| Title | Cover song networks: analysis and accuracy increase |
| Publication Type | Conference Paper |
| Year of Publication | 2010 |
| Conference Name | Net-Works 2010 |
| Authors | Serrà, J., Zanin M., & Herrera P. |
| Conference Start Date | 08/06/2010 |
| Conference Location | Zaragoza, Spain |
| Keywords | community detection, cover songs, Information retrieval, Music, networks |
| Abstract | The application of community detection in complex networks is
explored within the framework of cover song identification, i.e. the
automatic detection of different audio renditions of the same underlying
musical piece. In the last years this particular task has been widely
studied within the music information retrieval field as a query problem,
where one song was submitted and a list of possible matches was created
by the system. In this contribution we propose a new point of view:
songs are embedded in a complex weighted network, whose links represent
similarity (common musical content between songs). We analyze this
network and find a strong modular structure, with well-defined
communities and a clustering coefficient higher than expected. We then
perform clustering and community detection to identify groups of songs
that are versions the same musical piece. Importantly, the information
gained through this process is used to increase the overall accuracy of
the system. Results show that accuracy increments of 5 percent points
can be easily achieved. A further out-of-sample test provides evidence
that this increase can be potentially higher.
|
