Gaze Patterns During Classical Music Perception
Ko Watanabe; Yuki Matsuda; Rinko Hayakawa (Hrsg.)
International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp-2026), October 11-15, Shanghai, China, ISBN 979-8-4007-2539-5/2026/11, Association for Computing Machinery, 10/2026.
Zusammenfassung
Music education research has largely assessed performers, while listener expertise remains difficult to evaluate objectively without intrusive questionnaires or laboratory-grade sensing. We present CLEG, a system that estimates classical-music listening expertise from RGB-camera-based gaze behavior and explains the cues behind each prediction. From 22 listeners (13 experienced and 9 novice), sliding-window gaze features enabled a Random Forest classifier to separate experienced from novice listeners with an F1 Score of 0.670 at the window level and 0.747 with per-participant majority voting under leave-one-participant-out cross-validation (LOPOCV). SHAP analysis identified maximum fixation duration as the leading cue, suggesting that expertise leaves a recoverable and interpretable signal in commodity-camera gaze data. These results point toward scalable listener-side assessment that can complement performer-focused evaluation in music education.
