By Jean-Julien Aucouturier and François Pachet.
Proceedings of the 2002 IEEE International Conference on Multimedia and Expo (ICME), vol. 1, pp. 105–108, Lausanne, Switzerland, August 2002. DOI 10.1109/ICME.2002.1035729.
Read the paper (PDF) · IEEE Xplore
About the paper
The paper treats playlist generation as a constraint problem. An adaptive search algorithm improves a sequence of songs by reducing penalties for unmet criteria, such as listener preferences. The authors tested the method on a catalogue of 20,000 tracks and expanded it by duplication to measure performance at 200,000 entries.
Abstract summary
Automatically generating a playlist means finding an ordered sequence that satisfies several criteria at once. The authors formulate this as a constraint satisfaction problem and use adaptive search to find sequences in large music catalogues. They report experiments on how the method scales and describe its use in the Cuidado project.
Related research
A Combinatorial Approach to Content-Based Music Selection (2000) sets out the earlier constraint-based approach to music selection.
Related projects
Playlist