Social Influence-Based Similarity Measures for User-User Collaborative Filtering Applied to Music Recommendation

Social characteristics present in current music streaming services allow to use methods for endowing these systems with more reliable recommendation functionalities. There are many proposals in the literature that take advantage of that information and use it in the context of recommender systems. However, in the specific application domain of music the studies are much more limited, and the methods developed for other domains cannot be often applied since they require social interaction data that are not available in the streaming systems. In this paper, we present a method to determine social influence of users uniquely from friendship relations. The degree of influence obtained is used to define new similarity metrics for collaborative filtering (CF) where more weight is given to more influential users.