Improving Music Recommendations with a Weighted Factorization of the Tagging Activity

Collaborative filtering systems for music recommendations are often based on implicit feedback derived from listening activity. Hybrid approaches further incorporate additional sources of information in order to improve the quality of the recommendations. In the context of a music streaming service, we present a hybrid model based on matrix factorization techniques that fuses the implicit feedback derived from the users’ listening activity with the tags that users have given to musical items. In contrast to existing work, we introduce a novel approach to exploit tags by performing a weighted factorization of the tagging activity. We evaluate the model for the task of artist recommendation, using the expected percentile rank as metric, extended with confidence intervals to enable the comparison between models. Thus, our contribution is twofold: (1) we introduce a novel model that uses tags to improve music recommendations and (2) we extend the evaluation methodology to compare the performance of different recommender systems.