A symbolic representation approach of EEG signals for emotion recognition

Emotion recognition based on electroencephalogram (EEG) signals provides a direct access to inner state of a user, which is considered an important factor in Human-Machine-Interaction (HMI). Traditional feature extraction methods for EEG signals always suffer from high dimension and unsatisfactory interpretability. In this paper, we propose a novel symbolic representation approach of EEG signals for emotion recognition. By applying the Symbolic Aggregate approXimation(SAX) algorithm, the continuous EEG signals are represented as discrete symbol strings. The bag of words model and Latent Semantic Indexing algorithm are then performed to extract and select the word features from the symbolic strings as the discriminative features in a Support Vector Machine based classifier for emotion classification. The evaluations on DEAP dataset show that our proposed approach outperforms the three typical methods stably. Meanwhile, the symbolic representation is shown helpful to improve the interpretability of similar EEG signals. The more important issue is that this approach brings a new way to represent the EEG signal. It is helpful to introduce the natural language processing techniques to EEG signal analysis and classification research.