Beyond Frequency Bands: Toward a Representation-Centric Paradigm for Next-Generation EEG-Based Emotion Recognition

Abstract

Electroencephalography (EEG)-based emotion recognition is an established and increasingly important technology for affective computing, mental health assessment (MHA), brain–computer interfaces (BCIs), and human–machine interaction (HMI). Despite significant advances, many EEG-based emotion recognition systems (ERSs) continue to rely substantially on predefined frequency bands and handcrafted spectral features, which may not fully capture the complex neurophysiological dynamics underlying emotional processing. This editorial argues that the next generation of EEG-based emotion recognition should move beyond conventional frequency-band analysis toward representation-driven and physiologically grounded paradigms that link neural dynamics to meaningful affective and physiological representations. We discuss major limitations of traditional EEG processing pipelines, including sensitivity to inter-subject variability, limited temporal modeling, poor cross-dataset generalization, and inadequate interpretability. We further outline emerging directions based on deep neural representations, self-supervised learning, multimodal integration, and physiologically informed modeling. A conceptual framework is proposed in which EEG signals are transformed into semantically meaningful neural representations that preserve temporal dynamics and neurophysiological relevance. Such a transition may support more robust ERSs with improved cross-subject adaptation, explainability, and translational potential. Moving beyond frequency-band-centered analysis may therefore represent an important step toward reliable, generalizable, and interpretable EEG emotion recognition technologies for future biomedical and neuroengineering applications.