Developing a context-aware automatic playlist generator (CAAPG)

Aremu, Habeeb Bidemi
Journal Title
Journal ISSN
Volume Title
The current digitization of music and the sheer volume of the musical content available to listeners on local devices, such as mobile phones and IPod has been revolutionary. This trend has changed the way humans interact and experience their music. Music listeners can listen to their songs on the move. The most recent trend in the music industry is that users can organize and search for their songs based on emotions. However, most users have to manually create their playlists for particular situations. The work that this entails is cumbersome and sometimes negates the experience of the listener. The intuitive response to this problem is developing an automatic playlist generating (APG) system. Research on APG mostly focuses on using traditional metadata and audio similarity methods to create a playlist. In addition APG is seen as a static problem [1]. This means that APG is seen as a problem that does not change, however music listeners are always changing their listening habits. This thesis supports and follows from the argument made in Chi chung-yi’s work - that the APG problem is a continuous optimization problem. Additionally, in this paper I also argue that the best way to give users a good listening experience is to understand the user’s preference(s) depending on the context. Context here simply mean the features that make up the environmental space in which the system is being used. The main idea in this paper is to show the importance of emotional categorization in the generation of playlist content, while simultaneously mapping those categories to the user’s context based on the users past activities on the system. Reinforcement learning is the method used in this thesis to generate a personalized playlist, based on the context of use and the user’s emotional preference. After implementing the system we use two hypothetical users to simulate the use of our system. Various metrics are defined to measure the performance of this approach.
Thesis submitted to the Department of Computer Science, Ashesi University College, in partial fulfillment of Bachelor of Science degree in Computer Science, April 2014
automatic playlist generating system, music playlist, metadata