Recommendation assisted playlist creation

Abstract

This paper proposes a new mechanism for the creation of music playlists called recommendationassisted playlist creation (RAPC). It uses a recommender system to assist the user in the manual creation of a playlist. Based on a seed, tracks are proposed to the user. The user can select one or more of these tracks or request a new set of recommendations. The recommender system uses the user's track selections to improve each subsequent round of recommendations. This novel mechanism is implemented using The Echo Nest as its recommender system. This implementation of RAPC is evaluated in a user study which compares it to two other mechanisms: (1) manual creation, where users can search and browse a catalog and (2) automated creation using The Echo Nest's artist radio. Participants were asked to create 10 track playlists using all three mechanisms. The results show that playlists created using RAPC: (1) are better than the playlists created using the other mechanisms (p=0.03271) (2) contain a greater number of artists than those created using the manual mechanism (p=0.001953) (3) contain more tracks that users have not listened to recently compared to playlists created using the manual mechanism (p=0.003906) (4) contain less dissatisfactory tracks compared to playlists created using the automated mechanism (p=0.0009766).