EEG Based Emotion Identification Using Unsupervised Deep Feature Learning

Capturing user’s emotional state is an emerging way for implicit relevance feedback in information retrieval (IR). Recently, EEG-based emotion recognition has drawn increasing attention. However, a key challenge is effective learning of useful features from EEG signals. In this paper, we present our on-going work on using Deep Belief Network (DBN) to automatically extract high-level features from raw EEG signals. Our preliminary experiment on the DEAP dataset shows that the learned features perform comparably to the use of manually generated features for emotion recognition.