Personalized Music Playlist Recommendation Based on the User's Preference or the Mood of User's Preference

Abstract

Listening to music has become one of the closest activities in day to day life. People listen to various types of music as a hobby, to relief stress and many other reasons are there. The size of the digital music libraries are expanding rapidly. There is a thrust in finding a proper way of classifying or selecting a good playlist from an available music database. So the researchers have started looking into methods of automatic music classification. This research is a continuation of music classification and playlist generation research. Main focus is on Sri Lankan Sinhala music. Features were extracted from music tracks adhering to the music information retrieval (MIR) technologies and the classification was done using Hidden markov model (HMM). The genre classification was successful, where the results are evaluated with the supervision of a musician. Classified data was used as a training set for future classifications. The next step was to use the classified song database for music recommendation. The type of music we listen is differing from once occasion to another mainly based on the mood of that time or the preference of the user. Playlists were generated according to the basic information provided by the users about their preference. Users‟ feedback was taken to find the successfulness of the system. According to the results it was able to conclude that the genre classification using music information for Sinhala songs and the playlist generation is successful. Removing the limitation of file types that can be used to genre classification and update the user preference automatically based on the music listened by the user were identified as good improvements to the system.