CognEmoSense: A Continual Learning and Context-Aware EEG Emotion Recognition System Using Transformer-Augmented Brain-State Modeling

The real-time decoding of emotional states from neural signals has emerged as a critical frontier in affective computing and human-computer interaction. Electroencephalography (EEG), due to its high temporal resolution and non-invasive nature, remains a preferred modality for emotion recognition systems. However, existing models often suffer from poor generalization, high inter-subject variability, and inability to adapt to evolving emotional baselines over time. To address these limitations, we propose CognEmoSense—a context-aware, transformer-augmented, and continual learning-enabled EEG-based emotion recognition framework. Unlike traditional static classifiers, CognEmoSense is designed to dynamically adapt to individual brain response patterns and contextual variations. The proposed architecture integrates a temporal transformer encoder for sequence modeling, a contextual embedding layer that incorporates environmental and behavioral cues, and a continual learning module based on Elastic Weight Consolidation (EWC) to prevent catastrophic forgetting. By combining EEG signal windows (1–3s) with context vectors (e.g., time of day, stress cues), the model generates personalized, temporally stable emotion classifications. We trained and evaluated CognEmoSense using an extended version of the DEAP dataset, augmented with additional participants and multimodal annotations. Our model achieved an average accuracy of 88.0%, F1-score of 89.0%, and RMSE of 2.4 in predicting emotional valence-arousal classes—outperforming CNN-LSTM and SVM baselines by significant margins (p < 0.01). Furthermore, few-shot adaptation experiments showed that CognEmoSense can personalize to a new user with less than 10 seconds of calibration data, maintaining over 84% performance. Comparative analysis also confirmed improved robustness under noisy, real-world EEG conditions. These results highlight the feasibility of personalized, adaptive, and context-integrated emotion decoding from EEG data using transformer-based neural architectures. CognEmoSense sets a new benchmark for intelligent neural interfaces and offers promising directions for affective computing applications in mental health, adaptive learning, and brain-computer interfaces (BCIs).