Visualization and clustering of tagged music data

Abstract

The process of assigning keywords to a special group of objects is often called tagging and becomes an important character of community based networks like Flickr, YouTube or Last.fm. This kind of user generated content can be used to define a similarity measure for those objects. The usage of Emergent-Self-Organizing-Maps (ESOM) and U-Map techniques to visualize and cluster this sort of tagged data to discover emergent structures in collections of music is reported. An item is described by the feature vector of the most frequently used tags. A meaningful similarity measure for the resulting vectors needs to be defined by removing redundancies and adjusting the variances. In this work we present the principles and first examples of the resulting U-Maps.