Influence of music liking on EEG based emotion recognition

Abstract

Abstract In studies of emotions, music is usually used to induce the emotions that are measured on the arousal-valence/arousal-valence-dominance scales. However, the influence of music liking (that depends on individual preference and appraisal) on the induced emotions is often ignored. This work presents a novel study on the influence of liking on arousal, valence, and dominance using a signal processing and pattern recognition framework. Emotion recognition was performed using a feature-level fusion of three features together with feature selection method and a classifier. The features were derived from wavelet decomposition of EEG, pairwise functional connectivity, and graph-theoretic measures that reflect characteristics of an individual electrode, pair of electrodes, and topological properties of the brain networks, respectively. Here, classification is done between the high/low categories of each of the arousal, valence, and dominance scales under three different cases of music liking. The study shows that the classification performances of arousal, valence, and dominance were 22.50%, 14.87%, and 19.44% above the empirical chance-level, respectively. The fusion framework gave up to 5% relative improvement over individual features. The study indicates that liking influences classification performance and also the temporal dynamics of emotional experience across these scales. We observe an inverted U relationship between the level of liking and arousal and dominance classification performance. We also analyzed the feature and electrode usage and specific aspects of brain activity at different levels of liking. This reveals the importance of high-frequency bands and hemispheric features in emotion recognition.