A Case-Based Reasoning Framework for Music Playlist Recommendations

Recommender Systems have recently become a fundamental part of various applications intending to support the users when searching for information, products and services that could be interested in at a given moment. While the majority of the current Recommender Systems generate isolate item recommendations based on user-item interactions, mainly expressed through ratings, this paper presents a system generating recommendations of sets of items. The objective of this system is to generate accurate recommendations in domains where users are looking for a more complete experience, in terms of joint item selections, and where the complexity of the related information cannot be evaluated and treated efficiently using the actual systems, like in the music playlist recommendations. In order to identify the underlying relations between songs, their styles and the artists that perform those, a hybrid Case-Based Reasoning approach combined with a graph model is used. This framework was designed with the scope to overcome the semantic gap present in multimedia recommendations, and at the same time, to be able to perform better in cold start situations. The experimentation results presented, show that the proposed approach is able to deliver recommendations of equal and higher accuracy than some of the widely used recommendation methods.