A Deeper Look into Web-based Classification of Music Artists

We extensively evaluate a Web-based approach to automatic classification of musical artists into genres and explore several directions for improvements. The investigated steps to advance the current approach include the finding of optimal query terms to initially retrieve a high fraction of relevant pages from the Web and furthermore the development of a filtering technique to identify and exclude irrelevant pages among the retrieved ones. Since these extensions do not yield any substantial improvement, we propose an alternative approach that performs equally well in the task of genre classification but reduces the effort to be made to a minimum. Based on the gained insights we can conclude that Web-based artist classification operates mainly on the basis of genrerelevant proper nouns and thus can be considered to be a special form of co-occurrence analysis.