Abstract
Exemplified in the substantial amount of published research in music genre recognition, mood recognition and autotagging, content-based music information retrieval (MIR) advances an "engineering approach'': build a system producing the most "correct'' answers in datasets appearing throughout the literature. However, it has been clearly shown that much of this research is deficient in: formally and explicitly defining problems and use cases; identifying and testing underlying assumptions;
and using evaluation with the validity to address relevant hypotheses. This means that a system so engineered might not even be considering the through it answers "correctly''. It could thus be worthless for addressing real-world problems that must consider (e.g., music description). To emphasise the critical points above, and encourage a new approaches to research that address real-world problems,
we present four challenges for MIR researchers.
and using evaluation with the validity to address relevant hypotheses. This means that a system so engineered might not even be considering the through it answers "correctly''. It could thus be worthless for addressing real-world problems that must consider (e.g., music description). To emphasise the critical points above, and encourage a new approaches to research that address real-world problems,
we present four challenges for MIR researchers.
Originalsprog | Engelsk |
---|---|
Publikationsdato | 2015 |
Status | Udgivet - 2015 |
Begivenhed | Digital Music Research Network 9 - Queen Mary University of London, London, Storbritannien Varighed: 16 dec. 2014 → 16 dec. 2014 |
Workshop
Workshop | Digital Music Research Network 9 |
---|---|
Lokation | Queen Mary University of London |
Land/Område | Storbritannien |
By | London |
Periode | 16/12/2014 → 16/12/2014 |