Possible supervisor: Heiner Stuckenschmidt
These days Internet radio becoming reality. A number of audio feeds are now available on the web. Philips is already working on a stand alone media station supporting internet radio streams. At the moment internet radio is not really different from conventional radio concerning the limited ability to influence the content of the radio program. The idea of smart internet radio is now to exploit the possibilities of web technology, in order to provide internet radio that is customized to match the users interests by combining different audio streams and selecting the most appropriate stream. The idea is that the based on model of the users interests an appropriate audio stream is selected and played.
Besides more technical problems like dealing with different audio formats and timing of audio sequences, there are some conceptual problems to be addressed. These problems originate from the need to compare user interests with metadata provided with the audio streams. Besides information about artists, songs or artists are normally assigned to different musical genres. In principle, information about genres can be used to select songs the user is likely to be interested in. The problem with this approach is that there is neither a universal agreement on a fixed set of genres nor is there an agreement on the correct classification of songs into genres. The idea is now to use semantic web technologies, in particular explicit representation of terminologies and their intended meanings. While hierarchies of genres can often be found on the web sites of audio content providers like mp3.com, the classification of music genres that reflects the opinion of a user can often be obtained from the users file systems. Corresponding concept hierarchies can automatically be extracted from user data (a corresponding tool exists).
The task to be addressed in this project is to explore different approaches for matching heterogeneous classification hierarchies in the area of musical genres. This includes the generation of appropriate test data, the application of different algorithms and a comparison of the results. The final result of the thesis should be a suggestion about the choice of a matching approach for smart internet radio.
Candidates should have knowledge in programming with JAVA or PROLOG. Experience with using semantic web languages, especially the Web Ontology language are useful.