Probability based playlist generation based on music similarity and user customization

Playlist generation may be defined as the generation of a list of music items or songs based on a preexisting set of songs. A common approach would be to suggest a number of songs based on the currently playing song. An exhaustive algorithm, based on similarity calculation between two songs, would involve computation of similarity of musical attributes of all of the songs with respect of the reference song. This paper introduces a methodology focused towards decreasing the computational complexity of such an algorithm. The suggested algorithm uses a probability based model inspired by Ant Colony Optimization (ACO) to select only a small number of candidate songs for the playlist, instead of the entire database. The calculation of similarity and probability values is modified based on user feedback.