EEGEmotions-27: A Large-Scale EEG Dataset Annotated With 27 Fine-Grained Emotion Labels

초록

A fundamental debate in emotion research concerns whether emotions should be conceptualized as dimensional constructs or as discrete categories. Conventional models-often limited to basic emotions or bipolar dimensions such as valence and arousal-fail to capture the richness and subtlety of subjective affective experiences elicited by emotional stimuli. Most existing EEG-based emotion datasets are built upon these frameworks and therefore exhibit limitations in reflecting the full complexity and diversity of emotional states. To address this gap, we present EEGEmotions-27, a novel large-scale EEG dataset annotated with 27 fine-grained emotion categories, offering significantly higher resolution in affective representation compared to currently available public EEG datasets. To evaluate the utility of the dataset, we trained a deep learning-based emotion classification model, which achieved an average classification accuracy of 62.24%-a performance level substantially exceeding the 3.70% chance baseline for 27-class classification. This result establishes a new standard for the challenging task of fine-grained emotion recognition using consumer-grade EEG hardware.

키워드