Emotion Recognition Using Text Embedding Models: Wearable and Wireless EEG Without Fixed EEG Channel Configurations

Emotion recognition methods using Artificial Intelligence (AI) and wearable/wireless Electroencephalography (wEEG) are promising, as wEEG signals effectively and conveniently capture brain activities related to emotions. However, conventional AI models require separate development for each wEEG channel configuration, limiting adaptability and increasing costs. To address this gap, this paper proposes a framework for leveraging text embedding models to transform wEEG signals into a standardised representation for different wEEG channel setups to be compatible with a single AI model. This approach enhances scalability, adaptability, and resource efficiency, making AI-driven emotion recognition more cost-effective and accessible. Our proposed method achieves an accuracy of 0.9368 and 0.9484 with snowflake-arctic-embed-l-v2.0 with 2-second epoching and multilingual-e5-large-instruct using 5-second epoching. This proposed method can be effectively applied across various wEEG channel configurations to support tasks to improve or explore human well-being, such as stress monitoring or emotion self-regulation.