In recommender system, the content-based approach is trending since the arrival of deep learning and word embedding techniques. Otherwise the advent of folksonomies and the se- mantic web brings a better understanding of user profiles and item features. In this paper, we are focusing on music recommendations and we introduce a new preference index inte- grated in a content-based recommender system. By testing our approach on Last.fm dataset, we show that the use of terms from a folksonomy to describe the music content associated in addition to music information from the semantic weballows to improve the process of music recommendation.