Content-Aware Collaborative Music Recommendation Using Pre-trained Neural Networks

Although content is fundamental to our music listening preferences, the leading performance in music recommendation is achieved by collaborative-filtering-based methods which exploit the similarity patterns in user’s listening history rather than the audio content of songs. Meanwhile, collaborative filtering has the well-known “cold-start” problem, i.e., it is unable to work with new songs that no one has listened to. Efforts on incorporating content information into collaborative filtering methods have shown success in many non-musical applications, such as scientific article recommendation. Inspired by the related work, we train a neural network on semantic tagging information as a content model and use it as a prior in a collaborative filtering model. Such a system still allows the user listening data to “speak for itself”. The proposed system is evaluated on the Million Song Dataset and shows comparably better result than the collaborative filtering approaches, in addition to the favorable performance in the cold-start case.