A Data and Query Model for Dynamic Playlist Generation

Motivated by the increasing amount of digital music on the WWW, this paper presents a data and query model for dynamic playlist generation. Queries are continuous, i.e., songs are retrieved one at a time, allowing the user to dynamically influence the retrieval. The model can support arbitrary similarity measures, requiring only the identity properly of the metric space to be obeyed. Using the similarity measure, we are able to retrieve songs similar to a given seed song and avoid retrieval of songs similar to disliked songs. Additionally, the model allows combined querying on both music metadata and musical content. Vie model has been implemented in a prototype system that is able to support a very large number of both songs and simultaneous users.