A study of heterogeneity in recommendations for a social music service

Abstract

We present a preliminarily study on the influence of different sources of information in Web 2.0 systems on recommendation. Aiming to identify which are the sources of information (ratings, tags, social contacts, etc.) most valuable for recommendation, we evaluate a number of content-based, collaborative filtering and social recommenders on a heterogeneous dataset obtained from Last.fm. Moreover, aiming to investigate whether and how fusion of such information sources can benefit individual recommendation approaches, we propose various metrics to measure coverage, overlap, diversity and novelty between different sets of recommendations. The obtained results show that, in Last.fm, social tagging and explicit social networking information provide effective and heterogeneous item recommendations. Moreover, they give first insights on the feasibility of exploiting the above non performance recommendation characteristics by hybrid approaches.