Mobile Music

By Megan Cooley-Klein

The downloading of digital music is more popular with Millenials than buying CDs or listening to the radio. Many newly created service providers, such as the popular website Pandora.com and iTunes’ application Genius, use feedback from each individual user to create personalized playlists, or recommend other songs to the user. Roland T. Rust, David Bruce Smith Chair in Marketing, Michel Wedel, PepsiCo Professor of Consumer Science, and Tuck Siong Chung, Nanyan Tech University, Singapore, have created a new, more powerful model to automatically download music based on songs users like, called an Adaptive Personalization System (APS).

Most music recommendation systems belong in one of two categories: content filtering, or collaborative filtering. Content filtering systems make music recommendations based on past preferences and similarity, while collaborative filtering systems predict the user’s music preferences based on other known preferences. The APS, on the other hand, creates customized playlists depending on how long a user listens to a particular song. For example, if a user only listens to a song for about 2 seconds, then skips ahead, it is assumed that the listener does not like that song. The system can then predict the listening duration of other songs, and recommends ones with longer predicted listening duration.

One key aspect of the APS is that the system automatically downloads and creates playlists for you based on the amount of time you listened to a song. This minimizes the amount of work users have to do, which the authors found increases positive feedback. The disadvantage is that when users first start to use the system, they have to listen to a playlist created randomly by the APS in order for the system to begin to collect data. However, this can be easily improved upon by using playlist already stored by the user in their mobile devices. The APS, which is meant to be used in mobile devices and MP3 players, works in real time, so it is updated more often with user feedback, to which the system responds accordingly.

In order to find out if their system was more effective than other benchmark systems, the authors did a study to compare the two systems; people participating in the study were given the APS on a Palm PDA and instructed to test it out. The majority of the people studied were 18 to 21 years old, 63% female, and 37% male; this more or less represented the target demographic. The authors discovered that users of the APS listened to more songs picked by the system for a longer amount of time. Interestingly, the authors also found that the participants in the study were able to decide pretty quickly if they didn’t like a song, and that most of them had relatively focused song tastes.

Further research may look into the possibility of choosing song order (which is not available in the current APS version) and having different playlists for different contexts— a separate playlist for exercise songs, or songs to listen to on the way to work.

For more information about the APS and the related study, contact Rust,rrust@rhsmith.umd.edu, or Wedel, mwedel@rhsmith.umd.edu.