Adaptive multimodale Exploration von Musiksammlungen

Current online music portals and systems for the management of private music collections mainly concentrate on the support for directed search for music pieces (e.g. on the basis of describing metadata like title, artist, etc.). However, empirical studies have shown that the typical behaviour of customers in classic music retail stores is much more characterized by an undirected browsing in the offered catalogue without being able to articulate a clear information demand. To improve support for this access paradigm, this thesis describes the conceptualization, implementation and evaluation of a computer game-like exploration environment for music collections, which allows its users to freely and autonomously explore a comprehensibly structured music library. Such a structure is based on a quantifiable notion of similarity between songs. Therefore, in the first part of this thesis content-based, expert-based, collaborative and text-based approaches for distance calculation are discussed and assessed. To model the multiple facets of music similarity, a subset of these measures, which cover different aspects of music, are combined into a summarizing distance function. The generation of self-organizing maps allows the topology-preserving arrangement of pieces on a two-dimensional map, on which similar songs are placed close to each other. Furthermore, it allows the detection of cluster boundaries, that separate homogeneous groups of similar pieces. This information can be used to build a three-dimensional virtual landscape, which the user can freely navigate in. A main contribution of this thesis is the supplementation of this visual presentation by permanent spatialized acoustic playback of the placed pieces, which makes a media break-free communication of the actual exploration subject possible and thereby allows for a more informed navigation of the user in the landscape. To prevent users' acoustic overstimulation we developed and implemented concepts for perception focussing and intelligent selection of songs for playback. Besides the development of possibilities to customize the environment (e.g. by moving songs or changing the height profile of the terrain), we also let the environment adapt to the user automatically: On the basis of his interaction with the system, a re-weighting of the individual similarity measure components is learned, which explains the structure provided by the user best. The development of this adaptive distance measure which allows to model a personalized similarity notion can be regarded as a further main contribution of this thesis. A concluding evaluation shows the importance of the developed multimodal presentation for the user's orientation in the virtual environment and proves the performance of the adaptive distance function.

Grantor dc:publisher

Publikationsserver der RWTH Aachen University

Year dc:date

2009

Author dc:creator

• Lübbers, Dominik

Contributors dc:contributor

• Jarke, Matthias

• info:eu-repo/classification/ddc/004
• Visualisierung
• Explorative Datenanalyse
• Benutzeroberfläche
• Raumklang
• Informatik
• Auralisierung
• Benutzeradaption
• auralisation
• user adaptation

Statement dc:rights

• info:eu-repo/semantics/openAccess

Language dc:language

ger

Identifier

https://publications.rwth-aachen.de/search?p=id:%22RWTH-CONV-113755%22

OAI record oai:identifier

https://publications.rwth-aachen.de/oai2d?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Apublications.rwth-aachen.de%3A51465