Item-Based Learning for Music Emotion Prediction Using EEG Data

Recently, emotion recognition has gained a lot of attentions particularly in the domain of music recommendation since music can induce emotions. There have been many attempts to predict emotions using Electroencephalograph (EEG) signal, a recording of brain electrical activity. However, almost all of them just employed conventional classification techniques. In this paper, we present a novel emotion prediction algorithm by applying an item-based collaborative filtering (CF). A new EEG-based similarity score is invented to be used in the proposed method. Two-dimensional emotion model is employed, and both valence and arousal levels are discretized into two classes. The experiments were conducted on two data sets: benchmark data and our own collected data. The results show that our method does not only outperform the traditional CF, but also existing classification techniques, i.e., C4.5, SVM, and MLP.