MPlist: Context Aware Music Playlist

Abstract

The choice of music in everyday life is greatly influenced by contextual circumstances surrounding music listeners. Music lovers create music playlists for various contexts and activities they are engaged in, and this is done manually by updating and loading new playlist each time a user changes activity or context. This does not make music listening enjoyable as much time and effort is spent on searching for songs that befit the current context and activity. This paper proposes a personalized context-aware music recommendation system, called MPlist, that dynamically and automatically creates a music playlist for music lovers based on their context (i.e., current location and activity), listening preference, nearby users listening profiles, other users listening preferences and music from labels and tags mined from music experts and the web. MPlist collects data from multiple sensors in a user’s smart mobile device and uses them to infer the user’s context and activity, thereby generating a playlist based on contextual preference. This approach has the advantage of solving the well-known cold start problem yet giving music lovers a personalized anywhere anytime music listening experience. MPlist classifier is built using both kNN and rule-based learning algorithms using sensor datasets and context-aware listening profile dataset. The contextanalytic engine infers basic user activities and sends all the inferred content to the content provider so that the server can learn the music preferences given a particular context. The system exhibit performance as follows: accuracy is 0.944, Fmeasure is 0.945, and RMSE is 0.0978. This tends to suggest that context-aware music recommendation systems is probably what music lover expect from music stores.