Ana Belén Gil González

A collaborative filtering method for music recommendation using playing coefficients for artists and users

Proposal of a collaborative filtering (CF) method for music recommendation.The method is based on user and artist characterization.Only playing information that can be implicitly obtained is needed.The proposal can be applied for both rating prediction and item recommendation.The method outperforms other CF approaches. The great quantity of music content available online has increased interest in music recommender systems. However, some important problems must be addressed in order to give reliable recommendations. Many approaches have been proposed to deal with cold-start and first-rater drawbacks; however, the problem of generating recommendations for gray-sheep users has been less studied. Most of the methods that address this problem are content-based, hence they require item information that is not always available. Another significant drawback is the difficulty in obtaining explicit feedback from users, necessary for inducing recommendation models, which causes the well-known sparsity problem. In this work, a recommendation method based on playing coefficients is proposed for addressing the above-mentioned shortcomings of recommender systems when little information is available. The results prove that this proposal outperforms other collaborative filtering methods, including those that make use of user attributes.

Recommendation of Songs in Music Streaming Services: Dealing with Sparsity and Gray Sheep Problems

The interest for providing users with suitable recommendations of songs and playlists has increased since online services for listening to music have become popular. Many methods for achieving this objective have been proposed, some of them addressed to solve well-known problems of recommender systems. However, music application domain has additional drawbacks such as the difficulty for obtaining content information and explicit ratings required by the most reliable recommender methods. In this work, a proposal for improving collaborative filtering methods is presented, whose main advantage is the use of data obtainable easily and automatically from music platforms. The method is based on a procedure for deriving ratings from user implicit behavior as well as on a new way of managing the gray-sheep problem without using content information.