Music Mood Classification Based on Lifelog

Abstract

Music mood analysis is crucial for music applications that involve search and recommendation. At present, the classification of music mood mainly depends on manual annotation or music information from audio and lyrics. However, manual annotation requires a large number of users, and the process of acquiring and processing audio or lyric information is complicated. Therefore, we need a new and simple method to analyze music mood. As music mood itself is specific psychological feelings caused by various music elements acting together on the user’s hearing, in this paper we try to use information related to users, like their physiology and activities information. The development of wearable devices provides us with the opportunity to record user’s lifelog. The experiment results suggest that the classification method based on user information can effectively identify music mood. By integrating with the classification method based on music information, the recognition effect can be further improved.