EMOTION BASED MUSIC RECOMMENDATION SYSTEM USING WEARABLE PHYSICOLOGICAL SENSORS

Abstract

Many music recommendation systems rely on collaborative or content-based engines,
yet they often overlook a crucial aspect: the user's mood. The user's music choices are
influenced not only by past preferences or content features but also by their current
emotional state. This paper proposes an innovative framework for emotion-based
music recommendations, leveraging data from wearable physiological sensors.
Specifically, the user's emotions are inferred through a wearable device equipped with
galvanic skin response (GSR) and photoplethysmography (PPG) sensors. This
emotional data is then integrated as supplementary information into existing
recommendation engines, enhancing their performance. The paper addresses the
challenge of emotion recognition as arousal and valence prediction using multi-
channel physiological signals. Experimental validation is conducted on GSR and PPG
data from 32 subjects, with and without feature fusion, employing decision trees,
random forests, support vector machines, and k-nearest neighbors algorithms. The
comprehensive experiments on real data demonstrate the accuracy and efficacy of the
proposed emotion classification system, which can seamlessly integrate into any
recommendation engine without plagiarism concerns.