The way we listen to music has been changing fundamentally in past two decades with the increasing availability of digital recordings and portability of music players. Up to date research in music recommendation attracted millions of users to online, music streaming services, containing tens of millions of tracks (e.g. Spotify, Pandora). The main focus of up to date research in recommender systems has been algorithmic accuracy and optimization of ranking metrics. However, recent work has highlighted the importance of other aspects of the recommendation process, including explanation, transparency, control and user experience in general. Building on these aspects, this dissertation explores user interaction, control and visual explanation of music related mood metadata during recommendation process. It introduces a hybrid recommender system that suggests music artists by combining mood-based and audio content filtering in a novel interactive interface. The main vehicle for exploration and discovery in music collection is a novel visualization that maps moods and artists in the same, latent space, built upon reduced dimensions of high-dimensional artist-mood associations. It is not known what the reduced dimensions represent and this work uses hierarchical mood model to explain the constructed space. Results of two user studies, with over 200 participants each, show that visualization and interaction in a latent space improves acceptance and understanding of both metadata and item recommendations. However, too much of either can result in cognitive overload and a negative impact on user experience. The proposed visual mood space and interactive features, along with the aforementioned findings, aim to inform design of future interactive recommendation systems.