Autotagging Music Using Supervised Machine Learning

Social tags are an important component of “Web2.0” music recommendation websites. In this paper we propose a method for predicting social tags using audio features and supervised learning. These automatically-generated tags (or “autotags”) can furnish information about music that is untagged or poorly tagged. The tags can also serve to smooth the tag space from which similarities and recommendations are made by providing a set of comparable baseline tags for all tracks in a recommender system.