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

Abstract

Electroencephalography (EEG)-based emotion recognition has become a key enabling technology for affective computing, Mental Health Assessment (MHA), Brain–Computer Interfaces (BCIs), and human–machine interaction (HMI). However, many existing systems continue to rely on predefined frequency bands and handcrafted spectral features, limiting their ability to capture the complexity of emotional brain dynamics and generalize across individuals and recording conditions. In this editorial, we advocate a shift toward a representation-centric paradigm in which adaptive signal conditioning, Self-Supervised Learning (SSL), multimodal integration, and physiological knowledge jointly support learning Neuro-Semantic EEG Representations (NSERs). Rather than replacing spectral information, this perspective embeds it within richer neural representations that preserve temporal dynamics, neurophysiological relevance, and semantic interpretability. We further propose a conceptual framework that positions these representations as the foundation for more robust, explainable, and transferable EEG-based emotion recognition systems. Finally, we highlight key research priorities and translational opportunities that may accelerate the development of clinically relevant affective neurotechnologies.