Attention Evaluation with Eye Tracking Glasses for EEG-Based Emotion Recognition

Abstract

Attention of subjects in EEG-based emotion recognition experiments determines the quality of EEG data. Traditionally, self-assessment with questionnaires is used to evaluate the attention degree of subjects in experiments. However, this kind of self-assessment approach is subjective and inaccurate. Low quality EEG data from subjects without attention might influence the experiment evaluation and degrade the performance of affective models. In this paper, we extract scanpaths of subjects while watching emotion clips with eye tracking glasses and propose an attention evaluation method with spacial-temporal scanpath analysis. Based on the assumption that subjects with attention have similar scanpath patterns under the same clips, our approach clusters these similar scanpath patterns and evaluate the attention degree. Experimental results demonstrate that our proposed approach can cluster EEG features under attentive conditions effectively and significantly improve the classification performance. The mean accuracy of emotion recognition based on clustered high quality data is 81.70%, whereas the mean accuracy of using the whole dataset is 68.54%.