Geoshuffle: Location-Aware, Content-based Music Browsing Using Self-organizing Tag Clouds

In the past few years the computational capabilities of mobile phones have been constantly increasing. Frequently these smartphones are also used as portable music players. In this paper we describe GeoShuffle – a prototype system for content-based music browsing and exploration that targets such devices. One of the most interesting aspects of these portable devices is the inclusion of positioning capabilities based on GPS. GeoShuffle adds location-based and time-based context to a user’s listening preferences. Playlists are dynamically generated based on the location of the user, path and historical preferences. Browsing large music collections having thousands of tracks is challenging. The most common method of interaction is using long lists of textual metadata such as artist name or genre. Current smartphones are characterized by small screen real-estate which limits the amount of textual information that can be displayed. We propose selforganizing tag clouds, a 2D tag cloud representation that is based on an underlying self-organizing map calculated using automatically extracted audio features. To evalute the system the Magnatagatune database is utilized. The evaluation indicates that location and time context can improve the quality of music recommendation and that selforganizing tag clouds provide faster browsing and are more engaging than text-based tag clouds.