MusicGalaxy - An Adaptive User-Interface for Exploratory Music Retrieval

Sometimes users of a music retrieval system are not able to explicitly state what they are looking for. They rather want to browse a collection in order to get an overview and to discover interesting content. A common approach for browsing a collection relies on a similarity-preserving projection of objects (tracks, albums or artists) onto the (typically two-dimensional) display space. Inevitably, this implicates the use of dimension reduction techniques that cannot always preserve neighborhood and thus introduce distortions of the similarity space. MusicGalaxy is an interface for exploring large music collections (on the track level) using a galaxy metaphor that addresses the problem of distorted neighborhoods. Furthermore, the interface allows to adapt the underlying similarity measure to the user’s way of comparing tracks by weighting different facets of music similarity. Figure 1 shows a screenshot of the interface visualizing a music collection. Each track is displayed as a star (i.e. a point) with its brightness and (to some extend) its hue depending on a predefined importance measure (here a play count obtained from last.fm ‐ other measures such as a general popularity or ratings are possible). A spatially well distributed subset of the collection (specified by filters) is additionally displayed as a album cover for orientation. The arrangement of the stars is computed using multidimensional scaling (MDS) relying on a set of descriptive features to be extracted beforehand. (Alternatively, feature information may also be annotated manually or collected from external sources.) MDS is a popular neighborhoodpreserving projection technique that attempts to preserve the distances (dissimilarities) between the objects in the projection. The result of the MDS is optimal w.r.t. the minimization of the overall distance distortions. Thus, fixing one distorted neighborhood is not possible without damaging others. However, if the user shows interest in a specific neighborhood, this one can get a higher priority and be temporarily fixed (to some extend) at the cost of the other neighborhoods. To this end, an adaptive distortion technique called SpringLens [1] is applied that is guided by the user’s focus of interest. The SpringLens is a complex overlay of multiple fish-eye lenses divided into primary and secondary focus. The primary focus is a single large fish-eye lens used to zoom into regions of interest. At the same time, it compacts the surrounding space but does not hide it from the user to preserve overview. While the user can control the primary focus, the secondary focus is automatically adapted. It consists of a varying number of