Posted on 04 December 2012

Part 6 in a series of videos recorded from ACM MIRUM 2012 in Nara, Japan.

Jen-Yu Liu presents a method to infer a person's age and gender from the person's music listening history. First, information about user listening history is collected using the API. User information includes age, gender, nationality, which songs were played, and when they were played. Next, two types of features are computed: context and content features. For a single user, context features represent when that user is likely to listen to music (which month, which day of the week, which hour of the day), and content features describe which songs and artists are favored by the user, along with the songs' acoustic and musical descriptors. Finally, these features are used to train two binary SVMs: one for gender, and one for age (above vs. below age 24).