Genre Classification Using Graph Representations of Music

A song can be represented by a graph, where nodes and edges represent individual pitchduration tuples and co-occurrence of multiple notes respectively. A set of features can be derived from said graph for use in a variety of classification algorithms. In an attempt to derive meaning and utility from these graph features, we tackled the issue of genre classification–a highly subjective form of categorization in and of itself. We aimed to create a high performing method of genre classification by examining the capabilities of the algorithms SVM, Naive Bayes, multinomial logistic regression, and KNN using the aforementioned features as inputs.