Automatic Playlist Composition in a Dynamic Music Landscape
Résumé
Music is both a social and a personal experience. Our goal is to automatically compose music playlists that combine the personal expectations of the listener with the expertise of professional playlist composers (or DJs). This paper begins by introducing an environment in the form of a music landscape that can be used for indexing music, navigating through a music collection and manually or automatically generating playlist. This paper briefly describes the first two issues and focuses on the last one. An evaluation of the environment and the process is then detailed, showing that although satisfied with the results produced, users want further possibilities of interaction and more semantic expressivity. We build on these user desires and provide new, more interactive landscapes. We conclude that playlist composition should not be left to computers alone and should require a high level human interaction, if not for necessity, at least for more pleasure.