Personalized Music Playlist Generation Method for Placing the Listener in a Positive Mood

Abstract

This study proposes a method for the automatic design of personalized playlists that places a listener in a positive mood (uplifting or relaxing feelings) based on individual's impressions of audio tracks. This method designs a playlist with gradually changing personal impressions of audio tracks and induces individuals' moods to be uplifting or relaxing. It estimates the subjective impressions of all audio tracks in the music database using the bagging method, a type of ensemble learning, and creates a personalized playlist. This study investigated the changes in a listener's mood before and after listening to personalized playlists, non-personalized playlists, and playlists with randomly selected tracks using a psychological scale. Consequently, many personalized playlists have shifted the listener into a positive mood more than non-personalized playlists or random design playlists.

Impression and Pleasure-based Music Playlist Generation Method for Placing the Listener in a Positive Mood

“…The scales were constructed with four categories: feeling very much, feeling a little, feeling not much, and not feeling at all. These responses were converted to numerical values for computational purposes, from feeling very much (4) to not feeling at all (1). The final score of the short MMS method was the sum of the five items on all eight scales.…”

“…This study used five words to represent the impressions of musical contents concerning the activation-deactivation level of a song. These were low (very inactive), low-middle (moderately inactive), middle (slightly inactive and slightly active), middle-high (moderately active), and high (very active), as in our previous study [1]. Users assigned one of the five impressions that they felt most from a song as the representative (primary) impression of that song.…”

“…2 will be the low-middle impression based on the mode of three users' PI X , namely, low-middle, middle, and low-middle. However, when there is more than one mode of PI X in the song, the CI is defined as (1) where CI is rounded half up to the nearest whole number; PI X is set to the values; that is, low is 5, low-middle is 4, middle is 3, middle-high is 2, and high is 1, for each impression; n is the number of users who answered PI X for the song. For example, as shown in Table 1, the CI of song No.…”

“…This study used the same 28 musical features as in our previous study [1]. Fifteen musical features of music were obtained from the music information retrieval (MIR) toolbox [12] and 13 features were obtained from the Spotify Web API audio features object [13].…”

“…Our previous paper proposed an automatic music playlist design method for placing listeners in a positive mood (uplifting or relaxing) based on an individual's impression of music [1]. In the previous study, a comparative experiment between three types of playlists was performed.…”